New machine learning tool for SSC 95% precise in skin screening

Using a laptop, a computer machine learning tool was able to distinguish, with 95% accuracy, normal skin images from images collected from people with systemic sclerosis (SSc), the study showed.

This tool can potentially quickly diagnose SSc and accurately classify the range of conditions, the scientists said.

Moreover, according to researchers, it could be “easily applied in a clinical setting, providing a simple, inexpensive and accurate verification tool.”

Study, “Classification of deep learning for systemic sclerosis of the skin using the MobileNetV2 model, “It was published in a journal Engineering in Medicine and Biology.

SSc, also called systemic scleroderma, is an autoimmune disease characterized by widespread accumulation of scar tissue in the skin and internal organs. Because there are a number of SSc symptoms, early diagnosis and determining the extent of disease progression can present significant challenges for clinicians and can result in delayed treatment and disease management.

Therefore, objective and consistent methods are needed to diagnose SSc.

Computer-based machine learning techniques are already widely used to process and analyze large amounts of data generated for various purposes. Machine learning has shown potential in diagnosing autoimmune diseases, predicting outcomes, and classifying disorders such as rheumatoid arthritis, inflammatory bowel disease, and lupus.

Now, scientists from the Department of Biomedical Engineering at the University of Houston have used a type of machine learning called deep neural networks to analyze and learn from skin images collected from SSC patients and healthy individuals to predict the presence and extent of SSC.

“Our preliminary study, which aims to show the effectiveness of the proposed network architecture, promises to characterize the SSc,” said Dr. Metin Akay, founder of the Department of Biomedical Engineering and lead author of the study, at the university. Media Release.

“By scanning images, the network learns from existing images and decides which new image is normal or is in the early or late stages of the disease,” Akay said.

Deep learning organizes computer programs known as algorithms into layers that can learn and make decisions. The team used a deep learning network called the convolutional neural network (CNN), which is commonly used in engineering, medicine, and biology.

Combining a modified CNN architecture (UNet) with additional layers of algorithms, the team developed a new mobile training module called MobileNetV2.

It is significant that the proposed network was used on a standard laptop.

The training phase of the network included 834 images from 20 people without scleroderma and 678 images from 20 patients with early and late phase SSc, for a total of 1,512 images. In the validation phase, 104 normal and 84 SSc images were used.

“By scanning images, the network learns from existing images and decides which new image is normal or is in the early or late stages of the disease,” Akay said.

It took less than five hours to complete, the training phase diagnosed all 1,512 sets of images without misdiagnosis, giving 100% accuracy and 100% sensitivity. The sensitivity of the test is its ability to correctly identify the disease.

In the validation phase, the networks correctly diagnosed 81 of the 84 SSc image sets and misdiagnosed three, yielding a sensitivity of 96.4%. It should be noted that 101 out of 104 normal images were correctly classified, with an overall accuracy of 96.8% and a specificity of 97.1%. It should be noted that the specificity is the ability of the test to correctly identify people without disease.

Final test phase – analyzing 104 normal and 84 SSc images – correctly diagnosed 80 of 84 SSc set images with a sensitivity of 95.2%. It also correctly diagnosed 99 of 104 normal images, giving an overall accuracy of 95.2% and a specificity of 95.1%.

By comparison, the team ran the unmodified CNN architecture on the same laptop. The process took more than 14 hours to train and confirm 65 of 84 SSc images with a sensitivity of 77.3%. Likewise, the test phase correctly diagnosed 60 of the 84 SSc images, giving a sensitivity of 71.4%.

“These results showed that the MobileNetV2 architecture is more accurate and efficient than CNN for classifying normal and SSc skin images,” the team wrote.

To differentiate between early and late stage SSc, an additional 1,887 images were analyzed – 1,041 normal, 423 early stage, and 423 late stage SSc. The screening phase correctly diagnosed 178 of the 183 normal, early, and late phase SSc images, giving an overall accuracy of 97.2%.

Two of the 40 images at the early stage were misclassified as late stages, while two more were misclassified as common. In addition, one in 103 normal images is misclassified as early stage SSc. All 40 late stages of SSc were correctly diagnosed.

Accordingly, the study phase correctly diagnosed 182 of 192 normal, early and late phase SSc images, giving an overall accuracy of 94.8%. Again, the unmodified CNN architecture was less precise in identifying different SSc degrees.

“Our ultimate goal is to use this approach as a fast and reliable method for assessing the severity of SSc using images,” the researchers wrote. “The proposed network architecture could determine a high-accuracy diagnosis within minutes. This saves significant time and money compared to current diagnosis. ”

“We believe that the proposed network architecture could be easily applied in a clinical setting,” Akay concluded, noting that it is “a simple, inexpensive, and accurate tool for SSC verification.”

Steve holds a PhD in Biochemistry from the University of Toronto School of Medicine, Canada. For 18 years he worked as a medical scientist, both in industry and in academia, where his research focused on discovering new drugs to treat inflammatory disorders and infectious diseases. Steve recently moved away from the lab and engaged in scientific communications, where he helps make information about medical sciences more accessible to everyone.

Total Posts: 134

Patrícia holds a PhD in Medical Microbiology and Infectious Diseases from Leiden University Medical Center in Leiden, the Netherlands. She studied applied biology at the Universidade do Minho and was a postdoctoral researcher at the Institute of Molecule Medicine in Lisbon, Portugal. Her work focused on the molecular genetic properties of infectious agents such as viruses and parasites.