Feasibility Study on the Use of Infrared Cameras for Skin Cancer Detection under a Proposed Data Degradation Model
<p>Example of a plastic marker used to select the region of interest. The region of interest is indicated in red, and the suspicious lesion is highlighted in blue.</p> "> Figure 2
<p>Detection scheme used to evaluate the feasibility of using different IR cameras in detecting skin cancer using active thermography.</p> "> Figure 3
<p>Schematic of the proposed degradation model, which addresses 3 areas: temporal, spatial, and thermal resolution. Giving rise to a process composed of 5 stages.</p> "> Figure 4
<p>Samples of the measurement behaviors from different IR cameras on a blackbody stabilized at 40 °C in a room with controlled ambient temperature at 20 °C using an AC unit. The order of measurements taken with the different cameras is as follows: (<b>a</b>) Xenics Gobi-640, (<b>b</b>) Opgal Therm-App and (<b>c</b>) Seek Thermal CompactPRO.</p> "> Figure 5
<p>Sample of the results of the jump correction produced by the NUC in the Xenics Gobi-640 camera. (<b>a</b>) Uncorrected measurements; (<b>b</b>) corrected measurements.</p> "> Figure 6
<p>Sample of the results of the jump correction produced by the NUC in the Seek Thermal CompactPRO camera. (<b>a</b>) Uncorrected measurements; (<b>b</b>) corrected measurements.</p> "> Figure 7
<p>Sample of the degradation performed on a high-quality video to mimic Opgal Therm-App camera features. (<b>a</b>) Image captured with the Xenics Gobi-640 camera, (<b>b</b>) image captured at the same instant of time and same area with the Opgal Therm-App camera, (<b>c</b>) Xenics image adapted to Opgal camera features, (<b>d</b>,<b>e</b>) correspond to representative TRCs of the video acquired with the Xenics and Opgal cameras, respectively. (<b>f</b>) TRC from the adaptation.</p> "> Figure 8
<p>Sample of the degradation performed on a high-quality video to mimic Seek Thermal CompactPRO camera features. (<b>a</b>) Image captured with the Xenics Gobi-640 camera, (<b>b</b>) image captured at the same instant of time and same area with the Seek Thermal CompactPRO camera, (<b>c</b>) Xenics image adapted to Seek camera features, (<b>d</b>,<b>e</b>) correspond to representative TRCs of the video acquired with the Xenics and Seek cameras, respectively. (<b>f</b>) TRC from the adaptation.</p> "> Figure A1
<p>Sample results of the degradation of an IR video to the characteristics of different simulated cameras in the case of a benign lesion. The top row shows an image from the video and the bottom row shows characteristic TRCs of the mole and non-mole areas. (<b>a</b>,<b>e</b>) Correspond to high-quality data captured with the QmagiQ camera; (<b>b</b>,<b>f</b>) data adapted to Xenics Gobi-640; (<b>c</b>,<b>g</b>) data adapted to Opgal Therm-App; (<b>d</b>,<b>h</b>) data adapted to Seek Thermal CompactPRO.</p> "> Figure A2
<p>Sample results of the degradation of an IR video to the characteristics of different simulated cameras in a case of a malignant lesion, diagnosed as MM. The top row shows an image from the video and the bottom row shows characteristic TRCs of the mole and non-mole areas. (<b>a</b>,<b>e</b>) Correspond to high-quality data captured with the QmagiQ camera; (<b>b</b>,<b>f</b>) data adapted to Xenics Gobi-640; (<b>c</b>,<b>g</b>) data adapted to Opgal Therm-App; (<b>d</b>,<b>h</b>) data adapted to Seek Thermal CompactPRO.</p> "> Figure A3
<p>Sample results of the degradation of an IR video to the characteristics of different simulated cameras in the case of a malignant lesion, diagnosed as BCC. The top row shows an image from the video and the bottom row shows characteristic TRCs of the mole and non-mole areas. (<b>a</b>,<b>e</b>) Correspond to high-quality data captured with the QmagiQ camera; (<b>b</b>,<b>f</b>) data adapted to Xenics Gobi-640; (<b>c</b>,<b>g</b>) data adapted to Opgal Therm-App; (<b>d</b>,<b>h</b>) data adapted to Seek Thermal CompactPRO.</p> "> Figure A4
<p>Sample results of the degradation of an IR video to the characteristics of different simulated cameras in the case of a malignant lesion, diagnosed as SCC. The top row shows an image from the video and the bottom row shows characteristic TRCs of the mole and non-mole areas. (<b>a</b>,<b>e</b>) Correspond to high-quality data captured with the QmagiQ camera; (<b>b</b>,<b>f</b>) data adapted to Xenics Gobi-640; (<b>c</b>,<b>g</b>) data adapted to Opgal Therm-App; (<b>d</b>,<b>h</b>) data adapted to Seek Thermal CompactPRO.</p> ">
Abstract
:1. Introduction
2. Theoretical Background
2.1. Skin Cancer
2.2. Active Thermography
2.3. Microbolometer Technology
3. Materials and Methods
3.1. Infrared Imagers
3.2. Dataset
3.2.1. Data Acquisition Process
3.2.2. Image Registration Algorithm
- Manually select the corners of the plastic marker in the visible image and the first IR image. For subsequent images in the IR sequence, the corners are automatically detected using the selections from the previous image as references.
- Estimate an affine transformation matrix that maps motion between consecutive images (one matrix is estimated for each pair of images).
- Apply the inverse transformation to each image to align the image sequence relative to the first IR image.
3.3. Skin Cancer Screening
- Lesion selection. A mask is created over the visible image to delineate the lesion area, generated manually by outlining the pigmented region. Consequently, two sets of TRCs are established: one designated as L, comprising TRCs within the lesion area delineated by the mask, and the other denoted as N, consisting of TRCs within the non-lesion area.
- Initial temperature estimation. The subsequent step involves the selection of TRCs, which depends on the initial temperature of each TRC. In order to select curves whose initial temperature is less affected by non-uniformities in the cooling process, each TRC is modeled using a double exponential function, defined as follows:
- TRCs selection. A reference temperature is calculated as , where E is the expectation operator and is the vector function of the initial temperature of the TRCs of the modeled lesion area within the L set.The selection of points to process considers a margin of error of with respect to . In this way, the set of points to be used is defined as follows:
- Representative TRCs. For each set and , a representative TRC is computed as the average TRCs among the selected pixels within each set, generating the average curves and .
- Feature extraction. From the representative TRCs, a combination of features is extracted (features vector). In this case, the feature vector is as follows:, which are detailed in Section 3.3.1.
- Classifier. As the final step, the feature vector is processed by a classifier, which defines if the lesion is suspicious or not. In this study, we evaluated the performance of several machine learning techniques, including K-nearest neighbors (KNN), SVM with RBF kernel, random forest, and eXtreme Gradient Boosting (XGBoost). The random forest classifier achieved the best results, as detailed in Section 4.2. The results for the other classification techniques are provided in Appendix B.
3.3.1. Feature Extraction Techniques
- Euclidean distance (d). This feature is calculated as the norm of the difference , normalized by the amount of points, i.e., as shown in the following equation:The concept behind this feature is as follows: A small Euclidean distance indicates similar curves, making it highly likely that the lesion is benign. Conversely, a large Euclidean distance indicates significant differences in thermal recovery between the curves. This suggests that the lesion is likely malignant, as its thermal recovery behavior deviates from that of normal tissue [3].
- Energy difference (). Let , i.e., the unbiased TRC of the area. The energy difference is calculated as follows:This feature is closely related to d. However, because of the triangular inequality , quantifies smaller differences than d.
- Statistical similitude features. Here, six base features are defined to measure the similarity of a set of TRCs to a normalized modeled TRC, using the dual exponential model shown in (1). It is assumed here that the modeled TRC has K sample points and, thus, . With this, its inner product is computed by , where , and T denotes the column-vector transpose. is obtained by computing the five model parameters using a non-linear least-squares fitting approach. These parameters are averaged to obtain the descriptive TRC for each class (namely, cancerous TRCs and non-cancerous TRCs). Then, is forced to have a unit norm, i.e., .Let and denote a model TRC and an arbitrary TRC, respectively, both modeled and normalized (here, it is assumed that ). The following characteristics of projection, correlation, and Euclidean distance are described below.
- (a)
- Projection. The projection of onto is calculated as . This operation is performed for a set of at least 10 TRCs, and then the mean projection, , and the standard deviation of the projection, , are calculated over this set of projections.
- (b)
- Correlation. The correlation of onto is calculated using the following expression:
- (c)
- Distance. This parameter is calculated according to (3) but using as a reference and a family of at least 10 TRCs, . The mean value and standard deviation are calculated over this set of distance values.Considering a model of malignant and benign TRCs, and , respectively, for each data cube, the projection characteristics and , correlations and , and distances and are calculated for the sets of TRCs and . In this way, 24 features are extracted, and grouped into four categories based on the origin of the model curve and the analysis group. For example, the malignant model over non-lesion area allows us to extract , , , , , and features.
3.4. Proposed Data Adaptation Process
- Temporal resolution. This edge only contemplates the temporary downsampling stage. As the temporal resolution of the camera to adapt is lower than the camera that captures the data, the sequence of IR images is downsampled following the proportion , where and denote the sample rate of the high-quality camera and the camera to adapt, respectively.
- Spatial resolution. This aspect addresses two areas: spatial resolution and optical distortion.
- (a)
- Spatial downsampling. The QmagiQ camera has a high instantaneous field of view (IFOV); however, the size of its FPA is smaller than the FPA of the cameras to model, in terms of the number of pixels. To avoid introducing artificial elements with unknown effects, the spatial dimensions of the images were not modified.
- (b)
- Point spread function (PSF). Images captured by each camera are usually affected by blurring due to optical distortions of the lens of each camera. The point spread function (PSF) applied corresponds to a 2D model, which was calculated as described by Jara et al. [48]. This model considers both optical and electronic aberrations.In order to transfer the optical response of the camera to be simulated to the high-quality videos, the PSF obtained from the lower-quality camera is applied to each frame of the higher-quality camera. Thus, the modified frame is the result of the spatial convolution of the image with the PSF of the camera to simulate :It is important to mention that usually the PSF is used to correct the optical distortions. In this study, it is used to degrade the images; such a worst-case scenario is to be evaluated. So it is expected that using the real camera can improve the performance of the detection algorithm by correcting the optical distortion.
- Thermal resolution. In order to simulate the thermal sensitivity of a camera, adding the characteristic noise of the camera to be modeled is proposed.The IR images contain spatial and temporal noise. According to Feng et al. [49], the noise affecting images captured by microbolometer IR cameras contains low-, medium-, and high-frequency spatial noise, along with horizontal and vertical component noise and non-uniformities.For applications utilizing active thermography, temporal noise is equally or more important than spatial noise affecting the camera.
- (a)
- Spatial noise. The spatial noise characteristics of the camera to be simulated are added according to the following expression: , where is the spatial noise resulting from blackbody radiator measurements as described in Section 3.5.1. This spatial noise already contains the low-, medium-, and high-frequency components proposed by Feng et al. [49] since we extracted them from actual measurements.
- (b)
- Temporal noise. The temporal noise is composed of two components: low frequency () and high frequency (). Where the low-frequency noise is associated with ambient temperature fluctuations, while the high-frequency noise is related to the chemical and electrical characteristics of each camera component.Therefore, the response of each detector over time is degraded according to the following expression:To reiterate, the subscripts mean that we are evaluating the noise component at the -th pixel, at the k-th frame.
3.5. Noise Characterization of Imagers
3.5.1. Spatial Noise Characterization
3.5.2. Temporal Noise Characterization
4. Results
4.1. Validation of the Degradation Model
4.2. Feasibility Study of Using Low-Cost IR Cameras in Skin Cancer Detection
5. Discussion and Conclusions
- Adjustment of the image size. The model does not adjust the size of high-quality images to match those captured with lower-quality cameras when the FPA of the higher-quality camera is larger. This approach was not considered because it requires an interpolation process, which may introduce noise that is not characteristic of the camera being simulated.
- Shape and size of the detector. This feature is critical for determining the minimum size of detectable objects. However, since the size of skin lesions is significantly larger than the detector size, this characteristic was not considered. The size of a typical detector in microbolometer technology is approximately 20 μm; with the right optics, it would be possible to detect a 40 × 40 μm lesion.
- Temporal noise introduced by the shutter. When analyzing the temporal variations in the cameras due to changes in ambient temperature, the camera adjusts the offset and modifies the gain. However, incorporating this feature is problematic because the camera’s logic for applying these adjustments is unknown and cannot be determined.
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
Abbreviations
BCC | basal cell carcinoma |
SCC | squamous cell carcinoma |
AK | actinic keratosis |
MM | malignant melanoma |
fps | frames per second |
UV | ultraviolet |
FOV | field of view |
FPA | focal plane array |
LWIR | longwave infrared |
IR | infrared |
IRT | infrared thermography |
QWIP | quantum well-infrared photodetector |
NUC | non-uniformity correction |
ROI | region of interest |
SVM | support vector machine |
RBF | radial basis function |
KNN | K-nearest neighbors |
XGBoost | eXtreme gradient boosting |
TRC | thermoregulation curve |
DWELL | quantum dots in a well |
NETD | noise equivalent temperature difference |
PSF | point spread function |
TPR | true positive rate |
TNR | true negative rate |
PPV | positive predictive value |
Appendix A. IR Video Degradation Sample
Appendix B. Performance of the Detection Algorithm Using Different Machine Learning Techniques
Camera Adaptation | QmagiQ (Original) | Xenics | Opgal | Seek | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Index | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD |
Accuracy (%) | 48.28 | 100.00 | 78.39 ± 8.13 | 44.83 | 100.00 | 76.59 ± 8.15 | 41.38 | 96.55 | 75.37 ± 8.40 | 41.38 | 96.55 | 70.85 ± 8.83 |
TPR (%) | 11.11 | 100.00 | 70.87 ± 15.39 | 9.09 | 100.00 | 66.57 ± 15.92 | 5.88 | 100.00 | 65.80 ± 16.19 | 7.69 | 100.00 | 58.97 ± 16.33 |
TPR MM (%) | 0.00 | 100.00 | 53.41 ± 42.83 | 0.00 | 100.00 | 45.45 ± 42.45 | 0.00 | 100.00 | 52.73 ± 42.87 | 0.00 | 100.00 | 58.18 ± 42.16 |
TPR BCC (%) | 0.00 | 100.00 | 71.44 ± 17.98 | 0.00 | 100.00 | 66.77 ± 19.28 | 0.00 | 100.00 | 65.87 ± 19.11 | 0.00 | 100.00 | 59.03 ± 19.56 |
TPR SCC (%) | 0.00 | 100.00 | 82.25 ± 31.04 | 0.00 | 100.00 | 82.26 ± 31.26 | 0.00 | 100.00 | 75.85 ± 35.20 | 0.00 | 100.00 | 60.65 ± 39.37 |
TNR (%) | 38.89 | 100.00 | 83.44 ± 9.97 | 41.18 | 100.00 | 83.25 ± 10.18 | 40.00 | 100.00 | 81.72 ± 10.48 | 26.67 | 100.00 | 78.75 ± 11.41 |
PPV (%) | 11.11 | 100.00 | 74.26 ± 14.24 | 16.67 | 100.00 | 73.00 ± 15.00 | 14.29 | 100.00 | 70.83 ± 14.79 | 14.29 | 100.00 | 65.21 ± 15.92 |
Camera Adaptation | QmagiQ (Original) | Xenics | Opgal | Seek | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Index | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD |
Accuracy (%) | 51.72 | 100.00 | 85.03 ± 7.31 | 51.72 | 100.00 | 84.12 ± 7.28 | 58.62 | 100.00 | 84.55 ± 7.33 | 48.28 | 100.00 | 79.73 ± 7.99 |
TPR (%) | 11.11 | 100.00 | 82.09 ± 15.60 | 12.50 | 100.00 | 80.33 ± 15.45 | 22.22 | 100.00 | 80.25 ± 13.86 | 15.38 | 100.00 | 73.15 ± 15.12 |
TPR MM (%) | 0.00 | 100.00 | 73.57 ± 38.00 | 0.00 | 100.00 | 69.63 ± 39.45 | 0.00 | 100.00 | 72.96 ± 38.23 | 0.00 | 100.00 | 68.69 ± 39.78 |
TPR BCC (%) | 0.00 | 100.00 | 81.77 ± 17.07 | 0.00 | 100.00 | 81.26 ± 17.11 | 0.00 | 100.00 | 81.23 ± 15.71 | 11.11 | 100.00 | 74.02 ± 17.31 |
TPR SCC (%) | 0.00 | 100.00 | 90.72 ± 25.04 | 0.00 | 100.00 | 85.83 ± 28.51 | 0.00 | 100.00 | 82.03 ± 31.62 | 0.00 | 100.00 | 73.18 ± 36.24 |
TNR (%) | 43.75 | 100.00 | 87.12 ± 9.47 | 44.44 | 100.00 | 86.65 ± 9.43 | 52.94 | 100.00 | 87.42 ± 8.77 | 29.41 | 100.00 | 84.20 ± 10.04 |
PPV (%) | 14.29 | 100.00 | 81.47 ± 12.70 | 25.00 | 100.00 | 80.53 ± 12.89 | 20.00 | 100.00 | 81.15 ± 12.48 | 25.00 | 100.00 | 75.86 ± 14.15 |
Camera Adaptation | QmagiQ (Original) | Xenics | Opgal | Seek | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Index | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD | Min. | Max. | AVG±SD |
Accuracy (%) | 55.17 | 100.00 | 85.69 ± 6.76 | 51.72 | 100.00 | 82.66 ± 7.41 | 55.17 | 100.00 | 82.64 ± 7.36 | 48.28 | 100.00 | 80.25 ± 7.75 |
TPR (%) | 25.00 | 100.00 | 85.18 ± 12.31 | 28.57 | 100.00 | 81.26 ± 13.17 | 21.43 | 100.00 | 80.19 ± 13.59 | 14.29 | 100.00 | 79.19 ± 14.02 |
TPR MM (%) | 0.00 | 100.00 | 76.21 ± 36.24 | 0.00 | 100.00 | 64.17 ± 40.81 | 0.00 | 100.00 | 67.43 ± 39.53 | 0.00 | 100.00 | 71.17 ± 38.85 |
TPR BCC (%) | 16.67 | 100.00 | 84.39 ± 14.50 | 14.29 | 100.00 | 81.38 ± 15.42 | 0.00 | 100.00 | 80.01 ± 16.06 | 11.11 | 100.00 | 78.19 ± 16.61 |
TPR SCC (%) | 0.00 | 100.00 | 96.36 ± 15.48 | 0.00 | 100.00 | 94.48 ± 18.60 | 0.00 | 100.00 | 91.22 ± 23.38 | 0.00 | 100.00 | 88.92 ± 25.86 |
TNR (%) | 35.00 | 100.00 | 86.05 ± 9.18 | 50.00 | 100.00 | 83.67 ± 9.95 | 47.37 | 100.00 | 84.34 ± 9.91 | 38.46 | 100.00 | 80.93 ± 10.92 |
PPV (%) | 22.22 | 100.00 | 80.45 ± 12.24 | 25.00 | 100.00 | 77.05 ± 13.21 | 20.00 | 100.00 | 77.62 ± 13.11 | 22.22 | 100.00 | 73.85 ± 13.40 |
References
- Bagavathiappan, S.; Lahiri, B.; Saravanan, T.; Philip, J.; Jayakumar, T. Infrared thermography for condition monitoring—A review. Infrared Phys. Technol. 2013, 60, 35–55. [Google Scholar] [CrossRef]
- Lahiri, B.; Bagavathiappan, S.; Jayakumar, T.; Philip, J. Medical applications of infrared thermography: A review. Infrared Phys. Technol. 2012, 55, 221–235. [Google Scholar] [CrossRef] [PubMed]
- Godoy, S.E.; Ramirez, D.A.; Myers, S.A.; von Winckel, G.; Krishna, S.; Berwick, M.; Padilla, R.S.; Sen, P.; Krishna, S. Dynamic infrared imaging for skin cancer screening. Infrared Phys. Technol. 2015, 70, 147–152. [Google Scholar] [CrossRef]
- Godoy, S.E.; Hayat, M.M.; Ramirez, D.A.; Myers, S.A.; Padilla, R.S.; Krishna, S. Detection theory for accurate and non-invasive skin cancer diagnosis using dynamic thermal imaging. Biomed. Opt. Express 2017, 8, 2301–2323. [Google Scholar] [CrossRef] [PubMed]
- Kaczmarek, M.; Nowakowski, A. Active dynamic thermography in medical diagnostics. In Application of Infrared to Biomedical Sciences; Springer: Singapore, 2017; pp. 291–310. [Google Scholar] [CrossRef]
- Pauk, J.; Ihnatouski, M.; Wasilewska, A. Detection of inflammation from finger temperature profile in rheumatoid arthritis. Med. Biol. Eng. Comput. 2019, 57, 2629–2639. [Google Scholar] [CrossRef] [PubMed]
- Frize, M.; Adéa, C.; Payeur, P.; Gina Di Primio, M.D.; Karsh, J.; Ogungbemile, A. Detection of rheumatoid arthritis using infrared imaging. In Proceedings of the Medical Imaging 2011: Image Processing, Lake Buena Vista, FL, USA, 14–16 February 2011; Dawant, B.M., Haynor, D.R., Eds.; International Society for Optics and Photonics, SPIE: Bellingham, WA, USA, 2011; Volume 7962, p. 79620M. [Google Scholar] [CrossRef]
- Nur, R.; Frize, M. Image processing of infrared thermal images for the detection of necrotizing enterocolitis. In Proceedings of the Medical Imaging 2013: Image Processing, Lake Buena Vista, FL, USA, 10–13 February 2013; International Society for Optics and Photonics: Bellingham, WA, USA, 2013; Volume 8669, p. 86692M. [Google Scholar] [CrossRef]
- Alwashmi, M.F. The Use of Digital Health in the Detection and Management of COVID-19. Int. J. Environ. Res. Public Health 2020, 17, 2906. [Google Scholar] [CrossRef] [PubMed]
- Taylor, W.; Abbasi, Q.H.; Dashtipour, K.; Ansari, S.; Shah, S.A.; Khalid, A.; Imran, M.A. A Review of the State of the Art in Non-Contact Sensing for COVID-19. Sensors 2020, 20, 5665. [Google Scholar] [CrossRef] [PubMed]
- Verstockt, J.; Verspeek, S.; Thiessen, F.; Tjalma, W.A.; Brochez, L.; Steenackers, G. Skin Cancer Detection Using Infrared Thermography: Measurement Setup, Procedure and Equipment. Sensors 2022, 22, 3327. [Google Scholar] [CrossRef] [PubMed]
- Buzug, T.M.; Schumann, S.; Pfaffmann, L.; Reinhold, U.; Ruhlmann, J. Functional infrared imaging for skin-cancer screening. In Proceedings of the 2006 International Conference of the IEEE Engineering in Medicine and Biology Society, New York, NY, USA, 30 August–3 September 2006; pp. 2766–2769. [Google Scholar] [CrossRef]
- Çetingül, M.P.; Herman, C. Quantification of the thermal signature of a melanoma lesion. Int. J. Therm. Sci. 2011, 50, 421–431. [Google Scholar] [CrossRef]
- Di Carlo, A.; Elia, F.; Desiderio, F.; Catricalà, C.; Solivetti, F.M.; Laino, L. Can video thermography improve differential diagnosis and therapy between basal cell carcinoma and actinic keratosis? Dermatol. Ther. 2014, 27, 290–297. [Google Scholar] [CrossRef]
- Magalhaes, C.; Vardasca, R.; Rebelo, M.; Valenca-Filipe, R.; Ribeiro, M.; Mendes, J. Distinguishing melanocytic nevi from melanomas using static and dynamic infrared thermal imaging. J. Eur. Acad. Dermatol. Venereol. 2019, 33, 1700–1705. [Google Scholar] [CrossRef] [PubMed]
- Magalhaes, C.; Tavares, J.M.R.; Mendes, J.; Vardasca, R. Comparison of machine learning strategies for infrared thermography of skin cancer. Biomed. Signal Process. Control 2021, 69, 102872. [Google Scholar] [CrossRef]
- Soto, R.F.; Godoy, S.E. A novel feature extraction approach for skin cancer screening using active thermography. In Proceedings of the 2023 IEEE Latin American Conference on Computational Intelligence (LA-CCI), Recife, Brazil, 29 October–1 November 2023; pp. 1–6. [Google Scholar] [CrossRef]
- Bu, C.; Xu, H.; Mao, Z.; Zhang, D.; Pu, C. Non-destructive testing theoretical study on skin tumor detection using long-pulsed infrared thermal wave testing technology. Therm. Sci. 2019, 23, 1401–1408. [Google Scholar] [CrossRef]
- Barros, T.C.; Figueiredo, A.A.A. Three-dimensional numerical evaluation of skin surface thermal contrast by application of hypothermia at different depths and sizes of the breast tumor. Comput. Methods Programs Biomed. 2023, 236, 107562. [Google Scholar] [CrossRef]
- Narayanamurthy, V.; Padmapriya, P.; Noorasafrin, A.; Pooja, B.; Hema, K.; Nithyakalyani, K.; Samsuri, F. Skin cancer detection using non-invasive techniques. RSC Adv. 2018, 8, 28095–28130. [Google Scholar] [CrossRef]
- Dhar, N.K.; Dat, R.; Sood, A.K. Advances in infrared detector array technology. Optoelectron.-Adv. Mater. Devices 2013, 7, 149–188. [Google Scholar] [CrossRef]
- Martyniuk, P.; Antoszewski, J.; Martyniuk, M.; Faraone, L.; Rogalski, A. New concepts in infrared photodetector designs. Appl. Phys. Rev. 2014, 1, 041102. [Google Scholar] [CrossRef]
- Khodayar, F.; Sojasi, S.; Maldague, X. Infrared thermography and NDT: 2050 horizon. Quant. InfraRed Thermogr. J. 2016, 13, 210–231. [Google Scholar] [CrossRef]
- Fraiwan, L.; AlKhodari, M.; Ninan, J.; Mustafa, B.; Saleh, A.; Ghazal, M. Diabetic foot ulcer mobile detection system using smart phone thermal camera: A feasibility study. Biomed. Eng. Online 2017, 16, 117. [Google Scholar] [CrossRef] [PubMed]
- Fraiwan, L.; Ninan, J.; Al-Khodari, M. Mobile application for ulcer detection. Open Biomed. Eng. J. 2018, 12, 16. [Google Scholar] [CrossRef]
- Van Doremalen, R.; Van Netten, J.; Van Baal, J.; Vollenbroek-Hutten, M.; Van der Heijden, F. Validation of low-cost smartphone-based thermal camera for diabetic foot assessment. Diabetes Res. Clin. Pract. 2019, 149, 132–139. [Google Scholar] [CrossRef]
- Niri, R.; Lucas, Y.; Treuillet, S.; Douzi, H. Smartphone-Based Thermal Imaging System for Diabetic Foot Ulcer Assessment. 2019. Available online: https://hal.science/hal-02161044 (accessed on 25 May 2024).
- Villa, E.; Arteaga-Marrero, N.; Ruiz-Alzola, J. Performance Assessment of Low-Cost Thermal Cameras for Medical Applications. Sensors 2020, 20, 1321. [Google Scholar] [CrossRef]
- Kirimtat, A.; Krejcar, O.; Selamat, A.; Herrera-Viedma, E. FLIR vs SEEK thermal cameras in biomedicine: Comparative diagnosis through infrared thermography. BMC Bioinform. 2020, 21, 88. [Google Scholar] [CrossRef] [PubMed]
- Arteaga-Marrero, N.; Bodson, L.C.; Hernández, A.; Villa, E.; Ruiz-Alzola, J. Morphological Foot Model for Temperature Pattern Analysis Proposed for Diabetic Foot Disorders. Appl. Sci. 2021, 11, 7369. [Google Scholar] [CrossRef]
- Gulshan; Arora, A.S. Automated prediction of diabetes mellitus using infrared thermal foot images: Recurrent neural network approach. Biomed. Phys. Eng. Express 2024, 10, 025025. [Google Scholar] [CrossRef]
- Xue, E.Y.; Chandler, L.K.; Viviano, S.L.; Keith, J.D. Use of FLIR ONE smartphone thermography in burn wound assessment. Ann. Plast. Surg. 2018, 80, S236–S238. [Google Scholar] [CrossRef]
- Li, F.; Wang, M.; Wang, T.; Wang, X.; Ma, X.; He, H.; Ma, G.; Zhao, D.; Yue, Q.; Wang, P.; et al. Smartphone-based infrared thermography to assess progress in thoracic surgical incision healing: A preliminary study. Int. Wound J. 2023, 20, 2000–2009. [Google Scholar] [CrossRef]
- Simões, M.F.; Sousa, J.S.; Pais, A.C. Skin cancer and new treatment perspectives: A review. Cancer Lett. 2015, 357, 8–42. [Google Scholar] [CrossRef]
- Arens, E.A.; Zhang, H. The Skin’s Role in Human Thermoregulation and Comfort. 2006. Available online: https://escholarship.org/uc/item/3f4599hx (accessed on 25 May 2024).
- Nishida, N.; Yano, H.; Nishida, T.; Kamura, T.; Kojiro, M. Angiogenesis in cancer. Vasc. Health Risk Manag. 2006, 2, 213. [Google Scholar] [CrossRef]
- Bonmarin, M.; Le Gal, F.A. Thermal Imaging in Dermatology. In Imaging in Dermatology; Academic Press: Boston, MA, USA, 2016; pp. 437–454. [Google Scholar] [CrossRef]
- Gurjarpadhye, A.A.; Parekh, M.B.; Dubnika, A.; Rajadas, J.; Inayathullah, M. Infrared imaging tools for diagnostic applications in dermatology. SM J. Clin. Med Imaging 2015, 1, 1. [Google Scholar]
- AlZubaidi, A.; Ethawi, Y.; Schmölzer, G.; Sherif, S.; Narvey, M.; Seshia, M. Review of Biomedical Applications of Contactless Imaging of Neonates Using Infrared Thermography and Beyond. Methods Protoc. 2018, 1, 39. [Google Scholar] [CrossRef] [PubMed]
- Diakides, M.; Bronzino, J.D.; Peterson, D.R. Medical Infrared Imaging: Principles and Practices; CRC Press: Boca Raton, FL, USA, 2013. [Google Scholar]
- Bonmarin, M.; Le Gal, F.A. Lock-in thermal imaging for the early-stage detection of cutaneous melanoma: A feasibility study. Comput. Biol. Med. 2014, 47, 36–43. [Google Scholar] [CrossRef] [PubMed]
- QmagicQ-LLC. Datasheet QmagicQ Falcon, 2024. Available online: http://www.qmagiq.com/falcon256.html (accessed on 25 May 2024).
- Xenics. Datasheet Xenics Gobi-640, 2024. Available online: https://www.exosens.com/products/gobi (accessed on 25 May 2024).
- Nax-Instruments. Datasheet Opgal Therm-App, 2024. Available online: https://www.naxsg.com/product/therm-app-th/#tab-downloads (accessed on 25 May 2024).
- Seek-Thermal-Inc. Datasheet Seek Thermal CompactPRO, 2024. Available online: https://www.thermal.com/uploads/1/0/1/3/101388544/compactpro-sellsheet-usav1.pdf (accessed on 25 May 2024).
- Godoy, S.E. Communication-Theoretic Approach for Skin Cancer Detection Using Dynamic Thermal Imaging; The University of New Mexico: Albuquerque, NM, USA, 2015. [Google Scholar]
- Díaz, S.; Krohmer, T.; Moreira, Á.; Godoy, S.E.; Figueroa, M. An Instrument for Accurate and Non-Invasive Screening of Skin Cancer Based on Multimodal Imaging. IEEE Access 2019, 7, 176646–176657. [Google Scholar] [CrossRef]
- Jara, A.; Torres, S.N.; Machuca, G.; Coelho, P.; Viafora, L.A. Three-dimensional point spread function estimation method for mid-wave infrared microscope imaging. Appl. Opt. 2022, 61, 8467–8474. [Google Scholar] [CrossRef]
- Feng, T.; Jin, W.; Si, J. Spatial-noise subdivision evaluation model of uncooled infrared detector. Infrared Phys. Technol. 2021, 119, 103954. [Google Scholar] [CrossRef]
- Schober, P.; Boer, C.; Schwarte, L.A. Correlation coefficients: Appropriate use and interpretation. Anesth. Analg. 2018, 126, 1763–1768. [Google Scholar] [CrossRef]
- Yélamos, O.; Braun, R.P.; Liopyris, K.; Wolner, Z.J.; Kerl, K.; Gerami, P.; Marghoob, A.A. Usefulness of dermoscopy to improve the clinical and histopathologic diagnosis of skin cancers. J. Am. Acad. Dermatol. 2019, 80, 365–377. [Google Scholar] [CrossRef]
Camera | QmagiQ | Xenics Gobi-640 | Opgal Therm-App | Seek Thermal Compact PRO | |
---|---|---|---|---|---|
Characteristic | |||||
Focal distance [mm] | 50 | 25 | 6.8 | 12.5 | |
Spectral range [μm] | 8–14 | 8–14 | 7.5–14 | 7.5–14 | |
FOV [°] | 5.5 × 4.4 | 25.84 × 19.38 | 55.56 × 41.67 | 33.40 × 24.81 | |
Temporal resolution [fps] | 60 | 50 | 8.7 | 15 | |
FPA size [pixels] | 320 × 256 | 640 × 480 | 384 × 288 | 320 × 240 | |
NETD [mK] | 20 | 50 | 70 | 70 | |
Manufacturer | QmagiQ, LLC (Nashua, NH, USA) | Xenics nv (Leuven, Belgium) | Opgal Optronic Industries Ltd. (Karmiel, Israel) | Seek Thermal Inc. (Santa Barbara, CA, USA) |
Camera Adaptation | QmagiQ (Original) | Xenics | Opgal | Seek | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Index | Min. | Max. | AVG ± SD | Min. | Max. | AVG ± SD | Min. | Max. | AVG ± SD | Min. | Max. | AVG ± SD |
Accuracy (%) | 62.07 | 100.00 | 87.29 ± 6.60 | 51.72 | 100.00 | 84.33 ± 7.11 | 55.17 | 100.00 | 84.20 ± 7.09 | 51.72 | 100.00 | 82.13 ± 7.43 |
TPR (%) | 33.33 | 100.00 | 87.26 ± 11.49 | 28.57 | 100.00 | 83.03 ± 12.62 | 27.27 | 100.00 | 83.23 ± 12.58 | 28.57 | 100.00 | 79.77 ± 13.79 |
TPR MM (%) | 0.00 | 100.00 | 76.42 ± 36.29 | 0.00 | 100.00 | 66.91 ± 40.11 | 0.00 | 100.00 | 71.68 ± 38.14 | 0.00 | 100.00 | 74.98 ± 37.30 |
TPR BCC (%) | 20.00 | 100.00 | 87.23 ± 13.60 | 20.00 | 100.00 | 83.18 ± 14.96 | 0.00 | 100.00 | 83.60 ± 14.63 | 16.67 | 100.00 | 78.68 ± 16.40 |
TPR SCC (%) | 0.00 | 100.00 | 95.94 ± 15.81 | 0.00 | 100.00 | 94.74 ± 18.37 | 0.00 | 100.00 | 90.92 ± 24.09 | 0.00 | 100.00 | 87.66 ± 26.62 |
TNR (%) | 47.62 | 100.00 | 87.39 ± 8.98 | 47.62 | 100.00 | 85.28 ± 9.50 | 44.44 | 100.00 | 84.94 ± 9.75 | 36.84 | 100.00 | 83.74 ± 10.23 |
PPV (%) | 28.57 | 100.00 | 82.45 ± 11.94 | 28.57 | 100.00 | 79.11 ± 12.90 | 25.00 | 100.00 | 78.91 ± 12.74 | 26.67 | 100.00 | 76.96 ± 13.25 |
Methodology | Detection Problem | TPR (%) | TNR (%) |
---|---|---|---|
Naked eye evaluation [51] | MM vs. benign | 71.00 | 81.00 |
Dermoscopy evaluation [51] | MM vs. benign | 90.00 | 90.00 |
Passive thermography + deep learning [16] | MM vs. benign | 94.12 | 98.41 |
Passive thermography + deep learning [16] | Malignant vs. benign | 62.81 | 57.58 |
Active thermography − QmagiQ camera | Malignant vs. benign | 87.26 | 87.39 |
Active thermography − Xenics camera | Malignant vs. benign | 83.03 | 85.28 |
Active thermography − Opgal camera | Malignant vs. benign | 83.23 | 84.94 |
Active thermography − Seek camera | Malignant vs. benign | 79.77 | 83.74 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Soto, R.F.; Godoy, S.E. Feasibility Study on the Use of Infrared Cameras for Skin Cancer Detection under a Proposed Data Degradation Model. Sensors 2024, 24, 5152. https://doi.org/10.3390/s24165152
Soto RF, Godoy SE. Feasibility Study on the Use of Infrared Cameras for Skin Cancer Detection under a Proposed Data Degradation Model. Sensors. 2024; 24(16):5152. https://doi.org/10.3390/s24165152
Chicago/Turabian StyleSoto, Ricardo F., and Sebastián E. Godoy. 2024. "Feasibility Study on the Use of Infrared Cameras for Skin Cancer Detection under a Proposed Data Degradation Model" Sensors 24, no. 16: 5152. https://doi.org/10.3390/s24165152