[go: up one dir, main page]

CN112967778B - Accurate medicine application method and system for inflammatory bowel disease based on machine learning - Google Patents

Accurate medicine application method and system for inflammatory bowel disease based on machine learning Download PDF

Info

Publication number
CN112967778B
CN112967778B CN202110287375.0A CN202110287375A CN112967778B CN 112967778 B CN112967778 B CN 112967778B CN 202110287375 A CN202110287375 A CN 202110287375A CN 112967778 B CN112967778 B CN 112967778B
Authority
CN
China
Prior art keywords
model
image
data
features
disease
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110287375.0A
Other languages
Chinese (zh)
Other versions
CN112967778A (en
Inventor
于泽宽
周锟
耿道颖
吴兴旺
王乐
韩方凯
刘晓
陈卫强
李强
王侠
刘学玲
吕锟
王容
杜鹏
文剑波
韩秋月
张海燕
杜成娟
王娜
李璇璇
吴昊
耿岩
李郁欣
张军
尹波
曹鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huashan Hospital of Fudan University
Original Assignee
Huashan Hospital of Fudan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huashan Hospital of Fudan University filed Critical Huashan Hospital of Fudan University
Priority to CN202110287375.0A priority Critical patent/CN112967778B/en
Publication of CN112967778A publication Critical patent/CN112967778A/en
Application granted granted Critical
Publication of CN112967778B publication Critical patent/CN112967778B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G06F18/24155—Bayesian classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30028—Colon; Small intestine

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Epidemiology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Primary Health Care (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Quality & Reliability (AREA)
  • Computing Systems (AREA)
  • Pathology (AREA)
  • Mathematical Physics (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Medicinal Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention provides a machine learning-based accurate medicine application method and system for inflammatory bowel disease, and relates to the technical field of medical image processing, wherein the method comprises the following steps: step S1: acquiring energy spectrum CT data of inflammatory bowel disease and preprocessing the energy spectrum CT data to manufacture a disease activity-medication label data set; step S2: performing focus region segmentation on the preprocessed energy spectrum CT data; step S3: will be the focus region ROI of inflammatory bowel disease IBD Extracting image histology characteristics; constructing a migration model; step S4: a non-invasive drug evaluation model of inflammatory bowel disease was obtained. The invention can help guide the selection of treatment regimens for IBD patients, assess prognosisFurther shortens the treatment time of IBD patients and has better clinical practicability.

Description

Accurate medicine application method and system for inflammatory bowel disease based on machine learning
Technical Field
The invention relates to the technical field of medical image processing, in particular to a machine learning-based accurate medicine application method and system for inflammatory bowel disease.
Background
Inflammatory bowel disease (inflammatory bowel disease, IBD) is a clinically common chronic, non-specific inflammatory disease of the intestinal tract. Along with the severity of the development of inflammatory bowel disease, different clinical treatment schemes can be caused, so that the accurate evaluation of inflammatory bowel disease has important significance for the curative effect and personalized treatment of medicines. Endoscopy is an effective means of estimating and assessing disease and therapeutic outcome clinically. However, the use of endoscopes is limited by the complex operation of the endoscope, the poor patient tolerance, and the inability to fully assess intestinal lesions. Therefore, finding an automatic inflammatory bowel disease condition assessment and noninvasive curative effect assessment method has important significance for clinical diagnosis and accurate treatment of inflammatory bowel disease.
The invention discloses a Chinese patent with publication number of CN110880361A, which discloses a personalized accurate medication recommendation method and a device, comprising: acquiring medical record data of a plurality of patients suffering from the same disease, wherein the medical record data comprises structural data, text data and image data; obtaining medication information of a patient from the text data; screening and obtaining a first medicine recommendation result of the target patient from the medicine information of a plurality of historical patients; combining and processing the medical record data of the patient to obtain characteristic information of the patient; screening at least one similar patient similar to the current disease characteristic information of the target patient from a plurality of historical patients; generating a second medicine recommendation result according to the medicine information of the similar patients; and obtaining the personalized medicine recommendation result of the target patient according to the first medicine recommendation result and the second medicine recommendation result. The technical scheme provided by the embodiment of the invention can solve the problem of low medicine application accuracy of patients in the prior art.
In recent years, the continuous maturation of computer technology provides conditions for the perfection and development of CT technology. The advent of spectral CT provides a new opportunity for non-invasive efficacy assessment of inflammatory bowel disease. The energy spectrum CT adopts a precious stone detector, a high-voltage instantaneous energy-changing generator and a dynamic zoom tube on the basis of the common CT, can well overcome the defects of low density resolution, hardening artifact, isophytic artifact, foreign object isophytic artifact and the like of the conventional CT, and acquires an optimal single energy image so as to distinguish and quantify substances. Currently, CDAI scores are used clinically for the assessment of Crohn's Disease (CD) in inflammatory bowel Disease. This scoring involves a variety of clinical criteria, the collection is complex and the results may not be consistent with objective disease. If various examinations indicate serious lesions, but CDAI is very low; CDAI is high but patient objective disease is mild. Current latest scoring systems include MRE scoring, CE-US scoring, and CE scoring, which, although simple in parameter collection, do not improve scoring accuracy. There is a lack of an automatic classification model that is more accurate by studying the imaging features.
The current medical modalities for IBD are mostly met by one treatment regimen for all patients, although with adequate efficacy for some patients, most patients with poor efficacy. In recent years, accurate medical treatment is continuously accepted and adopted, and the accurate medical treatment can customize the treatment method according to the personalized characteristics of each patient. In the selection of the treatment scheme of IBD, a precise medical method can be adopted, and the influence of individual molecular biology and environmental factors on the curative effect of the treatment scheme is considered to promote better clinical decision. However, the current research lacks a unified mapping between the severity of inflammatory bowel disease and the medication guidance, and is difficult to construct a proper inflammatory bowel disease drug evaluation model so as to achieve the aim of accurate treatment.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a machine learning-based accurate medicine application method and system for inflammatory bowel disease.
According to the accurate medicine application method and system for inflammatory bowel disease based on machine learning, the scheme is as follows:
in a first aspect, a machine learning-based method of accurately administering an inflammatory bowel disease is provided, the method comprising:
step S1: acquiring energy spectrum CT data of inflammatory bowel disease, preprocessing the energy spectrum CT data, and manufacturing a disease activity-medication label data set based on the processed data;
step S2: the focus region segmentation is carried out on the pretreated energy spectrum CT data by using an end-to-end convolutional neural network to obtain the focus region ROI of inflammatory bowel disease IBD ；
Step S3: will be the focus region ROI of inflammatory bowel disease IBD Performing image histology feature extraction to obtain a feature s= { s 1 ,s 2 ,…,s n }，n＝1,2，…，176；
Constructing a migration model, namely performing target task T classification on 176 features, migrating the target task T classification to a migration model performing target task T classification on 60 features, and obtaining 60 features most relevant to clinical medication efficacy through iterative optimization;
step S4: and integrating the selected 60 most relevant features, training a plurality of classifiers by using an image histology method, and training the classifiers by using an AdaBoost integrated learning method to obtain a noninvasive drug evaluation model for inflammatory bowel disease.
Preferably, the step S1 specifically includes the following steps:
through a CTE sequence thin layer scanning mode, energy spectrum CT adopts high 140kV and low 80kV energy instantaneous switching, and according to sampling data obtained under the two energies, the attenuation coefficient of a voxel in the energy range of 40-140 keV is determined to obtain 101 single energy diagrams, and the scanning layer thickness and the layer spacing are 5mm;
the energy spectrum CT adopts an adaptive statistical iterative reconstruction algorithm (daptive statistical iterative reconstruction, ASIR) and a filtered back projection algorithm (filtered back projection, FBP) to reconstruct a single energy image by combining ASiR with the level of 40 percent, namely an image is formed by mixing ASiR and FBP with the weight of 4:6, and the reconstruction layer thickness and layer spacing are 1.25mm;
data desensitization is carried out on the obtained reconstructed image, and an irregular image is removed;
preprocessing training data, adopting a de-averaging operation, and normalizing the amplitude of the data within the same range;
the disease condition-medication standard is formulated, and different medicines are adopted for treatment according to different severity of IBD disease.
Preferably, in the step S2, the focus area segmentation is performed on the obtained energy spectrum CT image, which specifically includes the following steps:
the CT image is subjected to downsampling and feature value extraction, the convolution kernel size is 3 multiplied by 3, the step length is 2, and max-pooling operation is performed after each group of convolution operation, so that the image is further reduced to 1/2 of the original image;
the method comprises the steps of carrying out a total of 4 times in the downsampling process to obtain characteristics, and finally obtaining 256 downsampling characteristic graphs of 16 multiplied by 16, wherein the number of convolution kernels used by each convolution layer is 32, 64, 128 and 256 respectively;
4 groups of deconvolution operations are used in the up-sampling process to expand the picture to 2 times of the original picture, and the characteristic diagram of the corresponding layer is cut and copied to form an up-convolution result;
after the up-sampling process is finished, a graph with the size of 256 multiplied by 256 is obtained, finally, a convolution kernel with the size of 1 multiplied by 1 is used for reducing the number of channels to 2, different labels are used for marking, and CT image segmentation is completed, so that the CT image is obtainedROI IBD Wherein the number of convolution kernels used for each convolution layer is 256, 128, 64, 32, respectively;
and evaluating the segmentation result by adopting a loss function in the model training process.
Preferably, the loss function is constructed as follows:
L total ＝L dice +L CE
wherein the method comprises the steps ofX and Y are model segmentation results and doctor labeling results for comparing similarity between the model segmentation results and the doctor labeling results, and the range is 0,1]The method comprises the steps of carrying out a first treatment on the surface of the M represents the number of categories, y c Is a one-hot vector, the element has only 0 or 1 two values, if the category is the same as the category of the sample, 1 is taken, otherwise, 0 is taken; p (P) c Representing the probability that the predicted sample belongs to c.
Preferably, the step S3 includes:
using WORC toolkit method for the ROI obtained in step S2 IBD Extracting image characteristics from the energy spectrum CT of the image, and quantifying the information contained in the image;
and constructing a migration model by adopting SVM and a range loss function.
Preferably, the constructing the migration model by using the SVM and the range loss function includes:
defining D as a set of 176 image features and disease degree labels, and the source task s as a disease classification prediction task and labeling the disease degree label y t = { CD remission, CD mild, CD moderate, CD severe, UC remission, UC mild, UC moderate, UC severe } is defined as {0,1,2,3,4,5,6,7}, target task t = s is defined as a multi-classification task;
defining a migration model for classifying target tasks T by 176 features and migrating to classifying target tasks T by T featuresThe model finds a feature set most relevant to the clinical medication curative effect, so that s is migrated to the multi-classification task t to optimize the overall performance;
for all (s, y t ) E D, IBD typing class-associated migration modelIs-> Wherein s is an input feature, phi(s) is a feature selector, F t As an SVM classifier, a loss function adopts a range loss function; finally, t (t=60) features most relevant to the clinical medication efficacy are obtained through continuous iterative optimization.
Preferably, the step S4 includes:
classifying IBD disease conditions through image features, and inputting the 60 features which are selected in the step S3 and most relevant to clinical medication curative effects into a plurality of basic classifiers for training;
adopting AdaBoost integrated learning algorithm, inputting each characteristic x= [ x ] defined as the previous step selection 1 ，x 2 ，…x k ]K=1, 2, …, k, where k=60 is the feature number;
the output is defined as a multi-classification problem, i.e. s t = {0,1,2,3,4,5,6,7}; the n basis classifiers are defined as h t (x) T=1, 2, …, n, x is an example feature;
the learned combined classifier is defined as a linear combination of 4 basic classifiersBy continually updating the iterations, when the base classifier h t (x) Based on distribution D t After generation, the classifier weights alpha t So that alpha is t h t (x) Is>Minimum;
each iteration utilizes the next round of basic classifier h t (x) Correction of H t-1 (x) Error, i.e. minimizing
After model training is completed, data of a test set is input, and for each case, the model automatically outputs a column vector s t ＝{s 1 ，s 2 ，…，s n }，n＝7；
Label y corresponding to the degree of illness t The highest value indicates the recommended treatment plan, thereby completing the goal of automatic accurate medication.
In a second aspect, a machine learning based accurate medication system for inflammatory bowel disease is provided, the system comprising:
model M1: acquiring energy spectrum CT data of inflammatory bowel disease, preprocessing the energy spectrum CT data, and manufacturing a disease activity-medication label data set based on the processed data;
model M2: the focus region segmentation is carried out on the pretreated energy spectrum CT data by using an end-to-end convolutional neural network to obtain the focus region ROI of inflammatory bowel disease IBD ；
Model M3: will be the focus region ROI of inflammatory bowel disease IBD Performing image histology feature extraction to obtain a feature s= { s 1 ,s 2 ,…,s n }，n＝1,2，…，176；
Constructing a migration model, namely performing target task T classification on 176 features, migrating the target task T classification to a migration model performing target task T classification on 60 features, and obtaining 60 features most relevant to clinical medication efficacy through iterative optimization;
model M4: and integrating the selected 60 most relevant features, training a plurality of classifiers by using an image histology method, and training the classifiers by using an AdaBoost integrated learning method to obtain a noninvasive drug evaluation model for inflammatory bowel disease.
Preferably, the module M1 is specifically as follows:
module M1.1: through a CTE sequence thin layer scanning mode, energy spectrum CT adopts high 140kV and low 80kV energy instantaneous switching, and according to sampling data obtained under the two energies, the attenuation coefficient of a voxel in the energy range of 40-140 keV is determined to obtain 101 single energy diagrams, and the scanning layer thickness and the layer spacing are 5mm;
module M1.2: the energy spectrum CT adopts an adaptive statistical iterative reconstruction algorithm (daptive statistical iterative reconstruction, ASIR) and a filtered back projection algorithm (filtered back projection, FBP) to reconstruct a single energy image by combining ASiR with the level of 40 percent, namely an image is formed by mixing ASiR and FBP with the weight of 4:6, and the reconstruction layer thickness and layer spacing are 1.25mm;
module M1.3: data desensitization is carried out on the obtained reconstructed image, and an irregular image is removed;
module M1.4: preprocessing training data, adopting a de-averaging operation, and normalizing the amplitude of the data within the same range;
module M1.5: the disease condition-medication standard is formulated, and different medicines are adopted for treatment according to different severity of IBD disease.
Preferably, the module M2 performs focal region segmentation on the obtained energy spectrum CT image, which specifically includes the following steps:
module M2.1: the CT image is subjected to downsampling and feature value extraction, the convolution kernel size is 3 multiplied by 3, the step length is 2, and max-pooling operation is performed after each group of convolution operation, so that the image is further reduced to 1/2 of the original image;
the method comprises the steps of carrying out a total of 4 times in the downsampling process to obtain characteristics, and finally obtaining 256 downsampling characteristic graphs of 16 multiplied by 16, wherein the number of convolution kernels used by each convolution layer is 32, 64, 128 and 256 respectively;
module M2.2: 4 groups of deconvolution operations are used in the up-sampling process to expand the picture to 2 times of the original picture, and the characteristic diagram of the corresponding layer is cut and copied to form an up-convolution result;
after the up-sampling process is finished, a graph with the size of 256 multiplied by 256 is obtained, finally, a convolution kernel with the size of 1 multiplied by 1 is used for reducing the number of channels to 2, different labels are used for marking, CT picture segmentation is completed, and the ROI is obtained IBD Wherein the number of convolution kernels used for each convolution layer is 256, 128, 64, 32, respectively;
module M2.3: and evaluating the segmentation result by adopting a loss function in the model training process.
Compared with the prior art, the invention has the following beneficial effects:
1. the convolutional neural network is applied to the disease classification of IBD, so that the noninvasive curative effect artificial intelligent evaluation based on the energy spectrum CT image is realized, and various limitations caused by complex scoring adopted clinically are overcome;
2. the IBD accurate drug administration method provided by the invention has higher sensitivity and accuracy, can help guide the selection of the treatment scheme of the IBD patient, evaluate the prognosis, further shorten the treatment time of the IBD patient, and has better clinical practicability.
Drawings
Other features, objects and advantages of the present invention will become more apparent upon reading of the detailed description of non-limiting embodiments, given with reference to the accompanying drawings in which:
FIG. 1 is a flow chart of the overall process of the present invention.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the present invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications could be made by those skilled in the art without departing from the inventive concept. These are all within the scope of the present invention.
The embodiment of the invention provides a machine learning-based accurate medicine application method for inflammatory bowel disease, which comprises the following steps of;
example 1:
in order to solve the problems of complex evaluation of inflammatory bowel disease conditions and irregular medication in clinic, the invention provides a machine learning-based accurate medication method for inflammatory bowel disease, which comprises the following steps:
step S1: the method comprises the steps of obtaining energy spectrum CT data of inflammatory bowel disease and preprocessing the energy spectrum CT data, wherein the preprocessing comprises the steps of reconstructing a small intestine vein phase CT image with the layer thickness of 1.25mm and the layer spacing of 0.8cm by adopting 40% ASiR, and performing desensitization, cleaning, resampling and the like on the CT data to obtain CT data with uniform resolution and same gray scale distribution. Creating a patient activity-medication label dataset based on the processed data, specifically:
by adopting a CTE sequence thin layer scanning mode, the energy spectrum CT adopts high 140kV and low 80kV energy instantaneous switching, and according to sampling data obtained under the two energies, the attenuation coefficient of a voxel in the energy range of 40-140 keV is determined to obtain 101 single energy diagrams, and the scanning layer thickness and the layer spacing are 5mm. The energy spectrum CT adopts an adaptive statistical iterative reconstruction algorithm (daptive statistical iterative reconstruction, ASIR) and a filtered back projection algorithm (filtered back projection, FBP) to reconstruct a single energy image by combining ASiR with a level of 40%, namely an image is formed by mixing ASiR and FBP with a weight of 4:6, the reconstruction layer thickness and the layer spacing are 1.25mm, and compared with the conventional CT, the image noise can be reduced, the density resolution can be improved, and the radiation dose can be reduced.
And (3) performing data desensitization on the acquired reconstructed image and eliminating the nonstandard image so as to improve the consistency of the data. Preprocessing training data, adopting a de-averaging operation, normalizing the amplitude of the data within the same range, and reducing interference caused by the difference of the value ranges of the data in each dimension.
The disease condition-medication standard is formulated, and different medicines are adopted for treatment according to different severity of IBD disease. IBD is classified as ulcerative colitis (Ulcerative Colitis, UC) and Crohn's Disease (CD), wherein ulcerative colitis and Crohn's Disease are classified as remission, mild, moderate and severe depending on the severity of the condition. Formulating the disease degree label y t = { CD remission, CD mild, CD moderate, CD severe, UC remission, UC mild, UC moderate, UC severe }. According to 2018 diagnosis and treatment of inflammatory bowel diseaseThe opinion is formulated into drug administration standard, and the 5-aminosalicylic acid drug is suitable for mild colon type patients. Glucocorticoids are used in patients with more than moderate CD. Severe patients were hormone treated with complete treatment regimens as shown in table 1:
table 1: treatment regimen for IBD
Step S2: the focus region segmentation is carried out on the pretreated energy spectrum CT data by using an end-to-end convolutional neural network to obtain the focus region ROI of inflammatory bowel disease IBD . The focus region segmentation is carried out on the obtained energy spectrum CT image, and the method concretely comprises the following steps:
and (3) carrying out downsampling on the CT image to extract characteristic values, wherein the convolution kernel size is 3 multiplied by 3, the step size is 2, and carrying out max-pooling operation after each group of convolution operation to further shrink the image to 1/2 of the original image, and carrying out total 4 times in the downsampling process to obtain characteristics, so as to finally obtain 256 downsampling characteristic diagrams of 16 multiplied by 16, wherein the number of convolution kernels used by each convolution layer is 32, 64, 128 and 256 respectively. The up-sampling process uses 4 groups of deconvolution operations to expand the picture to 2 times of the original picture, cuts and replicates the feature pictures of the corresponding layers to form an up-convolution result, after the up-sampling process is finished, a picture with the size of 256 multiplied by 256 is obtained, finally, a convolution kernel with the size of 1 multiplied by 1 is used for reducing the number of channels to 2, different labels are used for marking, CT picture segmentation is completed, and the ROI is obtained IBD Wherein the number of convolution kernels used for each convolution layer is 256, 128, 64, 32, respectively.
And evaluating the segmentation result by adopting a loss function in the model training process. Wherein, the loss function is constructed as follows:
L total ＝L dice +L CE
wherein the method comprises the steps ofX and Y are model segmentation results and doctor labeling results for comparing similarity between the model segmentation results and the doctor labeling results, and the range is 0,1]The method comprises the steps of carrying out a first treatment on the surface of the M represents the number of categories, y c Is a one-hot vector, the element has only 0 or 1 two values, if the category is the same as the category of the sample, 1 is taken, otherwise, 0 is taken; p (P) c Representing the probability that the predicted sample belongs to c.
Step S3: will be the focus region ROI of inflammatory bowel disease IBD Performing image histology feature extraction to obtain a feature s= { s 1 ,s 2 ,…,s n N=1, 2, …,176. Constructing a migration model, and migrating 176 features to a migration model with 60 features for target task T classification, wherein 60 features most relevant to clinical medication efficacy are obtained through iterative optimization, and the method comprises the following steps of:
using WORC toolkit method for the ROI obtained in step S2 IBD The energy spectrum CT of the image is used for extracting image features and quantifying the information contained in the image. In order to avoid that some variables in the features become dominant due to larger measurement units, the data features are normalized to obtain 176 image features including 15 shape features, 18 gray features, 20 texture features, 120 wavelet features and 3 clinical medical record features. The complete feature types are shown in table 2:
table 2: feature selection
The migration model is built by adopting SVM and range loss functions, in particular:
defining D as a set of 176 image features and disease degree labels, and the source task s as a disease classification prediction task and labeling the disease degree label y t = { CD remission, CD mild, CD moderate, CD severe, UC remission, UC mild, UC moderate, UC severe } is defined as {0,1,2,3,4,5,6,7}, target task t=s is defined as a multi-classification task.
Definition of target task T by 176 featuresClassification, migration to migration model for target task T classification by T featuresThe model finds a feature set most relevant to the clinical medication efficacy, so that s is migrated to the multi-classification task t to optimize the overall performance.
For all (s, y t ) E D, IBD typing class-associated migration modelIs-> Wherein s is an input feature, phi(s) is a feature selector, F t As an SVM classifier, a loss function adopts a range loss function; finally, t (t=60) features most relevant to the clinical medication efficacy are obtained through continuous iterative optimization.
Step S4: integrating the selected 60 most relevant features, training a plurality of classifiers by using an image histology method, and training the classifiers by using an AdaBoost integrated learning method to obtain a noninvasive drug evaluation model of inflammatory bowel disease, wherein the method comprises the following steps of:
IBD disease classification is carried out through image features, and the 60 features which are selected in the step S3 and are most relevant to clinical medication curative effect are input into four basic classifiers for training, wherein the four basic classifiers are as follows: logistic regression classifiers, random forest classifiers, SVM classifiers, and naive bayes classifiers.
Adopting AdaBoost integrated learning algorithm, inputting each characteristic x= [ x ] defined as the previous step selection 1 ，x 2 ，…x k ]K=1, 2, …, k, where k is the number of features, k=6, and the output is defined as a multiple classification problem, i.e. s t = {0,1,2,3,4,5,6,7}; the n basis classifiers are defined as h t (x) T=1, 2, …, n, x is an example feature. The learned combined classifier is defined as a linear combination of n base classifiersBy continually updating the iterations, when the base classifier h t (x) Based on distribution D t After generation, the classifier weights alpha t So that alpha is t h t (x) Is a loss function of (2)Minimum.
Each iteration utilizes the next round of basic classifier h t (x) Correction of H t-1 (x) Error, i.e. minimizing After model training is completed, data of a test set is input, and for each case, the model automatically outputs a column vector s t ＝{s 1 ，s 2 ，…，s n N=7. Label y corresponding to the degree of illness t The highest value indicates the recommended treatment plan, thereby completing the goal of automatic accurate medication.
Example 2:
example 2 is a modification of example 1.
Referring to fig. 1, the invention provides a machine learning-based accurate medication method for inflammatory bowel disease, which comprises the following steps:
step (1): CT images are obtained through a CTE sequence thin-layer scanning mode, energy spectrum CT is switched instantaneously by high 140kV and low 80kV energy, and attenuation coefficients of voxels in an energy range of 40-140 keV are determined according to sampling data obtained under the two energies, so that a single-energy image is obtained. Then, CT data with uniform resolution and approximately same gray scale distribution are obtained by preprocessing methods such as data desensitization and resampling, and the data size is 256×256. And the disease condition-medication standard is formulated, and different medicaments are adopted for treatment according to different severity of IBD disease.
Step (2): segmentation of lesion areas of CT images of inflammatory bowel disease: segmenting a region of interest (region of interest, ROI, size 256×256) ROI by using the CT image obtained in the step (1) through a CNN network IBD The method comprises the steps of carrying out a first treatment on the surface of the The network consists of 3 convolutional layers of 3 x 3 steps of size 2, a ReLU activation function and a max-pulling layer of step 2.
Step (3): region of interest ROI for inflammatory bowel disease IBD The method comprises the steps of carrying out a first treatment on the surface of the And (3) extracting characteristics: using WORC toolkit method for the ROI obtained in step (2) IBD The method comprises the steps of carrying out a first treatment on the surface of the And extracting image characteristics and quantifying information contained in the image. Building a correlation migration model by adopting SVM and hinge loss function, and migrating 176 features to a migration model of T (t=60) features for target task T classificationFinally, t (t=60) features most relevant to the clinical medication efficacy are obtained through iterative optimization.
Step (4): the selected 60 most relevant features of the curative effect of clinical medication are integrated, a plurality of individual classifiers are trained, and the individual classifiers are subjected to additive model training by using an AdaBoost integrated learning method, so that an inflammatory bowel disease noninvasive medication curative effect evaluation model is constructed, and the automatic and accurate medication task is completed.
Referring to fig. 1, the pretreatment method in step (1) specifically includes:
data desensitization and washing; and carrying out data deformation on sensitive information in the original energy spectrum CT data acquired by the hospital according to a desensitization rule. Resampling data; the present invention resamples the entire dataset to a fixed isomorphic resolution, resampling all samples to a 256 x 256 size.
The disease condition-medication is standard, and different medicines are used for treatment according to the severity of IBD disease. IBD is classified as ulcerative colitis (Ulcerative Colitis, UC) and Crohn's Disease (CD), where ulcerative colitis and Crohn's Disease are classified as mild, moderate and severe depending on the severity of the condition.
Referring to fig. 1, the method for dividing a lesion area in step (2) specifically includes:
the invention adopts CNN segmentation network to segment the CT image obtained in the step (1) to obtain the region of interest (region of interest, ROI, size is 256 multiplied by 4) ROI IBD The method comprises the steps of carrying out a first treatment on the surface of the The invention randomly divides CT data with segmentation Mask into a training set (0.8), a verification set (0.1) and a test set (0.1), sends the data into a CNN segmentation network, and completes ROI segmentation through network training and network testing.
Referring to fig. 1, the feature extraction method described in step (3) includes:
1. ROI using WORC toolkit method IBD And extracting image characteristics and quantifying information contained in the image. To avoid some variables in the feature becoming dominant due to their larger units of measurement, the data features were normalized to obtain 176 image features.
2. Building a correlation migration model by adopting SVM and hinge loss function, and migrating 176 features to a migration model of T (t=60) features for target task T classificationFor all (s, y t ) E D, IBD classification associated migration model +.>Is-> Wherein s is an input feature, phi(s) is a feature selector, F t And finally, obtaining t (t=60) features most relevant to clinical medication efficacy through iterative optimization as an SVM classifier.
Referring to fig. 1, the non-invasive medication efficacy assessment method described in step (4) includes:
1. the selected 60 features most relevant to the clinical medication efficacy are input into four individual classifiers for training, and the four individual classifiers adopt a logistic regression classifier, a random forest classifier, an SVM classifier and a naive Bayes classifier.
2. The AdaBoost integrated learning algorithm is used for inputting various features x= [ x ] which are defined as the features selected in the previous step 1 ，x 2 ，…x k ]K=1, 2, …, k, where k is the number of features; the output is defined as a two-class problem (well defined as 0 for a certain drug treatment and 1 for a certain drug treatment), i.e. y= (0, 1); the n basis classifiers are defined as h t (x) T=1, 2, …, n, x is an example feature. The learned combined classifier is defined as a linear combination of n base classifiersBy continually updating the iterations, when the base classifier h t (x) After generating based on the distribution Dt, the classifier weights α t So that alpha is t h t (x) Is> Minimum.
3. Using the obtained next round of basic classifier h t (x) Correction of H t-1 (x) Error, i.e. minimizing The trained ensemble learning classifier can complete the evaluation of the therapeutic effect of the IBD disease drug treatment.
The embodiment of the invention provides a machine learning-based accurate medicine application method for inflammatory bowel disease, which applies a convolutional neural network to the disease classification of IBD, thereby realizing noninvasive therapeutic artificial intelligent evaluation based on energy spectrum CT images and overcoming various limitations caused by complex scoring adopted clinically; the IBD accurate administration method provided by the invention has higher sensitivity and accuracy, can help guide the selection of the treatment scheme of the IBD patient, evaluate prognosis, further shorten the treatment time of the IBD patient, and has better clinical practicability.
Those skilled in the art will appreciate that the invention provides a system and its individual devices, modules, units, etc. that can be implemented entirely by logic programming of method steps, in addition to being implemented as pure computer readable program code, in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Therefore, the system and various devices, modules and units thereof provided by the invention can be regarded as a hardware component, and the devices, modules and units for realizing various functions included in the system can also be regarded as structures in the hardware component; means, modules, and units for implementing the various functions may also be considered as either software modules for implementing the methods or structures within hardware components.
The foregoing describes specific embodiments of the present invention. It is to be understood that the invention is not limited to the particular embodiments described above, and that various changes or modifications may be made by those skilled in the art within the scope of the appended claims without affecting the spirit of the invention. The embodiments of the present application and features in the embodiments may be combined with each other arbitrarily without conflict.

Claims (3)

1. The accurate inflammatory bowel disease administration method based on machine learning is characterized by comprising the following steps of:
step S1: acquiring energy spectrum CT data of inflammatory bowel disease, preprocessing the energy spectrum CT data, and manufacturing a disease activity-medication label data set based on the processed data;
step S2: the focus region segmentation is carried out on the preprocessed energy spectrum CT data by using an end-to-end convolutional neural network to obtainTo the focus area of inflammatory bowel disease；
Step S3: will be in the focus of inflammatory bowel diseaseExtracting image histology characteristics to obtain characteristics，n=1,2，…，176；
Building a migration model, the target task will be performed by 176 featuresClassification, migration to by->Target task for individual features->The classified migration model obtains 60 features most relevant to the clinical medication efficacy through iterative optimization;
step S4: integrating the selected 60 most relevant features, training a plurality of classifiers through an image histology method, and training the classifiers through an AdaBoost integrated learning method to obtain a noninvasive medication evaluation model of inflammatory bowel disease;
the step S1 specifically includes the following steps:
step S1.1: through a CTE sequence thin layer scanning mode, energy spectrum CT adopts high 140kV and low 80kV energy instantaneous switching, and according to sampling data obtained under the two energies, the attenuation coefficient of a voxel in the energy range of 40-140 keV is determined to obtain 101 single energy diagrams, and the scanning layer thickness and the layer spacing are 5mm;
step S1.2: the energy spectrum CT adopts an adaptive statistical iterative reconstruction algorithm (daptive statistical iterative reconstruction, ASIR) and a filtered back projection algorithm (filtered back projection, FBP) to reconstruct a single energy image by combining ASiR with the level of 40 percent, namely an image is formed by mixing ASiR and FBP with the weight of 4:6, and the reconstruction layer thickness and layer spacing are 1.25mm;
step S1.3: data desensitization is carried out on the obtained reconstructed image, and an irregular image is removed;
step S1.4: preprocessing training data, adopting a de-averaging operation, and normalizing the amplitude of the data within the same range;
step S1.5: formulating a disease condition-medication specification, and adopting different medicaments for treatment according to different severity of IBD disease condition;
in the step S2, focus region segmentation is performed on the obtained energy spectrum CT image, which specifically includes the following steps:
step S2.1: the CT image is subjected to downsampling and feature value extraction, the convolution kernel size is 3 multiplied by 3, the step length is 2, and max-pooling operation is performed after each group of convolution operation, so that the image is further reduced to 1/2 of the original image;
the method comprises the steps of carrying out a total of 4 times in the downsampling process to obtain characteristics, and finally obtaining 256 downsampling characteristic graphs of 16 multiplied by 16, wherein the number of convolution kernels used by each convolution layer is 32, 64, 128 and 256 respectively;
step S2.2: 4 groups of deconvolution operations are used in the up-sampling process to expand the picture to 2 times of the original picture, and the characteristic diagram of the corresponding layer is cut and copied to form an up-convolution result;
after the up-sampling process is finished, a graph with the size of 256 multiplied by 256 is obtained, finally, a convolution kernel with the size of 1 multiplied by 1 is used for reducing the number of channels to 2, different labels are used for marking, and CT image segmentation is completed, so that the CT image is obtainedWherein the number of convolution kernels used for each convolution layer is 256, 128, 64, 32, respectively;
step S2.3: evaluating the segmentation result by adopting a loss function in the model training process;
the step S3 includes:
step S3.1: using WORC toolkit method for the step S2Extracting image characteristics from the energy spectrum CT of the image, and quantifying the information contained in the image;
step S3.2: constructing a migration model by adopting SVM and a range loss function;
the step S3.2 includes:
step S3.2.1: defining D as a set of 176 image features and disease degree labels, and the source task s as a disease classification prediction task and labeling the disease degree= { CD remission, CD mild, CD moderate, CD severe, UC remission, UC mild, UC moderate, UC severe } is defined as {0,1,2,3,4,5,6,7}, target task t = s is defined as a multi-classification task;
definition of target tasks by 176 featuresTClassifying, and migrating to target task by t featuresClassified migration model->The model finds a feature set most relevant to the clinical medication curative effect, so that s is migrated to the multi-classification task t to optimize the overall performance;
step S3.2.2: for all ofIBD typing Classification-associated migration model->Is optimized toThe method comprises the steps of carrying out a first treatment on the surface of the Wherein s is the input feature, < >>For feature selector, ++>As an SVM classifier, a loss function adopts a range loss function; finally obtain +.>Features most relevant to clinical medication efficacy;
the step S4 includes:
step S4.1: classifying IBD disease conditions through image features, and inputting the 60 features which are selected in the step S3 and most relevant to clinical medication curative effects into a plurality of basic classifiers for training;
step S4.2: adopting AdaBoost integrated learning algorithm to input each feature defined as the previous stepWherein->=60 is the feature number;
the output is defined as a multi-classification problem, i.eThe method comprises the steps of carrying out a first treatment on the surface of the The n base classifiers are defined as，/>Is an example feature;
the learned combined classifier is defined as a linear combination of 4 basic classifiersBy continuously updating the iterations, when the base classifier +.>Based on distribution->After generation, the weight of the classifier +.>Make->Is>Minimum;
step S4.3: each iteration utilizes the next round of basic classifierCorrect->Error, i.e. minimizing；
Step S4.4: after model training is completed, data of a test set is input, and for each case, the model automatically outputs column vectors，n=7；
Label corresponding to illness stateThe highest value indicates the recommended treatment plan, thereby completing the goal of automatic accurate medication.
2. The machine learning based accurate dosing method for inflammatory bowel disease of claim 1, wherein the loss function in step S2.3 is constructed as follows:
wherein the method comprises the steps ofThe method comprises the steps of carrying out a first treatment on the surface of the X and Y are model segmentation results and doctor labeling results for comparing similarity between the model segmentation results and the doctor labeling results, and the range is 0,1]The method comprises the steps of carrying out a first treatment on the surface of the M represents the number of categories, ">Is a one-hot vector, the element has only 0 or 1 two values, if the category is the same as the category of the sample, 1 is taken, otherwise, 0 is taken; />Representing the probability that the predicted sample belongs to c.
3. Accurate medicine system of inflammatory bowel disease based on machine learning, characterized by comprising:
model M1: acquiring energy spectrum CT data of inflammatory bowel disease, preprocessing the energy spectrum CT data, and manufacturing a disease activity-medication label data set based on the processed data;
model M2: the focus region segmentation is carried out on the pretreated energy spectrum CT data by using an end-to-end convolutional neural network to obtain the focus region of inflammatory bowel disease；
Model M3: will be in the focus of inflammatory bowel diseaseExtracting image histology characteristics to obtain characteristics，n=1,2，…，176；
Building a migration model, the target task will be performed by 176 featuresClassification, migration to by->Target task for individual features->The classified migration model obtains 60 features most relevant to the clinical medication efficacy through iterative optimization;
model M4: integrating the selected 60 most relevant features, training a plurality of classifiers through an image histology method, and training the classifiers through an AdaBoost integrated learning method to obtain a noninvasive medication evaluation model of inflammatory bowel disease;
the model M1 is specifically as follows:
model M1.1: through a CTE sequence thin layer scanning mode, energy spectrum CT adopts high 140kV and low 80kV energy instantaneous switching, and according to sampling data obtained under the two energies, the attenuation coefficient of a voxel in the energy range of 40-140 keV is determined to obtain 101 single energy diagrams, and the scanning layer thickness and the layer spacing are 5mm;
model M1.2: the energy spectrum CT adopts an adaptive statistical iterative reconstruction algorithm (daptive statistical iterative reconstruction, ASIR) and a filtered back projection algorithm (filtered back projection, FBP) to reconstruct a single energy image by combining ASiR with the level of 40 percent, namely an image is formed by mixing ASiR and FBP with the weight of 4:6, and the reconstruction layer thickness and layer spacing are 1.25mm;
model M1.3: data desensitization is carried out on the obtained reconstructed image, and an irregular image is removed;
model M1.4: preprocessing training data, adopting a de-averaging operation, and normalizing the amplitude of the data within the same range;
model M1.5: formulating a disease condition-medication specification, and adopting different medicaments for treatment according to different severity of IBD disease condition;
the focus region segmentation is carried out on the obtained energy spectrum CT image in the model M2, and the specific steps are as follows:
model M2.1: the CT image is subjected to downsampling and feature value extraction, the convolution kernel size is 3 multiplied by 3, the step length is 2, and max-pooling operation is performed after each group of convolution operation, so that the image is further reduced to 1/2 of the original image;
the method comprises the steps of carrying out a total of 4 times in the downsampling process to obtain characteristics, and finally obtaining 256 downsampling characteristic graphs of 16 multiplied by 16, wherein the number of convolution kernels used by each convolution layer is 32, 64, 128 and 256 respectively;
model M2.2: 4 groups of deconvolution operations are used in the up-sampling process to expand the picture to 2 times of the original picture, and the characteristic diagram of the corresponding layer is cut and copied to form an up-convolution result;
after the up-sampling process is finished, a graph with the size of 256 multiplied by 256 is obtained, finally, a convolution kernel with the size of 1 multiplied by 1 is used for reducing the number of channels to 2, different labels are used for marking, and CT image segmentation is completed, so that the CT image is obtainedWherein the number of convolution kernels used for each convolution layer is 256, 128, 64, 32, respectively;
model M2.3: evaluating the segmentation result by adopting a loss function in the model training process;
the model M3 includes:
model M3.1: model M2 is obtained using WORC toolkit methodExtracting image characteristics from the energy spectrum CT of the image, and quantifying the information contained in the image;
model M3.2: constructing a migration model by adopting SVM and a range loss function;
the model M3.2 comprises:
model M3.2.1: defining D as a set of 176 image features and disease degree labels, and the source task s as a disease classification prediction task and labeling the disease degree= { CD remission, CD mild, CD moderate, CD severe, UC remission, UC mild, UC moderate, UC severe } is defined as {0,1,2,3,4,5,6,7}, target task t = s is defined as a multi-classification task;
definition of target tasks by 176 featuresTClassifying, and migrating to target task by t featuresClassified migration model->The model finds a feature set most relevant to the clinical medication curative effect, so that s is migrated to the multi-classification task t to optimize the overall performance;
model M3.2.2: for all ofIBD typing Classification-associated migration model->Is optimized toThe method comprises the steps of carrying out a first treatment on the surface of the Wherein s is the input feature, < >>For feature selector, ++>As an SVM classifier, a loss function adopts a range loss function; finally obtain +.>Features most relevant to clinical medication efficacy;
the model M4 includes:
model M4.1: IBD disease classification is carried out through image features, and 60 features which are selected by the model M3 and most relevant to clinical medication curative effect are input into a plurality of basic classifiers for training;
model M4.2: adopting AdaBoost integrated learning algorithm to input each feature defined as the previous stepWherein->=60 is the feature number;
the output is defined as a multi-classification problem, i.eThe method comprises the steps of carrying out a first treatment on the surface of the The n base classifiers are defined as，/>Is an example feature;
the learned combined classifier is defined as a linear combination of 4 basic classifiersBy continuously updating the iterations, when the base classifier +.>Based on distribution->After generation, the weight of the classifier +.>Make->Is>Minimum;
model M4.3: each iteration utilizes the next round of basic classifierCorrect->Error, i.e. minimizing；
Model M4.4: after model training is completed, data of a test set is input, and for each case, the model automatically outputs column vectors，n=7；
Label corresponding to illness stateThe highest value indicates the recommended treatment plan, thereby completing the goal of automatic accurate medication.
CN202110287375.0A 2021-03-17 2021-03-17 Accurate medicine application method and system for inflammatory bowel disease based on machine learning Active CN112967778B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110287375.0A CN112967778B (en) 2021-03-17 2021-03-17 Accurate medicine application method and system for inflammatory bowel disease based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110287375.0A CN112967778B (en) 2021-03-17 2021-03-17 Accurate medicine application method and system for inflammatory bowel disease based on machine learning

Publications (2)

Publication Number Publication Date
CN112967778A CN112967778A (en) 2021-06-15
CN112967778B true CN112967778B (en) 2024-02-02

Family

ID=76279038

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110287375.0A Active CN112967778B (en) 2021-03-17 2021-03-17 Accurate medicine application method and system for inflammatory bowel disease based on machine learning

Country Status (1)

Country Link
CN (1) CN112967778B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113362932B (en) * 2021-06-24 2023-02-17 河北大学 A radiomics feature screening method based on CC-attention mechanism
CN113658700B (en) * 2021-08-16 2023-10-27 复旦大学附属中山医院 A non-invasive assessment method and system for portal hypertension based on machine learning
CN115691791A (en) * 2022-11-17 2023-02-03 吾征智能技术(北京)有限公司 Identification system, method, electronic equipment and storage medium of TCM insomnia types
CN117934480B (en) * 2024-03-25 2024-05-28 天津市第五中心医院 Lung CT image processing method and system for lung injury prediction

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109166105A (en) * 2018-08-01 2019-01-08 中国人民解放军南京军区南京总医院 The malignancy of tumor risk stratification assistant diagnosis system of artificial intelligence medical image
CN109461495A (en) * 2018-11-01 2019-03-12 腾讯科技（深圳）有限公司 A kind of recognition methods of medical image, model training method and server
CN110755104A (en) * 2019-11-05 2020-02-07 四川大学华西医院 Non-invasive assessment method of portal hypertension in liver cirrhosis based on 3D multi-channel convolutional neural network
CN111653356A (en) * 2020-04-20 2020-09-11 浙江大学 A new coronary pneumonia screening method and new coronary pneumonia screening system based on deep learning

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2569541B (en) * 2017-12-19 2020-08-19 Mirada Medical Ltd Method and apparatus for medical imaging
US20200342958A1 (en) * 2019-04-23 2020-10-29 Cedars-Sinai Medical Center Methods and systems for assessing inflammatory disease with deep learning

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109166105A (en) * 2018-08-01 2019-01-08 中国人民解放军南京军区南京总医院 The malignancy of tumor risk stratification assistant diagnosis system of artificial intelligence medical image
CN109461495A (en) * 2018-11-01 2019-03-12 腾讯科技（深圳）有限公司 A kind of recognition methods of medical image, model training method and server
CN110755104A (en) * 2019-11-05 2020-02-07 四川大学华西医院 Non-invasive assessment method of portal hypertension in liver cirrhosis based on 3D multi-channel convolutional neural network
CN111653356A (en) * 2020-04-20 2020-09-11 浙江大学 A new coronary pneumonia screening method and new coronary pneumonia screening system based on deep learning

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Enhanced Approach for Classification of Ulcerative Colitis Severity in Colonoscopy Videos Using CNN;Tejaswini, SVLL等;Advances in Visual Computing;25-37 *
人工智能在腹部影像的研究现状与展望;黄子星;宋彬;;放射学实践(第11期);110-114 *
深度学习技术在消化系统疾病诊疗中的应用;韩伟;周金池;魏延;张哲;赵曙光;;胃肠病学和肝病学杂志(第09期);116-120 *
肝脏肿瘤CT图像深度学习分割方法综述;马金林;邓媛媛;马自萍;;中国图象图形学报(第10期);78-100 *

Also Published As

Publication number Publication date
CN112967778A (en) 2021-06-15

Similar Documents

Publication Publication Date Title
CN112967778B (en) Accurate medicine application method and system for inflammatory bowel disease based on machine learning
US12266442B2 (en) Decision support system for medical therapy planning
Vijila Rani et al. Lung lesion classification scheme using optimization techniques and hybrid (KNN-SVM) classifier
VanBerlo et al. Accurate assessment of the lung sliding artefact on lung ultrasonography using a deep learning approach
Rodríguez et al. Computer aided detection and diagnosis in medical imaging: a review of clinical and educational applications
Ho et al. Feature-level ensemble approach for COVID-19 detection using chest X-ray images
Tummala et al. Liver tumor segmentation from computed tomography images using multiscale residual dilated encoder‐decoder network
Jerbi et al. Automatic classification of ultrasound thyroids images using vision transformers and generative adversarial networks
Rani et al. Radon transform-based improved single seeded region growing segmentation for lung cancer detection using AMPWSVM classification approach
Qin et al. Application of artificial intelligence in diagnosis of craniopharyngioma
Haq An overview of deep learning in medical imaging
Davamani et al. Biomedical image segmentation by deep learning methods
Hasan et al. Lung segmentation from chest X-ray images using deeplabv3plus-based cnn model
Vishmitha et al. Kidney stone detection using deep learning and transfer learning
Singh et al. Performance evaluation of breast lesion detection systems with expert delineations: a comparative investigation on mammographic images
Priya et al. An intellectual caries segmentation and classification using modified optimization-assisted transformer denseUnet++ and ViT-based multiscale residual denseNet with GRU
Ghosh et al. Ensemble based neural network for the classification of mura dataset
WO2023200990A1 (en) Automated methods for determining fibroglandular density on mammograms
Alghamdi et al. Lung cancer detection from LDCT images using deep convolutional neural networks
Hsu et al. A comprehensive study of age-related macular degeneration detection
Mhedbi et al. A Convolutional Neural Network based system for classifying malignant and benign skin lesions using mobile-device images
Jain Liver Disease Detection from CT scan images using Deep Learning and Transfer Learning
Anggraini et al. Optimizing hyperparameters in multiview convolutional neural network for improved breast cancer detection in mammograms
Sarkar et al. Precision Tumor Identification in Lung CT Scans: Leveraging Convolutional Neural Networks for Accurate Segmentation
Nancy Noella et al. Efficient computer-aided diagnosis of Alzheimer’s disease and Parkinson’s disease—a survey

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant