Deep Learning Performance in Analyzing Nailfold Videocapillaroscopy Images in Systemic Sclerosis.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Medicine
• Rheumatology Diagnostics

Background:

• Nailfold videocapillaroscopy (NVC) is crucial for diagnosing systemic sclerosis (SSc).
• Manual classification of NVC images can be subjective and time-consuming.
• Deep learning offers a potential solution for automated NVC image analysis.

Purpose of the Study:

• To classify NVC images from systemic sclerosis (SSc) patients and healthy controls using deep learning.
• To compare the performance of six different deep learning models for NVC image classification.
• To evaluate the diagnostic accuracy of deep learning against expert rheumatologists.

Main Methods:

• A dataset of 977 NVC images from 50 SSc patients and 30 healthy individuals was curated.
• Images were classified into normal, early, active, and late-stage SSc patterns by rheumatologists.
• Six deep learning models (MobileNetV3Large, ResNet152V2, Xception, VGG-19, InceptionV3, NASNetLarge) were trained and evaluated using accuracy, precision, recall, and F1 score.

Main Results:

• Deep learning models achieved high accuracy (90.6%–98.9%), precision (93.4%–98.9%), recall (90.6%–98.8%), and F1 scores (92%–98.9%).
• InceptionV3 model exhibited superior performance with 98.95% accuracy, 98.94% precision, 98.80% recall, and 98.88% F1 score.
• All models demonstrated excellent ROC AUC values (98.99%–100%), indicating robust diagnostic capability.

Conclusions:

• Deep learning models can effectively classify NVC images for systemic sclerosis diagnosis.
• The performance of AI models approaches that of experienced rheumatologists.
• This technology holds promise for improving the efficiency and accuracy of SSc diagnosis.