diff --git a/.github/copilot-instructions.md b/.github/copilot-instructions.md new file mode 100644 index 0000000..dddaba2 --- /dev/null +++ b/.github/copilot-instructions.md @@ -0,0 +1,91 @@ +# AI Coding Agent Guide for LarsonLab.github.io + +This repo is a Jekyll-based website (Mediumish theme) for the UCSF Larson Advanced Imaging Group. Agents should focus on content changes, theme/layout tweaks, and local dev with Jekyll. + +## Architecture & Key Locations +- Content: `_posts/` for blog posts (Markdown), `_pages/` for site pages. +- Presentation: `_layouts/` (page/post templates) and `_includes/` (reusable HTML snippets like `pagination.html`, `search-lunr.html`, `disqus.html`, `share.html`, star rating includes). +- Styles & Assets: `_sass/` partials, `assets/css/` (`main.scss`, `screen.css`), `assets/js/` (jQuery, Lunr search, theme scripts), `assets/images/` for images. +- Config: root `_config.yml` (site metadata, authors, plugins, permalinks), `site/_config.yml` (dev-only overrides like `host: 0.0.0.0`). +- Feeds/Search: `feed.xml` for RSS; Lunr search via `assets/js/lunr.js` and `assets/js/lunrsearchengine.js` with `_includes/search-lunr.html`. + +## Local Development +- Docker (recommended on Windows): + - `docker-compose up` using `jekyll/jekyll:latest` with `jekyll serve --force_polling` mapped to port 4000. +- Native Ruby (optional): + - `bundle install` + - `bundle exec jekyll serve` (use `--livereload` if desired). Ensure Ruby and Bundler are installed. +- Preview: open `http://localhost:4000`. + +## Content Conventions +- Post filenames: `YYYY-MM-DD-title-with-dashes.md` in `_posts/`. +- Required front matter for posts: + ```yaml + --- + layout: post + title: "Your Title" + author: peder + categories: [ education ] + image: assets/images/MRI_logo-retro.png + featured: false + hidden: false + --- + ``` + - `author` must match a key in `_config.yml` under `authors` (e.g., `peder`, `jess`). + - `categories` drive archives via `jekyll-archives`; ensure meaningful category values. +- Pages: place in `_pages/` with `layout: page` or appropriate layout. +- Images: store in `assets/images/` and reference with site-relative paths (e.g., `assets/images/...`). + +## Theme & Includes Patterns +- Modify layouts in `_layouts/` to change global templates (`default.html`, `post.html`, `archive.html`). +- Use `_includes/` for shared widgets: + - Comments: `disqus.html` uses site `disqus` shortname in `_config.yml`. + - Search: `search-lunr.html` with Lunr scripts in `assets/js/`. + - Social sharing: `share.html`. + - Ratings: `star_rating.html` and `star_rating_postbox.html` with styles in `_sass/_stars.scss`. + - Adsense: `adsense-under-header.html` (disabled unless `_config.yml` `adsense` is enabled). + +## Configuration & Plugins +- `_config.yml` controls: + - Site metadata (`name`, `title`, `description`, `logo`, `favicon`). + - Permalinks: `/:title/` (no date in URL). + - `include: ["_pages"]` to expose pages directory. + - Plugins: `jekyll-paginate`, `jekyll-sitemap`, `jekyll-feed`, `jekyll-seo-tag`, `jekyll-archives`, `jekyll-twitter-plugin`. + - Analytics/Comments: `google_analytics`, `disqus` shortname. + - Markdown: `kramdown` with Rouge highlighting and line numbers. +- Dev-only: `site/_config.yml` sets `host: 0.0.0.0` for container access. + +## Deployment +- GitHub Pages-style repo (`LarsonLab.github.io`). No CI config present; publishing likely via GitHub Pages with default Jekyll build. Avoid adding unsupported plugins for GitHub Pages unless building externally. + +## Common Workflows & Tips +- Add author: edit `_config.yml` under `authors` (include `name`, `display_name`, `email`, `web`, `git_name`, `twitter`, `description`). +- New post: create file in `_posts/` with required front matter; set `featured: true` to show as featured in templates that support it. +- Categories & Tags pages: see `_pages/categories.md` and `_pages/tags.md`; ensure categories/tags used in posts to populate. +- Search indexing: Lunr searches client-side; keep post front matter and content consistent for indexing. +- Windows file watch: container uses `--force_polling` to avoid missed reloads. +- Ads/Analytics: only active if configured in `_config.yml`. + +## Project-Specific Notes +- Theme is Mediumish; many defaults live in `assets/js/mediumish.js` and layouts. Match existing markup/classes when adding components. +- Avoid changing `permalink` unless prepared to update internal links. +- Future-dated posts may not render unless Jekyll `future` is enabled; publish with current or past dates if needed for visibility. + +## Examples +- Example post front matter (from current repo): + ```yaml + --- + layout: post + title: "Teaching MRI and Book" + author: peder + categories: [ education ] + image: assets/images/MRI_logo-retro.png + featured: true + --- + ``` +- Docker dev: `docker-compose up` then browse `http://localhost:4000`. + +## Agent Etiquette +- Keep changes minimal and consistent with current theme and structure. +- Do not introduce site-wide layout changes without confirming intent. +- When editing content, validate local preview to catch broken links or includes. diff --git a/_layouts/default.html b/_layouts/default.html index 510df92..efe0494 100644 --- a/_layouts/default.html +++ b/_layouts/default.html @@ -73,6 +73,14 @@
  • About +
    • + +
  • + Research Projects +
    • + +
  • + Software
  • - Publications + Publications
    • diff --git a/_pages/about.md b/_pages/about.md index e6d10c6..a7775af 100644 --- a/_pages/about.md +++ b/_pages/about.md @@ -5,7 +5,7 @@ permalink: /about comments: false --- -Our research group takes an engineering-driven approach to develop advanced medical imaging methods, with most of our projects focusing on MRI although dabbling with CT and PET as well. We have a broad range of projects, focusing on human or human-ready imaging technology, including +Our research group takes an engineering-driven approach to develop advanced medical imaging methods, with most of our projects focusing on MRI although dabbling with CT and PET as well. The group is lead by led by Prof. Peder Larson, and includes numerous collaborations with other researchers and clinicians. We have a broad range of projects, focusing on human or human-ready imaging technology, including * Metabolic MRI with hyperpolarized contrast agents * Simultaneous PET/MR imaging systems * Lung MRI @@ -13,8 +13,8 @@ Our research group takes an engineering-driven approach to develop advanced medi * Radiation treatment planning * Prostate and kidney cancer prediction using deep learning -We use tools such as signal processing, optimization methods, signal modeling, statistical estimation, MRI scanner programming, and deep learning, and apply these methods for human studies in oncology, urology, pulmonology, cardiology and neurology. +We use tools such as signal processing, optimization methods, signal modeling, statistical estimation, MRI scanner programming, and deep learning, and apply these methods for human studies in oncology, urology, pulmonology, cardiology and neurology. -Our team is in Byers Hall at the UCSF Mission Bay campus, as a part of the Quantitative Biosciences Institute (QBI) at UCSF. The primary facilities available for research include 3T and 7T MRI systems, Hyperpolarizers, an electronics shop, and a machine shop, all of which are part of the Surbeck Laboratory for Advanced Imaging and are supported in part by the NIH-funded Hyperpolarized MRI Technology Resource Center. We are also actively involved in development of technology for PET/MR systems, using the time-of-flight PET with 3-Tesla MRI at China Basin. +Our team is in Byers Hall at the UCSF Mission Bay campus, as a part of the Quantitative Biosciences Institute (QBI) at UCSF. The primary facilities available for research include 3T and 7T MRI systems, Hyperpolarizers, an electronics shop, and a machine shop, all of which are part of the Surbeck Laboratory for Advanced Imaging and are supported in part by the NIH-funded [Hyperpolarized MRI Technology Resource Center](https://hyperpolarizedmri.ucsf.edu/). We also work on clinical imaging systems in our hospitals at our China Basin research facility. These include 3T MRI scanners, a 0.55 T MRI scanners, and a time-of-flight PET/MR at China Basin. Want to learn more or interested in available positions? Contact peder.larson@ucsf.edu diff --git a/_pages/researchprojects.md b/_pages/researchprojects.md new file mode 100644 index 0000000..4535935 --- /dev/null +++ b/_pages/researchprojects.md @@ -0,0 +1,130 @@ +--- +layout: page +title: Research Projects +permalink: /researchprojects +comments: false +--- + +### Hyperpolarized Carbon-13 MRI: Methods + +Hyperpolarized Carbon-13 MRI is a relatively novel imaging method of collecting in vivo metabolic information in a minimally invasive imaging procedure. Applications for this technique span various anatomical organs, including: cardiac, kidneys, brain, liver, pancreas, and clinical significances, such as: disease and cancer staging, response to therapy and surgical planning. UCSF pioneered this work with the first in vivo human study in 2013 and continues to be a strong leader in the field. Our group currently focuses on kidney and cardiac applications as well as development of acquisition methods and analysis tools to improve speed/SNR of imaging procedures and quantification of results. + +Our group has developed a “push-button” image acquisition methodology for semi-autonomous Hyperpolarized MR scans [1]. The user defines a region of interest (ROI) relative to the image volume, and then initiates the scan at the start of injection. The software is designed to autonomously perform all required calibrations and trigger acquisition: determine experimental timing (bolus tracking), adjust and capture MRI system properties (fine CF, B0, B1 calibration), and initiate rapid metabolic imaging. This system is defined using the [RTHawk Research Platform](https://vista.ai/products/research-rthawk/) from Vista.ai which safely adds a layer of functionality to our clinical scanner. Our rapid metabolic imaging method consists of alternating, metabolite-selective sequences which utilize either spectral-spatial or spectrally selective excitations. Coupled with automated bolus-triggering and calibrations, with our current methodology we are able to consistently maximize SNR and characterize near real-time metabolism. + +Experiment Recordings: + +* [Annotated Demo](https://youtu.be/ifZV_-7y7sY) +* [Human Brain](https://youtu.be/Oq36Z7ayQ0g) +* [Human Kidney](https://youtu.be/Joc9LABNRbc) + +![HP Methods](../assets/images/research_projects/hpmethods.jpeg) + +The dynamic metabolite maps acquired after reconstruction provide relative measurements over time. However, in most applications, we are interested in looking at other metrics that may serve as biomarkers. This includes model-free metrics such as the area-under-the-curve (AUC) or AUC ratios (for example, the pyruvate to lactate AUC ratio), as well as time-to-peak (TTP). Additionally, we can use and fit our acquired dynamic signals to a pharmacokinetic model to estimate apparent rate constants such as the pyruvate to lactate rate constant (kPL). Our group has worked on and developed an inputless one-compartment model [2] (can be accessed in the [Hyperpolarized MRI Toolbox](https://github.com/LarsonLab/hyperpolarized-mri-toolbox)). + +1. Tang S, Milshteyn E, Reed G, Gordon J, Bok R, Zhu X, Zhu Z, Vigneron DB, Larson PEZ. A regional bolus tracking and real-time B1 calibration method for hyperpolarized 13 C MRI. Magn Reson Med. 2019 Feb;81(2):839-851. doi: 10.1002/mrm.27391. Epub 2018 Sep 18. PMID: 30277268; PMCID: PMC6289616. +2. Larson PEZ, Chen HY, Gordon JW, Korn N, Maidens J, Arcak M, Tang S, Criekinge M, Carvajal L, Mammoli D, Bok R, Aggarwal R, Ferrone M, Slater JB, Nelson SJ, Kurhanewicz J, Vigneron DB. Investigation of analysis methods for hyperpolarized 13C-pyruvate metabolic MRI in prostate cancer patients. NMR Biomed. 2018 Nov;31(11):e3997. doi: 10.1002/nbm.3997. Epub 2018 Sep 19. PMID: 30230646; PMCID: PMC6392436. + +### Hyperpolarized Carbon-13 MRI: Kidney Studies + +Chronic kidney disease and kidney cancer are two diametrically opposed pathologies but are connected in a number of ways and both pose significant public health challenges. UCSF Medical Center is a top 10 hospital in the nation and the best in Northern California for kidney care and each of its locations treats a large number of patients with kidney disease and kidney cancer. The wide-spread use of computed tomography (CT), ultrasound imaging (US) and magnetic resonance imaging (MRI) has led to an improvement on kidney disease and kidney cancer detection. However, the noninvasive prediction of renal tumor aggressiveness and rejection of kidney allograft remain a challenge with conventional imaging. As a highly energy dependent organ, the kidney requires mitochondrial oxidative phosphorylation and ATP generation for essential renal functions and protection against injury. Altered energy metabolism plays a central role in various kidney diseases and can be detected by hyperpolarized (HP) 13C metabolic MRI. To meet the clinical needs in kidney disease, our group is developing and applying HP 13C pyruvate MRI in patients with renal cell carcinoma (RCC) and patients with transplanted kidneys to identify novel imaging techniques and develop them into practical, useful diagnostic tools under the leadership of Drs. Jane Wang and Peder Larson. + +For the research on patients with renal tumors, our group has published a paper in Cancer, an international interdisciplinary journal of the American Cancer Society. The paper describes the first study of a cohort of renal tumor patients with HP 13C MRI technology and correlations to tumor pathology. So far, we showed that HP 13C MRI enabled improved prediction of high-grade RCCs and the potential of incorporating metabolism information from HP 13C MRI to improve the prediction of aggressive renal tumors. + +![HP Kidney 1](../assets/images/research_projects/hp_kidney1.png) + +For the research on patients with transplanted kidneys, our group has presented our initial experience in applying HP [1-13C]pyruvate MRI to assess energy metabolism in patients with kidney allograft, and this work is ongoing. + +![HP Kidney 2](../assets/images/research_projects/hp_kidney2.png) + +1. Tang S, Meng M V., Slater JB, et al. Metabolic imaging with hyperpolarized 13C pyruvate magnetic resonance imaging in patients with renal tumors—Initial experience. Cancer. 2021;127(15):2693-2704. doi:10.1002/cncr.33554 +2. Liu X, Lai YC, Cui D, et al. Initial experience of metabolic imaging with hyperpolarized [1‐13C] pyruvate MRI in kidney transplant patients. Journal of Magnetic Resonance Imaging. 2025 Apr;61(4):1969-78. doi: 10.1002/jmri.29580 + +### Hyperpolarized Carbon-13 MRI: Cardiac Studies + +UCSF is one of six sites conducting hyperpolarized 13C cardiac MRI experiments, comparing healthy volunteers to hypertrophic cardiomyopathy (HCM) patients in collaboration with Dr. Roselle Abraham’s HCM clinic, a designated Center of Excellence. Hyperpolarized 13C offers a distinct advantage over FDG-PET, CEST, and 1H MRS, enabling the assessment of multiple metabolic pathways and the tracking of pyruvate-to-lactate and pyruvate-to-bicarbonate flux in a quick, non-invasive manner. Given that HCM is considered a metabolic disease, insights into various metabolic phenotypes and genotypes provide a foundation for tailored therapies. + +In this ongoing study, HCM patients and healthy volunteers are injected with 1-13C pyruvate and scanned at 3T with a Helmholz “clamshell” transmit coil and an 8-channel “paddle” receive array. The HP 1-13C pyruvate scan uses an autonomous scanning protocol (RTHawk) including metabolite-specific imaging using a spectral-spatial pulse and spiral readout to acquire 2D short-axis images. We perform pharmacokinetic modeling using a unidirectional three-site “inputless” model with one physical compartment to assess the conversion of pyruvate-to-lactate (kPL) and pyruvate-to-bicarbonate (kPB). Subjects undergo fasted and fed studies, with a dose of oral glucose given in between scans to assess metabolic response and glycolysis-glucose oxidation coupling. + +So far, we have demonstrated high quality metabolic imaging across the human heart in both volunteers and HCM patients. We also showed that, in healthy participants, kPL and kPB increased after oral glucose consumption. Additionally, the correlation of kPB with blood glucose levels was highly significant. This aligns with the fact that a healthy, metabolically flexible heart utilizes fatty acids in the fasting state and transitions to more efficient oxidative phosphorylation when glucose is available. + +![HP Cardiac](../assets/images/research_projects/hpcardiac.png) +Kinetic rate maps in healthy volunteers before and after oral glucose load, overlaid on 1H anatomical images. These measures of LDH and PDH activity are relatively uniform across the LV myocardium in healthy volunteers. + +1. Chen, Hsin-Yu, Jeremy W. Gordon, Nicholas Dwork, Brian T. Chung, Andrew Riselli, Sanjay Sivalokanathan, Robert A. Bok, et al. “Probing Human Heart TCA Cycle Metabolism and Response to Glucose Load Using Hyperpolarized [2-13C]Pyruvate MRS.” NMR in Biomedicine (2023): e5074. +2. Larson, Peder E. Z., Shuyu Tang, Xiaoxi Liu, Avantika Sinha, Nicholas Dwork, Sanjay Sivalokanathan, Jing Liu, et al. “Regional Quantification of Cardiac Metabolism with Hyperpolarized [1-13C]-Pyruvate CMR Evaluated in an Oral Glucose Challenge.” Journal of Cardiovascular Magnetic Resonance 25, no. 1 (December 14, 2023): 77. + +### Lung MRI + +MRI of the lungs has the potential to provide a range of anatomical and functional information to characterize the composition, structure, ventilation and perfusion. Use of pulmonary MRI is limited by relatively low signal due to the lower tissue density, artifacts and rapid signal decay rates due to magnetic susceptibility differences between tissue and air in the lungs, and tissue motion. + +We are developing new techniques for high-resolution and functional MRI of the lungs. This includes the use of ultrashort echo time (UTE) pulse sequences that can efficiently capture the rapidly decaying signals. We have also pioneered novel data acquisition and reconstruction strategies that can account for motion while efficiently using data to maximize SNR. This includes dynamic 3D navigators [1], iterative motion compensation (iMoCo) reconstruction [2], and Motion-Compensated Low Rank (MoCoLoR) reconstruction [3]. In addition to providing high-resolution structural imaging, these methods can also measure ventilation by measuring changes across the respiratory cycle. + +These methods are being translated into clinical studies, with a focus on pediatric imaging where it is desirable to eliminate CT scans and their associated radiation exposure. They are also being used with newly available commercial mid-field (0.5 T) MRI systems, which in the lungs are very promising due to the reduced magnetic susceptibility effects and longer signal decay rates. + +![Lung MRI 1](../assets/images/research_projects/lung_mri_1.jpeg) +Illustration of the iterative motion-compensation (iMoCo) reconstruction for UTE Lung MRI data [2], where respiratory motion can be estimated from the data itself. After dividing the data up into different motion states, images of each state are created via a motion-resolved reconstruction and then registered to each other. This is fed into the final reconstruction which can then use data from the entire free-breathing scan, maximizing the SNR efficiency. + +![Lung MRI 2](../assets/images/research_projects/lung_mri_2.jpg) +Motion-compensated low-rank (MoCoLoR) reconstructions of UTE Lung MRI data across a range of patient research subjects with various ages, body sizes, and pathologies [3]. + +1. Jiang, Wenwen, Frank Ong, Kevin M Johnson, Scott K Nagle, Thomas A Hope, Michael Lustig, and Peder E.Z. Larson. “Motion Robust High Resolution 3D Free-Breathing Pulmonary MRI Using Dynamic 3D Image Self-Navigator.” Magnetic Resonance in Medicine, n.d., n/a-n/a. +2. Zhu, Xucheng, Marilynn Chan, Michael Lustig, Kevin M. Johnson, and Peder E. Z. Larson. “Iterative Motion-Compensation Reconstruction Ultra-Short TE (iMoCo UTE) for High-Resolution Free-Breathing Pulmonary MRI.” Magnetic Resonance in Medicine 83, no. 4 (2020): 1208–21. +3. Tan, Fei, Xucheng Zhu, Marilynn Chan, Matthew A. Zapala, Shreyas S. Vasanawala, Frank Ong, Michael Lustig, and Peder E. Z. Larson. “Motion-Compensated Low-Rank Reconstruction for Simultaneous Structural and Functional UTE Lung MRI.” Magnetic Resonance in Medicine 90, no. 3 (2023): 1101–13. + +### Myelin Imaging by UTE MRI + +The myelin sheath plays an important role in normal brain function and development by facilitating long-range conduction of electrical impulses across the brain and protecting nerve fibers from injury. Loss or damage of myelin is implicated in numerous neurological disorders, including the neurodegenerative disease of multiple sclerosis (MS) associated with gross white matter damage and characteristic foci of demyelination. The benefits of myelin imaging include, myelination monitoring during aging in healthy subjects, early detection of demyelination, and assessing the treatment and re-myelination process in patients. However, due to the specific chemical environment of the bilayer-bound protons, 75% of the myelin lipid protons have T2 values well below 1 ms, which lead to magnetization decay too fast to be captured by conventional MRI techniques with TEs of milliseconds or longer. + +The most promising method for direct measurement of myelin is ultra-short echo time (UTE) and/or zero echo time (ZTE) MRI pulse sequences. Most UTE sequences require the immediate application of the readout gradients after the completion of the RF pulses and a center-out k-space trajectory in each data acquisition to achieve the minimum possible TE, which is on the order of microseconds to capture the fast decaying signals of myelin protons. The main challenge for these methods is to separate myelin ultrashort-T2 signals from the much larger water proton signals using a combination of an inversion recovery (IR) suppression and a dual-echo subtraction. Other studies have fitted T2*-relaxometry based on multiple components exponential decay models to extract myelin signals, which shows potential for quantitative analysis. + +We are working to develop and apply novel UTE MRI sequences and/or post-processing methods for direct myelin imaging. For example, utilizing novel non-Cartesian k-space patterns, multiple TE acquisitions, combining other MRI techniques like balanced steady state free precession (bSSFP) to UTE MRI, and establishing comprehensive fitting models to separate signals originating from myelin. In addition, we are partnering with collaborators at UCSF Neurology to apply these UTE MRI based imaging modalities to better characterize lesions in MS patients. + +![Myelin Imaging](../assets/images/research_projects/myelin_ute.jpeg) + +1. Boucneau, Tanguy, Peng Cao, Shuyu Tang, Misung Han, Duan Xu, Roland G. Henry, and Peder E. Z. Larson. “In Vivo Characterization of Brain Ultrashort-T2 Components.” Magnetic Resonance in Medicine 80, no. 2 (August 1, 2018): 726–35. +2. Deveshwar, Nikhil, Jingwen Yao, Misung Han, Nicholas Dwork, Xin Shen, Emil Ljungberg, Eduardo Caverzasi, et al. “Quantification of the in Vivo Brain Ultrashort-T2 * Component in Healthy Volunteers.” Magnetic Resonance in Medicine, January 30, 2024. + +### PET/MR and PET Methods + +Positron emission tomography (PET) provides valuable information about tissue function such as metabolism, perfusion, and more. Hybrid PET/MRI systems are recently introduced commercially, and combine the functional information from PET tracers with the soft-tissue contrast from MRI. For PET/MR we have worked on motion management, quantitative imaging, and image reconstruction, and several of these techniques have broad applicability for PET. + +Spatially and quantitatively accurate PET requires knowledge of positron attenuation and tissue motion. For attenuation, this information is usually derived from CT, but is challenging to derive from MRI. We pioneered methods utilizing specialized ZTE MRI scans and also the use of deep learning for MR-based attenuation correction. Most recently, we integrated a Bayesian deep learning method directly with the PET reconstruction [1]. We have also integrated MRI-derived measurements of motion to correct PET data [2]. + +In addition to attenuation and motion correction, we have worked to more generally understand what information is shared versus what information is unique between PET and MRI. Our work generating synthetic PET data from MRI datasets illustrates typical PET distributions, which can be used for generating testing data but also identifying abnormal PET distributions [3]. + +![PET](../assets/images/research_projects/pet_mr.png) +PET/MRI reconstruction integrating MR-based attenuation correction maps and their expected uncertainty derived using a Bayesian CNN along with a maximum likelihood estimation of attenuation and activity (MLAA) method weighted based on the confidence of the attenuation maps [1]. + +1. Leynes, Andrew P., Sangtae Ahn, Kristen A. Wangerin, Sandeep S. Kaushik, Florian Wiesinger, Thomas A. Hope, and Peder E. Z. Larson. “Attenuation Coefficient Estimation for PET/MRI With Bayesian Deep Learning Pseudo-CT and Maximum Likelihood Estimation of Activity and Attenuation.” IEEE Transactions on Radiation and Plasma Medical Sciences, 2021, 1–1. +2. Yang, Jaewon, Mehdi Khalighi, Thomas A. Hope, Karen Ordovas, and Youngho Seo. “Technical Note: Fast Respiratory Motion Estimation Using Sorted Singles without Unlist Processing: A Feasibility Study.” Medical Physics 44, no. 5 (May 1, 2017): 1632–37. +3. Rajagopal, Abhejit, Yutaka Natsuaki, Kristen Wangerin, Mahdjoub Hamdi, Hongyu An, John J. Sunderland, Richard Laforest, Paul E. Kinahan, Peder E. Z. Larson, and Thomas A. Hope. “Synthetic PET via Domain Translation of 3-D MRI.” IEEE Transactions on Radiation and Plasma Medical Sciences 7, no. 4 (April 2023): 333–43. + +### Image Reconstruction + +MR images are obtained in the frequency domain, or k-space, and are transformed, typically with a FFT to generate images for research and diagnostic purposes. However, since MRI is a slower imaging modality compared to its peers like CT, recent research looked to develop methods to speed up the reconstruction process which can broadly be defined as model-based and deep learning based. + +Model-based methods, including parallel imaging and compressed sensing use physical constraints to speed up the scan. Parallel imaging techniques such as SENSE and GRAPPA leverage spatial information from multiple receive coil signals to reconstruct MR images. This can reduce the acquisition time since k-space is undersampled while maintaining image quality. Compressed sensing exploits the inherent sparsity of medical images in the wavelet domain and uses iterative non-linear reconstruction solvers to converge to a generated image. + +Deep-learning methods, and by extension, unrolled algorithms attempt to leverage the data consistency with traditional iterative reconstruction but also take advantage of learning from vast datasets to capture patterns and relationships on a population level that wouldn’t be easy using classic methods. These methods are data hungry and computationally expensive. + +Our research group has experience in developing and applying both types of methods, with our most recent work focusing on deep-learning methods. This has included developing a scan-specific, self-supervised method that does not require training or calibration data [1] and using deep generative models to synthesize large amounts of raw data from clinical images that can be used to train these models [2]. + +![Image Reconstruction 1](../assets/images/research_projects/img_recon_1.png) + +![Image Reconstruction 2](../assets/images/research_projects/img_recon_2.png) + +1. Leynes, Andrew P., Nikhil Deveshwar, Srikantan S. Nagarajan, and Peder E. Z. Larson. “Scan-Specific Self-Supervised Bayesian Deep Non-Linear Inversion for Undersampled MRI Reconstruction.” arXiv, December 9, 2023. +2. Deveshwar, Nikhil, Abhejit Rajagopal, Sule Sahin, Efrat Shimron, and Peder E. Z. Larson. “Synthesizing Complex-Valued Multicoil MRI Data from Magnitude-Only Images.” Bioengineering 10, no. 3 (March 2023): 358. + +### AI Methods for Cancer Imaging + +Biomedical imaging is an integral part of the management of cancer, from diagnosis to staging and monitoring response to treatments. Modern computer vision and AI techniques are set to revolutionize cancer imaging. We believe there are significant opportunities to create more reproducible assessments, automated quantifications of disease, and even discover new imaging characteristics of aggressive disease. + +In primary prostate cancer and localized kidney cancer, the main challenge is to improve our identification of aggressive, clinically-significant disease. In primary prostate cancer, MRI is typically used prior to biopsy, offering an opportunity to better localize biopsy samples or even rule out the need for biopsy. We have curated an internal database of over 1000 prostate cancer patients with MRI. We have also developed a specialized, mixed-supervision deep learning method to utilize all available clinical pathology data regardless of its specificity, and this was shown to improve performance of identification of clinically-significant prostate cancer [1]. We also showed it can be used across different medical centers data with federated learning, a distributed method that doesn’t require data sharing [2]. + +![AI Methods Cancer Imaging](../assets/images/research_projects/ai_cancer.png) + +We are also exploring projects using other imaging modalities where UCSF has significant experience such as PET for metastatic prostate cancer and CT for kidney cancers, all still aiming to improve our assessments of cancer with imaging. + +1. Rajagopal, Abhejit, Antonio C. Westphalen, Nathan Velarde, Tim Ullrich, Jeffry P. Simko, Hao Nguyen, Thomas A. Hope, Peder E. Z. Larson, and Kirti Magudia. “Mixed Supervision of Histopathology Improves Prostate Cancer Classification from MRI.” arXiv, December 12, 2022. +2. Rajagopal, Abhejit, Ekaterina Redekop, Anil Kemisetti, Rushikesh Kulkarni, Steven Raman, Karthik Sarma, Kirti Magudia, Corey W. Arnold, and Peder E. Z. Larson. “Federated Learning with Research Prototypes: Application to Multi-Center MRI-Based Detection of Prostate Cancer with Diverse Histopathology.” Academic Radiology, Special Issue: Adapting after COVID, 30, no. 4 (April 1, 2023): 644–57. diff --git a/_pages/software.md b/_pages/software.md new file mode 100644 index 0000000..61c566d --- /dev/null +++ b/_pages/software.md @@ -0,0 +1,83 @@ +--- +layout: page +title: Software +permalink: /software +comments: false +--- +We have developed and (try to) maintain software for hyperpolarized MRI experiments, RF pulse design, k-space trajectory design, image reconstruction, and educational software. They are all available on GitHub: + +* (Group specific organization) +* (Hyperpolarized MRI organization) +* (Pulmonary MRI organization) + +We welcome contributions and collaborators to these repositories. You can contribute via Pull Requests or Issues through GitHub, or you can reach out to us via email with any interests or comments. + +## Hyperpolarized MRI + +### Hyperpolarized-MRI-Toolbox + +We have compiled and developed a wide range of software tools for hyperpolarized MRI experiments in the [hyperpolarized-mri-toolbox](https://github.com/LarsonLab/hyperpolarized-mri-toolbox). This includes methods for designing radiofrequency (RF) pulses, imaging gradients, data reconstruction, and data analysis including pharmacokinetic modeling. + +Hyperpolarized-MRI-Toolbox. Available online at: doi: 10.5281/zenodo.1198915 + +### Hyperpolarized Pulse Sequences + +We have developed and supported development of pulse sequences for Bruker preclinical MRI and GE Healthcare clinical MRI scanners, as well as in the vendor-agnostic RTHawk Research software. We also have custom reconstruction pipelines for the RTHawk Research platform that are available. Since these rely on proprietary source code, they are only available as private repositories. + +Please see for how to request access. + + +## Pulse Sequence Design + +### Spectral-Spatial RF Pulse Design + +In the early days of hyperpolarized MRI, a UCSF-Stanford collaboration developed a comprehensive spectral-spatial RF pulse design package, including many optimizations and unpublished methods. Even though it was started in 2007, I think it's still near state-of-the-art today (2024) :) + + + +### Multiband RF Pulse Design + +This software specifically focuses on designing RF pulse with multiple different bands of arbitrary magnitude and phase spectral response using using convex optimization to minimize pulse duration, transition width and total energy with flexible trade-off. + + + +### Long-T2 Suppression Pulses for Ultra-short Echo Time (UTE) MRI + +Ultra-short echo time (UTE) and zero echo time (ZTE) magnetic resonance imaging (MRI) allows for visualization of semi-solid tissues with very short T2 relaxation times such as tendons, calcifications, myelin sheaths, and cortical bone, are normally invisible with conventional MRI techniques. +Long-T2 species suppression with RF pulses is one way to improve the contrast of short-T2 species: + + + +### Radial-Field-of-Views + +Code for designing 3D radial, 3D cones, and PROPELLER k-space trajectories +anisotropic FOV and resolution sampling patterns: + + + +## Image Reconstruction + +### Motion Management and Lung MRI + +Motion management is critical for lung MRI. We have developed reconstruction methods specifically to manage motion through data-driven motion estimation, estimation motion fields, and iterative reconstructions. There include +* Soft-gating, Motion-resolved, and Dynamic Image Navigator reconstructions + +Reference: + + +* iterative Motion Compensation (iMoCo) reconstruction for MRI. +. Reference: + +* Motion-compensated low-rank reconstruction for simultaneous structural and functional UTE lung MRI, . Reference: + +### Deep Learning based Image Reconstruction + +Deep non-linear inversion (DNLINV) is a scan-specific self-supervised method, meaning no training data is required, that utilizes Bayesian deep learning for undersampled MRI image reconstruction even without calibration data. Pre-print: + + + +## Educational Software + +Prof. Larson and the teaching assistants in Principles of MRI taught at UCSF (Biomedical Imaging 201) have developed a simple software package to reinforce and explore the basic principles of MRI. + +MRI-education-resources \ No newline at end of file diff --git a/_posts/2021-08-27-AI-Art-from-MRI.md b/_posts/2021-08-27-AI-Art-from-MRI.md index 8fd3e39..67b7be5 100644 --- a/_posts/2021-08-27-AI-Art-from-MRI.md +++ b/_posts/2021-08-27-AI-Art-from-MRI.md @@ -3,7 +3,7 @@ layout: post title: "AI-generated Art and MRI" author: peder categories: [ fun ] -image: assets/images/HPBrain_MRI-Spaceship-7.png +image: assets/images/HPBrain_MRI-Spaceship.jpg featured: true --- I recently went off on an exploration of generating images and art using AI, specifically generative adversarial networks (GANs). This was first inspired by a twitter feed, https://twitter.com/images_ai?lang=en, where they provide a nice Google Colab notebook. @@ -18,9 +18,9 @@ So I began playing around with the CoLab notebook, here's a few fun explorations In these examples, I used a random image seed, with the prompt "MRI Spaceship". The difference is based on the image training data (Wikiart versus ImageNet) -![MRI spaceship](../assets/images/MRI-spaceship-1.png) +![MRI spaceship](../assets/images/MRI-spaceship-1.jpg) -![MRI spaceship](../assets/images/MRI-spaceship-2.png) +![MRI spaceship](../assets/images/MRI-spaceship-2.jpg) ## Seeding with MRIs - Hyperpolarized Brain @@ -36,12 +36,14 @@ And the prompt "MRI Spaceship". The image evolved as ![Hyperpolarized 13C Brain MRI Spaceship 4](../assets/images/HPBrain_MRI-Spaceship-4.png) ![Hyperpolarized 13C Brain MRI Spaceship 5](../assets/images/HPBrain_MRI-Spaceship-5.png) ![Hyperpolarized 13C Brain MRI Spaceship 6](../assets/images/HPBrain_MRI-Spaceship-6.png) -![Hyperpolarized 13C Brain MRI Spaceship 7](../assets/images/HPBrain_MRI-Spaceship-7.png) +![Hyperpolarized 13C Brain MRI Spaceship 7](../assets/images/HPBrain_MRI-Spaceship.jpg) +![Hyperpolarized 13C Brain MRI Spaceship animation](../assets/images/HPBrain_MRI-Spaceship.gif) -I also tried with a prompt of something like "Moon | Volcano", but using WikiArt training: -![Hyperpolarized 13C Brain Volcano Moon](../assets/images/HPBrain_volcano-moon.png) +I also tried with a prompt of something like "Moon Volcano", but using WikiArt training: + +![Hyperpolarized 13C Brain Volcano Moon](../assets/images/HPBrain_volcano-moon.jpg) ## Seeding with MRIs - Prostate Diffusion @@ -59,8 +61,8 @@ This was getting fun, so I started with this seed image of a prostate diffusion **"Moon Carrot"** -![Moon Carrot](../assets/images/prostateDWI_moon-carrot-wikiart.png) +![Moon Carrot](../assets/images/prostateDWI_moon-carrot-wikiart.jpg) **"Vegetables Under the Moon"** -![Vegetables under the moon](../assets/images/prostateDWI_Vegetables-under-the-moon.png) +![Vegetables under the moon](../assets/images/prostateDWI_Vegetables-under-the-moon.jpg) diff --git a/_posts/2022-07-28-Synthesizing_CT_from_MRI.md b/_posts/2022-07-28-Synthesizing_CT_from_MRI.md index 0317411..4bc99ed 100644 --- a/_posts/2022-07-28-Synthesizing_CT_from_MRI.md +++ b/_posts/2022-07-28-Synthesizing_CT_from_MRI.md @@ -1,15 +1,15 @@ --- layout: post title: "Synthesizing CT from MRI in Radiation Oncology" -author: Jess Scholey +author: jess categories: [ RadiationOncology ] -image: assets/images/image_name.png +image: assets/images/MR-CT-dose_comparison.jpg featured: true --- In Radiation Oncology, photons, protons, or electrons are used to treat cancer. Medical imaging is crucial in this process and allows us to visualize a patient’s tumor and healthy organs. The most common imaging modality used in radiation oncology is a computed tomography (CT) scan because it provides both anatomical information and a map of photon attenuation, which is necessary for calculating dose delivered by the radiation. However, magnetic resonance imaging (MRI) is a powerful imaging modality that can provide very helpful information, such as excellent soft tissue contrast and functional information about tumor behavior. Therefore, MRI is becoming a very popular imaging choice in radiation oncology clinics. -Figure 1. The UCSF Radiation Oncology Team having fun testing MRI sequences for imaging the pelvis. ![Vida Volunteer](../assets/images/Vida_volunteer.JPG) +Figure 1. The UCSF Radiation Oncology Team having fun testing MRI sequences for imaging the pelvis. ![Vida Volunteer](../assets/images/Vida_volunteer.jpg) One drawback of utilizing MRI in radiation oncology is that it does not provide a map of photon attenuation which is crucial for dose calculation. This can be explained by looking at examples of bone versus air. Bone is most dense biological material and will highly attenuate a beam of radiation. Air, on the other hand (like in the lungs), is not very dense and will minimally attenuate a beam of radiation. On a CT scan, bone is bright while air is dark. On MRI scans, both bone and air appear as dark (due to low proton density of air and bone and very short transverse relaxation time of bone). Because MRI pulse sequences cannot inherently provide information on photon attenuation, patients will often receive both a CT scan (for photon attenuation information) and MRI (for anatomical visualization). diff --git a/_posts/2024-01-10-The-Sounds-of-MRI.md b/_posts/2024-01-10-The-Sounds-of-MRI.md new file mode 100644 index 0000000..95c7995 --- /dev/null +++ b/_posts/2024-01-10-The-Sounds-of-MRI.md @@ -0,0 +1,72 @@ +--- +layout: post +title: "The Sounds of MRI" +author: peder +categories: [ fun ] +image: assets/images/The%20Sound%20of%20MRI.jpg +featured: true +--- +When I tell people I work on MRI systems, one of the most common questions is about the sounds (or maybe described as noise, depending on your experience). Why is it so loud? Can you make it quieter? + +Jump to the bottom of the article if you just want to hear MRI machines making music + +## What causes the sounds of MRI? + +The sounds made during an MRI scan are primarily due to the magnetic field gradient coils. It occurs when the coils are switched on and off in the presence of a magnetic field (in MRI, the main B0 field). When the currents inside these coils are changed, there is a force exerted on the coils. This is called the Lorentz force, which is a force that is exerted on a charged patricle (currents) in a magnetic field. + +The pitch or frequencies created during the scan depend on the rate that the gradient coils are turned on and off. For example, to get middle C (440 Hz), the gradients should be switched at a rate of 1/440 Hz = 2.2 ms. + +## Sounds and the Pulse Sequence + +Different pulse sequences have very different patterns, depending on the overall timing as well as the rate of gradient switching rates. + +Echo-planar Imaging (EPI) is a great example, as it is typically done with the most rapid gradient switching possible. The switching rates is often less than 1 ms, leading to frequencies of 1 kHz and above. EPI readouts are also short (< 100 ms) which then leads to high frequency "chirps". + +![Echo planar imaging (EPI) sequence](../assets/images/EPI%20sequence.jpg) + +Diffusion-weighting gradients are another great as example, as they have some of the longest gradient pulse durations and thus slow switching rates. But they are also very high amplitude so have a strong volume. Thus diffusion-weighting leads to very low frequency sounds, in my experience to the point you feel the vibration in the table and your bones. In a diffusion pulse sequence, where $\Delta$ is typically 10-50 ms, leading to frequencies of 100 Hz and below. + +![Diffusion Pulse sequence, where $\Delta$ is on the order of 10s of ms](../assets/images/Spin-echo%20diffusion%20sequence.jpg) + +Given these patterns, you can identify some characteristics of the pulse sequence while you are being scanned. Some examples +* repeated high frequency patterns would indicate an EPI readout, e.g. for DWI or fMRI +* consistent, moderate frequency (a bit grinding) is probably a T1-weighted (short TR) scan +* large gaps in sound indicate long TR, for example in T2 weighted imaging +* other unqiue sequence timings are distinct, such as inversion recovery which is a blip (crusher gradient after inversion pulse), delay, up to 1 s of noise (readout), and another delay + +You can find sounds recorded for a full MRI exam here: + +## Making Quieter MRI scans + +Since the sounds in MRI are caused by the changing of the magnetic field gradient currents, a MRI scan can be made quieter by reduced the rate at which the gradients are changed. However, this is much easier said than done, as many MRI scans rely of rapid switching of these gradients to create contrast, to get high SNR, and/or to keep scan times from getting too long. + +The most prominent example of quieter MRI scans I'm aware of are zero echo time (ZTE) pulse sequences. I have worked with these sequences and it is shocking - you expect to hear the scanning easily from outside the room, but instead you have to strain and maybe turn up the scan room microphone volume to even hear the sequence! + +These work by having the gradients on prior to the RF pulse so the k-space trajectory begins immediately. Data is read out, and then the gradients are only slightly changed for the next repetition. This slow change in the gradients is the key feature that makes them silent. + +![ZTE Pulse Sequence](../assets/images/ZTE%20pulse%20sequence.jpg) + +More info: + +Ljungberg, E., Damestani, N.L., Wood, T.C., Lythgoe, D.J., Zelaya, F., Williams, S.C., Solana, A.B., Barker, G.J. and Wiesinger, F., 2021. Silent zero TE MR neuroimaging: current state-of-the-art and future directions. Progress in Nuclear Magnetic Resonance Spectroscopy, 123, pp.73-93. + +## The Most Annoying MRI Scan + +My team happens to work with one of the arguably most annoying sounding MRI scans. It is a 3D ultrashort echo time (UTE) sequence, basically a T1-weighted scan with a short TR, but with the twist that the gradient direction is being incoherently modified every TR. This leads to an extremely dissonant set of frequencies when the scanner is running + +[3D radial UTE sequence with golden angle ordering](../assets/audio/UTE%20golden%20spaceship.wav) + +[3D radial UTE sequence with bit-reversed ordering](../assets/audio/UTE%20reverse%20spaceship.wav) + + +## Making Music with MRI + +Some clever MRI experts have realized that we can do a reasonable job at controlling the sounds that are created by a MRI scanner. + +There are fun examples of playing songs on a MRI scanner (a pretty expensive instrument!). +* This example uses the MRI scanner for the iconic guitar riff in Smoke on the Water by Deep Purple: +* Intro by The XX - an MRI Cover, with some nice visualization as well: +* The Root Beer Rag by Billy Joel +* There is another beautiful example of MR fingerprinting, a quantitative imaging technique, making music while actually also producing good quality data: +Ma, D., Pierre, E.Y., Jiang, Y., Schluchter, M.D., Setsompop, K., Gulani, V. and Griswold, M.A. (2016), Music-based magnetic resonance fingerprinting to improve patient comfort during MRI examinations. Magn. Reson. Med., 75: 2303-2314. +[Bach's Cello Suite No. 1, based on a recording by Yo-Yo Ma, played on a MRI scanner](https://onlinelibrary.wiley.com/action/downloadSupplement?doi=10.1002%2Fmrm.25818&file=mrm25818-sup-0006-suppinfo06.mp3) diff --git a/_posts/2024-07-24-Tabletop_MRI_Education.md b/_posts/2024-07-24-Tabletop_MRI_Education.md new file mode 100644 index 0000000..d16bfd4 --- /dev/null +++ b/_posts/2024-07-24-Tabletop_MRI_Education.md @@ -0,0 +1,34 @@ +--- +layout: post +title: "Tabletop MRI for Education - Initial Experience" +author: peder +categories: [ education ] +image: assets/images/ilumr_larson-lab.jpg +featured: true +--- +As the ISMRM Annual Meeting 2023 in Toronto, I was approached by a group of UCSF PhD students telling me I had to see something. They led me over the to [Resonint](https://www.resonint.com/) booth in the vendor hall to experience the Ilumr tabletop MRI system. This enterprising group of students was clearly impressed, and rightfully so! + +I got a demo right on the exhibition floor with the system and was amazed. There was a sleek system, fully contained on a tabletop, running via JupyterLab-based web interface, and instantly delivering beautiful results. It is a beautiful example of modern engineering with thoughtful design towards education and accessibility. They are building open-source educational tools , and the team was friendly and open with me for all my questions. So, I began my quest to find the funds to buy one! + +Long story short: I was fortunate in early 2024 to receive support from the UCSF Department of Radiology and Biomedical Imaging to purchase a system. It arrived late April 2024, and now we have finally got to testing and learning. + +The experience so far has been great! Some learning curve to getting onto the system, but much less so than other MRI systems I've worked with. It's being tested by PhD students, post-docs, post-BS/MS research assistants, undergrads, and even high school students! Currently we are doing a short summer course, going through the [MRI Fundamentals labs](https://github.com/Resonint/ilumr-courseware) from Resonint. I plan to integrate this into my MRI course in the Fall , so stay tuned for more updates then. + +### Small flower with 3D RARE + +![Small Flower with 3D RARE](../assets/images/ilumr-flower.gif) + +### Celery + +![Celery](../assets/images/ilumr_cel2RAREinv.png) + +### Blackberry with 3D RARE +![Blackberry with 3D RARE](../assets/images/ilumr-blackberry.png) + +### MRI Education in Action +![Action!](../assets/images/ilumr_action1.jpg) + +![Action!](../assets/images/ilumr_action2.jpg) + +![Action!](../assets/images/ilumr_action3.jpg) + diff --git a/_posts/2024-08-31-What-Can-UTE-Image.md b/_posts/2024-08-31-What-Can-UTE-Image.md new file mode 100644 index 0000000..80a893b --- /dev/null +++ b/_posts/2024-08-31-What-Can-UTE-Image.md @@ -0,0 +1,47 @@ +--- +layout: post +title: "Violin Imaging and other fun UTE MRI scans" +author: peder +categories: [ fun ] +image: assets/images/violin_angled.png +featured: true +--- +Ultrashort echo time (UTE) MRI pulse sequences use specialized approaches for RF excitation and data acquisition to minimize the delay between excitation and acquisition in order to capture rapidly decaying signals with short T2* relaxation rates. These techniques also have a special place in my heart and research career, as they were the first MRI technique I worked on in my PhD. And I have been fortunate that it has been a gift that keeps on giving, spawning multiuple biomedical research projects imaging bones, the lungs, tendons and other connective tissues, and myelin. Thank you Dwight Nishimura for this gift! + +UTE MRI techniques (and here I would include zero echo time or ZTE methods as the shortest TE technique) can also acquire signals from objects that we normally would not expect to be able to image with MRI. I've accumulated a few of my favorites, from wood to rubber to plastics + +### Violins +At the end of my PhD, one of my fellow students Joelle Barral had the idea to image her violin with UTE. After making sure to remove the strings and other metal parts, the results were wonderful! + +![Violin MRI](../assets/images/stradivariUTE.png) + +You can distinguish the different components, the fingerboard and pegs stood out with high signal maybe due to higher density and/or longer T2*. The violin is made of wood but also has a varnish applied. I'm actually a cellist, but have yet to try cello imaging, maybe next! + + +### RF coils + +Some RF coils show significant signal on UTE MRI. In this surface coil, there is a lot of signal where the reactive components are placed around the loop. + +![Surface Coil UTE](../assets/images/UTE-surface_coil.png) + +Here's another example shoing RF coil cables, housing, and other plastic material in patient table and positioning. + +![RF coils and housing](../assets/images/ZTE_RF-coils.jpg) + +Weiger M, Pruessmann KP. Short-T2 MRI: principles and recent advances. Prog Nucl Magn Reson Spectrosc. 2019;114–115:237–70. https://doi.org/10.1016/j.pnmrs.2019.07.001 + +This imaging has some real-world use case for PET/MR and radiation therapy planning where photon attenuation of the coils maybe important to consider. E.g. Mootaz Eldib et al 2015 Phys. Med. Biol. 60 4705 DOI 10.1088/0031-9155/60/12/4705 + +### Other stuff + +I used to use a wood block as a phantom for UTE, which has quite low proton density but is visible with efficient techniques! Below you can see my wood block phantom, placed next to a pair of headphones and also in a plastic boot used to immobilize the ankle. + +![Wood in Headphones](../assets/images/UTE-wood_headphones.png) + +![Wood in Plastic Boot](../assets/images/UTE-wood_plastic.jpg) + +Rubber has become a more popular UTE phantom, particularly becuase its T2* is similar to some in vivo signals like cortical bone, and also it has more proton density than my wood phantoms. + +![Rubber](../assets/images/ZTE_rubber.jpg) + +Weiger M, Pruessmann KP. Short-T2 MRI: principles and recent advances. Prog Nucl Magn Reson Spectrosc. 2019;114–115:237–70. https://doi.org/10.1016/j.pnmrs.2019.07.001 \ No newline at end of file diff --git a/_posts/2025-02-05-Tabletop_MRI_Education.md b/_posts/2025-02-05-Tabletop_MRI_Education.md new file mode 100644 index 0000000..fd40385 --- /dev/null +++ b/_posts/2025-02-05-Tabletop_MRI_Education.md @@ -0,0 +1,64 @@ +--- +layout: post +title: "Tabletop MRI for Education - Teaching Experience" +author: peder +categories: [ education ] +image: assets/images/ilumr_larson-lab.jpg +featured: true +--- +Imagine being able to scoop up your MRI scanner, bring it into a classroom, and then live-demo MRI experiments alongside typical classroom materials. This was what I was able to do during my Fall 2024 UCSF course, "Biomedical Imaging 201: Principles of MRI," with the Ilumr 0.34 T Tabletop MRI system from [Resonint](https://www.resonint.com/). As a teacher, I loved the experience, it felt so fresh and active, and truly a way to bring the course material to life. My department published a blog and produced a short video, you can check them out below. Here I'll also share some more specific insights into my experience. + +## UCSF Blog Posts + + + + + + +My favorite quotes from the blog: + +"In Larson’s classroom ..., students have on their table a machine that vaguely resembles a rice cooker or a speaker about to throw down some serious bass." + +"It's a tabletop MRI scanner, the Ilumr from Resonint, weighing about as much as a toddler", + +"Larson's classes have taken on a playful yet informative tone, with everyday objects becoming subjects of MRI scans. From juicy blackberries to crunchy cauliflower florets, students image a variety of samples, learning about signal intensity, contrast, and spatial resolution. As the machine softly whirrs like a gentle air conditioner with occasional beeps, students witness the effects of motion artifacts by gently wiggling the sample tube during a scan, creating intriguing "ghost" images, or drop a metal contaminant, either steel or brass, into the sample tube to show how signal is destroyed and how artifacts pile up in the image." + +![Tabletop MRI in the Classroom](../assets/images/peder-larson-tabletop-MRI.jpg) + +## General Tips + +* **Startup**: Warming up the system takes ~30 minutes for temperature and field to stabilize. But, even if it is not stable, this is a good teachable moment to show how MRI is sensitive to field changes +* **Wobble**: Not essential to do every time, but once or twice the coil was detuned and SNR was poor. + +## Samples + +- **Doped water**: Most often I just used the water shim tube, for ease of use and consistency. You can demonstrate a lot with just that. Easy to see effects of spatial resolution, SNR, artifacts such as displacement and motion. +- **Structure Phantoms**: The structural set of phantoms provided are also nice for slice selection and more structure, but take more time to setup properly. +- **Lunch**: I often used a small piece of food from my lunch, and these ended up giving some of the most interesting images! Rolled up lettuce, pieces of rice, salami, and cheese, and cauliflower all worked very well. + +## Sequences +* **Advanced RARE**: I used this sequence most frequently, and it is extremely relevant given its high usage for in vivo MRI. +* **FISP**: This sequence was also quite efficient, so I used this on occasion. The unique steady-state of this sequence I find to be quite complex, so put it beyond scope of my course, which made me hesitant to use this more. + +## Experiments +* **Resonint Labs**: Resonint provides a fantastic set of Jupyter Notebooks to illustrate many concepts. I had students try these themselves and also used for some in class demos. +* **MRI System**: The Lab 1: Intro to NMR" nicely covered this, I liked the wobble to illustrate RF coils. I also had not explicitly covered shimming before so this was nice addition. +* **Contrast**: I did not have much success demonstrating the "weighting" concepts of contrast (e.g. T1, T2, PD), partly due to lack of phantoms, and not having a simple GRE sequence also made this more challenging. Chemical shift was also not easy to demonstrate given the low magnetic field strength. +* **Spin-Echoes**: Very easy to demonstrate with included Labs and Apps. +* **RF Pulses**: The Pulse Calibration App was a very clear demonstration of pulse power/flip angle. Slice selection was very easy to demonstrate by ensuring the image was resolved in the slice select direction. Lab 3 was great for this as well. +* **K-space and Fourier Transforms**: There were nice ways by imaging the RF pulse profiles and also by looking at the frequency spectrum from the samples to illustrate concepts like rect/sinc pairs and the stretch/shrink property of the Fourier Transform. The 1D SE App was useful for this. +* **Image Formation**: It was easy to demonstrate properties of frequency encoding by showing the spectrum of the signal. Phase encoding, as I expected, was not so easy to show. I took some parts of Lab 2 for this. +* **Image Characteristics**: I did some simple demonstrations using RARE sequence of tradeoffs between *scan time, resolution, FOV and SNR.* This was relatively simple, but I think cannot be understated how important this is in practical use of MRI. This is the constant exercise of MRI practitioners, to balance desired contrast with scan time, resolution and SNR, and I think is incredibly valuable. +* **Fast Imaging**: The RARE sequence app was great for this topic. But there was no EPI sequence. Since it is single channel, there's no opportunity for parallel imaging. I didn't attempt any compressed sensing. +* **Aliasing**: I had to go into the configuration files to figure out how to change the FOV, but with this was able to nicely demonstrate aliasing. +* **Motion**: I was able to induce motion artifacts by shaking the shim during imaging. It was better without averaging, since averaging can remove some of this artifact. +* **Susceptibility Effects**: I had some fun with this concept by inserting some non-magnetic metals into the sample tube. In some cases, the signal was completely lost, and in others, the signal was just distorted. +* **Flow**: This was not a classroom demonstration, but was a final project by a group of students using the Flow Imaging Kit I got with the system. Experimentally this was very successful! I made a simple batch of doped Gd water (~1:100 Gd agent:water) which was effective. + +## Future Ideas + +* **Simple gradient echo sequence**: It would have been nice to have a simple GRE sequence to demonstrate, for example, Ernst angle, T1 contrast, and T2* effects. Probably this is not very SNR efficient. It may also exist on the system but not as one of the main Apps. +* **EPI**: I'm not sure how well this would work on the system, but this is a workhorse sequence of modern MRI. +* **Pulse programming**: I did not get to this, but it would be a great to try to build new sequences to test, and also a way to show the students how the sequences are built and how they can be modified. It seems doable, but thus far have not had the time to do it. + + diff --git a/_posts/2025-09-17-Lung_MRI.md b/_posts/2025-09-17-Lung_MRI.md new file mode 100644 index 0000000..9ea59c6 --- /dev/null +++ b/_posts/2025-09-17-Lung_MRI.md @@ -0,0 +1,56 @@ +--- +layout: post +title: "My Journey into Lung MRI" +author: peder +categories: [ research ] +image: assets/images/lungMRI-logo_sample_images.png +featured: true +--- + +I was recently asked to participate in a "Pulmonary MR Virtual Office Hours", a session intended to allow experienced researchers to connect with others in the field and share their experience. +This is a new initiative by the newly formed ISMRM Pulmonary MR Study Group. It got me thinking about my own journey into lung MRI research as well as my current opinions in the area, which I thought I would share here. + +## The Journey + +I got into lung MRI very surreptitiously through a collaborator on a project that was given to me, and had no idea where it would go. UCSF had purchased a new PET/MR system, and I'd been tasked with developing sequences and research projects on the system. Dr. Thomas Hope identified an unmet need of pulmonary nodule detection when performing PET/MR, and also had seen Prof. Kevin Johnson's recently published seminal paper on [Optimized 3D ultrashort echo time pulmonary MRI](https://pmc.ncbi.nlm.nih.gov/articles/PMC4199575/). Prof. Johnson was a very willing collaborator and we were able to execute a successful study on [Detection of Small Pulmonary Nodules with Ultrashort Echo Time Sequences in Oncology Patients by Using a PET/MR System](https://pmc.ncbi.nlm.nih.gov/articles/PMC4699498/). + +Following this, Prof. Miki Lustig and myself expanded our collaboration with Kevin Johnson, who identified that motion artifacts were a key limitation when using UTE in the lungs. We started with several challenging pediatric imaging datasets from UW-Madison, and worked on [Motion robust high resolution 3D free‐breathing pulmonary MRI using dynamic 3D image self‐navigator](https://pmc.ncbi.nlm.nih.gov/articles/PMC6474413/). Eventually, we also started doing many more clinical studies at UCSF, with a focus on the challenging but impactful pediatric population, and built off these datasets to build new tools for high-resolution reconstructions in the presence of motion, including [Iterative motion‐compensation reconstruction ultra‐short TE (iMoCo UTE) for high‐resolution free‐breathing pulmonary MRI](https://pmc.ncbi.nlm.nih.gov/articles/PMC6949392/) which has be come a popular tool in the field. + +![iMoCo Pipeline](../assets/images/iMoCo-pipeline.jpg) + +![MoCoLoR reconstructions](../assets/images/MoCoLoR-examples.jpg) + +I learned from this journey to have an open mind about new research ideas, especially when working with collaborators you trust, and I try to maintain some time to explore some more high-risk, unfunded research directions. I also learned humility about pulse sequence programming - my success in this area has hinged in the amazing pulmonary UTE sequence from Kevin Johnson at UW-Madison, and if we tried to reproduce this ourselves I'm not sure I would still be in the field! + +## Current Priorities + +I am currently very excited about the potential for lung MRI. Our current priorities include functional contrasts and mid-field MRI. + +In numerous conversations with radiologists and pulmonologists I received feedback that, for MRI to be successful, it needs to go beyond what CT can provide. I believe MRI will never achieve the spatial resolution and speed of CT, but I think it can succeed by providing an array of contrasts. Reduced radiation exposure is also an advantage, although modern low-dose protocols and photon-counting CT systems are impressive in using very small doses of radiation. + +### Functional Imaging + +One theme from clinicians was they would want to see functional contrast, such as ventilation. We focused on performing respiratory-resolved reconstructions of free-breathing UTE data as in [Motion‐compensated low‐rank reconstruction for simultaneous structural and functional UTE lung MRI](https://pmc.ncbi.nlm.nih.gov/articles/PMC10501714/), from which ventilation maps can be derived. We have also begun to get experience with other exciting ventilation/perfusion mapping methods such as [Phase-resolved functional lung (PREFUL) MRI Method](https://doi.org/10.1002/mrm.26893). + +![Comparison of UTE and Hyperpolarized Xenon Ventilation](../assets/images/ventilation-UTE_vs_HPX.jpg) + +Based on a collaboration with Dr. Jonathan Rayment and Prof. Rachel Eddy at University of British Columbia [Three‐Dimensional Free‐Breathing Ultrashort Echo Time (UTE) 1H MRI Regional Ventilation: Comparison With Hyperpolarized 129Xe MRI and Pulmonary Function Testing in Healthy Volunteers and People With Cystic Fibrosis](https://pubmed.ncbi.nlm.nih.gov/40235063/), I don't think a UTE or other 1H approach is as sensitive as hyperpolarized xenon-129 gas MRI, but they have the advantage that it can be added onto existing MRI scans with no extra hardware or agents. + +### Mid-field MRI + +UCSF recently purchased a "mid-field" MRI system, the Siemens Healthineers 0.55T Free.Max. While many applications suffer from the reduced polarization compared to 1.5 and 3 T MRI scanners, I think pulmonary imaging is actually better at this field strength. This is due to the reduced susceptibility effects at the numerous air-tissue interfaces in the lungs, which results in for longer T2/T2* relaxation times. Not only does this counteract some of hte polarization loss in maintaining SNR, it also allows for other contrasts such as T2-weighted and diffusion-weighted imaging. A bonus is that mid-field systems have lower magnet and siting costs, making them more accessible. + +We are actively deploying lung MRI protocols at 0.55T, testing and running clinical scans, and determining how we can turn this system into a viable modality for clinical lung imaging. + +## The Future + +What does the future hold for lung MRI? I'm excited about the expansion of the lung MRI research community, both due to the free.max and similar mid-field systems as well as the recent FDA approval of hyperpolarized xenon-129 gas MRI (HPX). There were some excellent questions at the Virtual Office Hours that revealed areas where I think there is important work to do to move the field forward, including motion management, pulse sequence support, and leveraging open-source software. + +Managing motion in the lungs for clinical studies is still very challenging, particularly for our patients with compromised lung function who may breath irregularly, cough, and not be able to perform breath-holds. Still one of the challenges faced in my groups work is how to extract reliable estimates of motion. I think it is something that is not always well definied in the literature, and unspoken challenge that we all assume is a solved problem. Clinical deployment requires high reliabilities, I'd guess upwards of 95% success rates, and I think we are not there yet. + +Lung MRI also suffers from relatively poor vendor support. +Most vendors do not provide FDA-approved pulse sequences (e.g. UTE) that work well for lung MRI, although there are generally works-in-progress (WIP) research prototypes available. I think some hesitation is that the demand for lung MRI scans is small, so there is not a significant financial incentive. The field has benefitted immensively from Prof. Kevin Johnson's GE scanner sequencer, and I've also seen some great utility of the Siemens stack-of-spirals UTE sequence. We can work with clinicians to increase demand and also encourage vendor support whenever possible. + +I'm a huge proponent of open-source software, founding the [PulmonaryMRI GitHub Organization](https://github.com/PulmonaryMRI/), and I think we are in a bit of a golden age for open-source MRI. It is now widely accepted as beneficial to advancing MRI, and the introduction of open-source pulse sequence programming via tools such as Pulseq has opened up this realm that was historically tied into specific vendor environments. https://github.com/PulmonaryMRI/ currently hosts mostly reconstruction software, and I would love to see some more pulse sequences as well as image analysis tools become available. + +Finally, I've been impressed and fascinated by the progress of Hyperpolarized Xenon-129 gas MRI (HPX) as now a clinically reimbursable study. This is potentially a game-changer for lung MRI, if HPX can be a robust study ordered by clinicians in their normal workflow. The next couple of years will be important to see if this transition can occur from research to the clinic. Who knows, maybe I'll dabble in HPX someday! \ No newline at end of file diff --git a/_posts/2026-01-09-Teaching-Principles-of-MRI.md b/_posts/2026-01-09-Teaching-Principles-of-MRI.md new file mode 100644 index 0000000..c479e38 --- /dev/null +++ b/_posts/2026-01-09-Teaching-Principles-of-MRI.md @@ -0,0 +1,53 @@ +--- +layout: post +title: "Teaching MRI and Book" +author: peder +categories: [ education ] +image: assets/images/MRI_logo-retro.png +featured: true +--- + +For the last 15(!) years I’ve been teaching an introductory MRI course to graduate students, and I struggled to find a textbook that was both rigorous but also accessible to students with a wide variety of coming from backgrounds, including engineering and physics but also biology, chemistry, and neuroscience. Within the past year I decided to finalize my own textbook, which I am now proud to share as an online and open source (including code for generating figures and plots shown) book: + +[Principles of MRI](https://larsonlab.github.io/MRI-education-resources/Introduction.html) + +I just finished my first year relying solely on this book, and am writing this to share my justification for creating a new book, my experience using it, and also to encourage feedback. You can also read my previous post about teaching MRI which includes some similar discussion points [Learning MRI (with lectures too)](../Learning-Principles-of-MRI/). + +## Finding a Book Suitable for Engineers, Physicists, Biologists, Chemists, Neuroscientists, and More + +The target audience of my course are students in the [UCSF Masters of Science in Biomedical Imaging program](https://radiology.ucsf.edu/education/graduate-programs/msbi-program) that includes a broad range of backgrounds and aims to teach the fundamental principles of biomedical imaging modalities as well as how imaging is used in clinical applications. The program does not require prerequisites in engineering. The students come from varied backgrounds including engineering, physics, biology, chemistry, and neuroscience. They take this coures in their first quarter of the program. + +I first taught with [**Principles of MRI** by Dwight Nishimura](https://www.lulu.com/shop/dwight-nishimura/principles-of-magnetic-resonance-imaging/paperback/product-6355103.html?page=1&pageSize=4), the textbook written by my PhD advisor and what I learned MRI from. It is a great text - concise, consistent, clear, and rigorous - but my non-engineering students struggled. +In retrospect, I think I took for granted the engineering principles and also the signals and systems perspective that I already had when learning MRI, but many of my students did not have. However, I leaned heavily on the material from this book when creating my book. + +I also tried a few other books that are written for a learners without an engineering background, including [**MRI: From Picture to Proton** by McRobbie, Moore, Prince, and Graves](https://doi.org/10.1017/9781107706958) and [**MRI: The Basics** by Hashemi, Lisanti, and Bradley](https://shop.lww.com/MRI--The-Basics/p/9781496384355). +These books used lots of intuitive and visual examples to explain MRI concepts, as well as included many example images, which I think resonated with many of my students. +However, I found they lacked the mathematical rigor, including consistency of notation and fundamental signal equations, that I wanted to use when teaching MRI. + +## My Book + +In [Principles of MRI](https://larsonlab.github.io/MRI-education-resources/Introduction.html), I sought to create a book that was both rigorous and accurate, but also accessible to all of my students. The Learning Goals are stated in the Introduction, and generally aim towards a practical understanding of MRI such that students can be comfortable and competent running MRI scanners and working with MRI images. + +It includes key equations that I found I wanted to use in lectures and homework assignments, but often I have skipped over derivations or in some cases used simplified forms of equations. + +I created simulations to illustrate and provide visual examples of concepts. + +There are several practical sections, including tables of values and a few example images. + +In the Fall of 2025, I taught solely based on this book, and made a "final" round of edits to fill in any gaps (or eliminate material that was too complex), and now am reasonably confident in this version of the book. + + +## Online and Open Source + +When I first saw the use of the JupyterBook format, I was blow away. This, to me, is the future of publishing, as it can leverage text, images, simulations, and even interactive content. I experimented over the past 5 or so years with this format, which formed the basis of this book. + +The code is all there to use as is or modify. In homeworks, I have asked students to build off of these examples and experiment or make modifications. + +The format also allows for community contributions and feedback. If you use the book, I sincerely hope you will consider using this feature. +You can message me or leave a GitHub issue with comments, or even create a pull request for content changes you want to suggest. + +## Next Steps + +I still have some self doubt about this book. I wonder if it really was worth it, and whether I should have stuck with, for example, Dwight Nishimura's book, and provided any supplementary materials needed to support this. I worry I have made errors that will propagate! But I also really had a lot of fun creating this book, so maybe that in itself makes it worth it. + +I am in the midst of the next evolution of the course to include significant demonstrations using the [Ilumr](https://resonint.com/ilumr) tabletop MRI system. You can read about my experience [Tabletop MRI for Education - Teaching Experience](../Tabletop_MRI_Education/), and I aspire to release descriptions of demos I do on this scanner in class that can accompany the book. \ No newline at end of file diff --git a/assets/audio/UTE golden spaceship.wav b/assets/audio/UTE golden spaceship.wav new file mode 100644 index 0000000..493f463 Binary files /dev/null and b/assets/audio/UTE golden spaceship.wav differ diff --git a/assets/audio/UTE reverse spaceship.wav b/assets/audio/UTE reverse spaceship.wav new file mode 100644 index 0000000..19a8ca5 Binary files /dev/null and b/assets/audio/UTE reverse spaceship.wav differ diff --git a/assets/images/CT_MRI_bone.jpg b/assets/images/CT_MRI_bone.jpg new file mode 100644 index 0000000..797053f Binary files /dev/null and b/assets/images/CT_MRI_bone.jpg differ diff --git a/assets/images/Dixon_sCT.jpg b/assets/images/Dixon_sCT.jpg new file mode 100644 index 0000000..4edc6cc Binary files /dev/null and b/assets/images/Dixon_sCT.jpg differ diff --git a/assets/images/EPI sequence.jpg b/assets/images/EPI sequence.jpg new file mode 100644 index 0000000..9c91f20 Binary files /dev/null and b/assets/images/EPI sequence.jpg differ diff --git a/assets/images/HPBrain_MRI-Spaceship.gif b/assets/images/HPBrain_MRI-Spaceship.gif new file mode 100644 index 0000000..5f2eb36 Binary files /dev/null and b/assets/images/HPBrain_MRI-Spaceship.gif differ diff --git a/assets/images/HPBrain_MRI-Spaceship.jpg b/assets/images/HPBrain_MRI-Spaceship.jpg new file mode 100644 index 0000000..b44704f Binary files /dev/null and b/assets/images/HPBrain_MRI-Spaceship.jpg differ diff --git a/assets/images/HPBrain_volcano-moon.jpg b/assets/images/HPBrain_volcano-moon.jpg new file mode 100644 index 0000000..a176393 Binary files /dev/null and b/assets/images/HPBrain_volcano-moon.jpg differ diff --git a/assets/images/LarsonLab.GitHub.io_QRcode.png b/assets/images/LarsonLab.GitHub.io_QRcode.png new file mode 100644 index 0000000..644a9a6 Binary files /dev/null and b/assets/images/LarsonLab.GitHub.io_QRcode.png differ diff --git a/assets/images/MR-CT-dose_comparison.jpg b/assets/images/MR-CT-dose_comparison.jpg new file mode 100644 index 0000000..adb6136 Binary files /dev/null and b/assets/images/MR-CT-dose_comparison.jpg differ diff --git a/assets/images/MRI-spaceship-1.jpg b/assets/images/MRI-spaceship-1.jpg new file mode 100644 index 0000000..3c93f4f Binary files /dev/null and b/assets/images/MRI-spaceship-1.jpg differ diff --git a/assets/images/MRI-spaceship-2.jpg b/assets/images/MRI-spaceship-2.jpg new file mode 100644 index 0000000..189fdec Binary files /dev/null and b/assets/images/MRI-spaceship-2.jpg differ diff --git a/assets/images/MRI_logo-retro.png b/assets/images/MRI_logo-retro.png new file mode 100644 index 0000000..de39b7a Binary files /dev/null and b/assets/images/MRI_logo-retro.png differ diff --git a/assets/images/MoCoLoR-examples.jpg b/assets/images/MoCoLoR-examples.jpg new file mode 100644 index 0000000..4714e17 Binary files /dev/null and b/assets/images/MoCoLoR-examples.jpg differ diff --git a/assets/images/Spin-echo diffusion sequence.jpg b/assets/images/Spin-echo diffusion sequence.jpg new file mode 100644 index 0000000..9b26fec Binary files /dev/null and b/assets/images/Spin-echo diffusion sequence.jpg differ diff --git a/assets/images/The Sound of MRI.jpg b/assets/images/The Sound of MRI.jpg new file mode 100644 index 0000000..7fee37a Binary files /dev/null and b/assets/images/The Sound of MRI.jpg differ diff --git a/assets/images/UTE-surface_coil.png b/assets/images/UTE-surface_coil.png new file mode 100644 index 0000000..f55e028 Binary files /dev/null and b/assets/images/UTE-surface_coil.png differ diff --git a/assets/images/UTE-wood_headphones.png b/assets/images/UTE-wood_headphones.png new file mode 100644 index 0000000..387885a Binary files /dev/null and b/assets/images/UTE-wood_headphones.png differ diff --git a/assets/images/UTE-wood_plastic.jpg b/assets/images/UTE-wood_plastic.jpg new file mode 100644 index 0000000..6a044b3 Binary files /dev/null and b/assets/images/UTE-wood_plastic.jpg differ diff --git a/assets/images/Vida_volunteer.jpg b/assets/images/Vida_volunteer.jpg new file mode 100644 index 0000000..0a5e1a1 Binary files /dev/null and b/assets/images/Vida_volunteer.jpg differ diff --git a/assets/images/ZTE pulse sequence.jpg b/assets/images/ZTE pulse sequence.jpg new file mode 100644 index 0000000..6ce05f2 Binary files /dev/null and b/assets/images/ZTE pulse sequence.jpg differ diff --git a/assets/images/ZTE_RF-coils.jpg b/assets/images/ZTE_RF-coils.jpg new file mode 100644 index 0000000..44c39e0 Binary files /dev/null and b/assets/images/ZTE_RF-coils.jpg differ diff --git a/assets/images/ZTE_dixon.jpg b/assets/images/ZTE_dixon.jpg new file mode 100644 index 0000000..97bdd37 Binary files /dev/null and b/assets/images/ZTE_dixon.jpg differ diff --git a/assets/images/ZTE_rubber.jpg b/assets/images/ZTE_rubber.jpg new file mode 100644 index 0000000..1f5928e Binary files /dev/null and b/assets/images/ZTE_rubber.jpg differ diff --git a/assets/images/iMoCo-infant.jpg b/assets/images/iMoCo-infant.jpg new file mode 100644 index 0000000..982b78b Binary files /dev/null and b/assets/images/iMoCo-infant.jpg differ diff --git a/assets/images/iMoCo-pipeline.jpg b/assets/images/iMoCo-pipeline.jpg new file mode 100644 index 0000000..f090b93 Binary files /dev/null and b/assets/images/iMoCo-pipeline.jpg differ diff --git a/assets/images/ilumr-blackberry.png b/assets/images/ilumr-blackberry.png new file mode 100644 index 0000000..b8ca8af Binary files /dev/null and b/assets/images/ilumr-blackberry.png differ diff --git a/assets/images/ilumr-flower.gif b/assets/images/ilumr-flower.gif new file mode 100644 index 0000000..b3e660d Binary files /dev/null and b/assets/images/ilumr-flower.gif differ diff --git a/assets/images/ilumr_action1.jpg b/assets/images/ilumr_action1.jpg new file mode 100644 index 0000000..7a2d20c Binary files /dev/null and b/assets/images/ilumr_action1.jpg differ diff --git a/assets/images/ilumr_action2.jpg b/assets/images/ilumr_action2.jpg new file mode 100644 index 0000000..63aa4e4 Binary files /dev/null and b/assets/images/ilumr_action2.jpg differ diff --git a/assets/images/ilumr_action3.jpg b/assets/images/ilumr_action3.jpg new file mode 100644 index 0000000..597cd5b Binary files /dev/null and b/assets/images/ilumr_action3.jpg differ diff --git a/assets/images/ilumr_cel2RAREinv.png b/assets/images/ilumr_cel2RAREinv.png new file mode 100644 index 0000000..1c24bf4 Binary files /dev/null and b/assets/images/ilumr_cel2RAREinv.png differ diff --git a/assets/images/ilumr_larson-lab.jpg b/assets/images/ilumr_larson-lab.jpg new file mode 100644 index 0000000..065c1fa Binary files /dev/null and b/assets/images/ilumr_larson-lab.jpg differ diff --git a/assets/images/lungMRI-logo_sample_images.png b/assets/images/lungMRI-logo_sample_images.png new file mode 100644 index 0000000..dd26285 Binary files /dev/null and b/assets/images/lungMRI-logo_sample_images.png differ diff --git a/assets/images/peder-larson-tabletop-MRI.jpg b/assets/images/peder-larson-tabletop-MRI.jpg new file mode 100644 index 0000000..a396159 Binary files /dev/null and b/assets/images/peder-larson-tabletop-MRI.jpg differ diff --git a/assets/images/prostateDWI_Vegetables-under-the-moon.jpg b/assets/images/prostateDWI_Vegetables-under-the-moon.jpg new file mode 100644 index 0000000..5c62273 Binary files /dev/null and b/assets/images/prostateDWI_Vegetables-under-the-moon.jpg differ diff --git a/assets/images/prostateDWI_moon-carrot-wikiart.jpg b/assets/images/prostateDWI_moon-carrot-wikiart.jpg new file mode 100644 index 0000000..429825f Binary files /dev/null and b/assets/images/prostateDWI_moon-carrot-wikiart.jpg differ diff --git a/assets/images/research_projects/ai_cancer.png b/assets/images/research_projects/ai_cancer.png new file mode 100644 index 0000000..09c4e62 Binary files /dev/null and b/assets/images/research_projects/ai_cancer.png differ diff --git a/assets/images/research_projects/hp_kidney1.png b/assets/images/research_projects/hp_kidney1.png new file mode 100644 index 0000000..81ca8e4 Binary files /dev/null and b/assets/images/research_projects/hp_kidney1.png differ diff --git a/assets/images/research_projects/hp_kidney2.png b/assets/images/research_projects/hp_kidney2.png new file mode 100644 index 0000000..c5aa305 Binary files /dev/null and b/assets/images/research_projects/hp_kidney2.png differ diff --git a/assets/images/research_projects/hpcardiac.png b/assets/images/research_projects/hpcardiac.png new file mode 100644 index 0000000..a19a4d4 Binary files /dev/null and b/assets/images/research_projects/hpcardiac.png differ diff --git a/assets/images/research_projects/hpmethods.jpeg b/assets/images/research_projects/hpmethods.jpeg new file mode 100644 index 0000000..4c7be50 Binary files /dev/null and b/assets/images/research_projects/hpmethods.jpeg differ diff --git a/assets/images/research_projects/img_recon_1.png b/assets/images/research_projects/img_recon_1.png new file mode 100644 index 0000000..64c824d Binary files /dev/null and b/assets/images/research_projects/img_recon_1.png differ diff --git a/assets/images/research_projects/img_recon_2.png b/assets/images/research_projects/img_recon_2.png new file mode 100644 index 0000000..55199df Binary files /dev/null and b/assets/images/research_projects/img_recon_2.png differ diff --git a/assets/images/research_projects/lung_mri_1.jpeg b/assets/images/research_projects/lung_mri_1.jpeg new file mode 100644 index 0000000..f090b93 Binary files /dev/null and b/assets/images/research_projects/lung_mri_1.jpeg differ diff --git a/assets/images/research_projects/lung_mri_2.jpg b/assets/images/research_projects/lung_mri_2.jpg new file mode 100644 index 0000000..490499c Binary files /dev/null and b/assets/images/research_projects/lung_mri_2.jpg differ diff --git a/assets/images/research_projects/myelin_ute.jpeg b/assets/images/research_projects/myelin_ute.jpeg new file mode 100644 index 0000000..0894a7d Binary files /dev/null and b/assets/images/research_projects/myelin_ute.jpeg differ diff --git a/assets/images/research_projects/pet_mr.png b/assets/images/research_projects/pet_mr.png new file mode 100644 index 0000000..a2da6df Binary files /dev/null and b/assets/images/research_projects/pet_mr.png differ diff --git a/assets/images/stradivariUTE.png b/assets/images/stradivariUTE.png new file mode 100644 index 0000000..b696703 Binary files /dev/null and b/assets/images/stradivariUTE.png differ diff --git a/assets/images/ventilation-UTE_vs_HPX.jpg b/assets/images/ventilation-UTE_vs_HPX.jpg new file mode 100644 index 0000000..5e07094 Binary files /dev/null and b/assets/images/ventilation-UTE_vs_HPX.jpg differ diff --git a/assets/images/violin_angled.png b/assets/images/violin_angled.png new file mode 100644 index 0000000..48bb769 Binary files /dev/null and b/assets/images/violin_angled.png differ diff --git a/index.html b/index.html index 7e870b3..aa2b111 100644 --- a/index.html +++ b/index.html @@ -12,21 +12,15 @@

    Welcome!

    - Welcome to a developing home for our research group! We love to tackle problems in medical imaging using engineering-driven approaches to improve human health. Our work mostly focuses on MRI (although we dabble in CT and PET) and at or near the stage of human studies. + Welcome to a developing home for our research group, lead by Prof. Peder Larson and working closely with a network of researchers and clinicans. We love to tackle problems in medical imaging using engineering-driven approaches to improve human health. Our work mostly focuses on MRI (although we dabble in CT and PET) and on projects that at or near the stage of human studies.

    - We work on a range of projects, taking on new ones opportunistically, and this currently includes - Metabolic MRI with hyperpolarized contrast agents, - Simultaneous PET/MR imaging systems, - Lung MRI, - Myelin MRI, - Radiation treatment planning, and - Prostate and kidney cancer prediction using deep learning. +We work on a range of projects, taking on new ones opportunistically, and projects currently include Metabolic MRI with hyperpolarized contrast agents, Simultaneous PET/MR imaging systems, Lung MRI, Myelin MRI, Radiation treatment planning, and Prostate and kidney cancer prediction based on historical imaging data.

    - This website is very much in a beta phase, please also visit our UCSF Group Website for more information. + This website focuses on Research Resources, including a blog, research project descriptions, and software packages. More information about the group is available at our UCSF Group Website for more information.

    @@ -56,6 +50,7 @@

    Featured Posts

    {% endif %} +