Deep Learning Classification of Systemic Sclerosis Skin Using the MobileNetV2 Model

Area of Science:

• Biomedical engineering
• Artificial intelligence in medicine
• Dermatology

Background:

• Systemic sclerosis (SSc) is a rare autoimmune disease characterized by fibrosis, necessitating early diagnosis for effective management.
• Machine learning (ML) and deep learning (DL) show promise in medical applications but face challenges in resource-limited environments due to data and computational demands.
• Existing ML algorithms require significant training data and powerful hardware (GPUs), hindering their clinical adoption.

Purpose of the Study:

• To develop a novel, efficient mobile deep learning network for characterizing Systemic sclerosis (SSc) skin.
• To create an accurate and accessible diagnostic tool for SSc screening in clinical settings.
• To improve upon traditional Convolutional Neural Network (CNN) performance in SSc skin image analysis.

Main Methods:

• Proposed a novel mobile deep learning network integrating UNet, dense connectivity CNN, and a mobile training module.
• Utilized the MobileNetV2 architecture for enhanced computational efficiency and diagnostic accuracy in training.
• Implemented and fine-tuned the network on a standard laptop, comparing its performance against a traditional CNN.

Main Results:

• The proposed network achieved high accuracy (95.2% on test set) with rapid training (<5 hours) on a standard laptop.
• MobileNetV2 demonstrated superior performance compared to a standard CNN, achieving 94.8% accuracy vs. 82.9% on the test set for SSc classification.
• The model effectively classified normal, early, and late-stage SSc skin images with high precision.

Conclusions:

• The developed mobile deep learning network shows significant promise for accurate and efficient SSc skin characterization.
• This approach offers a potential solution for implementing advanced diagnostic tools in resource-constrained clinical environments.
• The proposed network could serve as a simple, inexpensive, and accurate screening tool for early SSc detection.