Artificial Intelligence Algorithms in Nailfold Capillaroscopy Image Analysis: A Systematic Review

Background Non-invasive imaging modalities offer a great deal of clinically significant information that aid in the diagnosis of various medical conditions. Coupled with the never-before-seen capabilities of Artificial Intelligence (AI), uncharted territories that offer novel innovative diagnostics are reached. This systematic review compiled all studies that utilized AI in Nailfold Capillaroscopy as a future diagnostic tool. Methods and Findings Five databases for medical publications were searched using the keywords artificial intelligence, machine learning, deep learning and nailfold capillaroscopy to return 105 studies. After applying the eligibility criteria, 10 studies were selected for the final analysis. Data was extracted into tables that addressed population characteristics, AI model development and nature and results of their respective performance. We found supervised deep learning approaches to be the most commonly used (n = 8). Systemic Sclerosis was the most commonly studied disease (n = 6). Sample size ranged from 17,126 images obtained from 289 participants to 50 images from 50 participants. Ground truth was determined either by experts labelling (n = 6) or known clinical status (n = 4). Significant variation was noticed in model training, testing and feature extraction, and therefore the reporting of model performance. Recall, precision and Area Under the Curve were the most used metrics to report model performance. Execution times ranged from 0.064 to 120 seconds per image. Only two models offered future predictions besides the diagnostic output. Conclusions AI has demonstrated a truly remarkable potential in the interpretation of Nailfold Capillaroscopy by providing physicians with an intelligent decision-supportive tool for improved diagnostics and prediction. With more validation studies, this potential can be translated to daily clinical practice.