سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / International Conference on Artificial Intelligence; City, Industry and Health

Data-Driven Finger Selection for Nailfold Capillaroscopy in SLE Using Unsupervised Learning and Diagnostic Scoring

نویسندگان :

Habibollah Jafari¹ Abdolamir Karbalaie²

1- Department of Computer Engineering, Na.C., Islamic Azad University, Najafabad, Iran 2- Department of Community Medicine and Rehabilitation, Umeå University, Sweden SE-901 87, Umeå, Sweden

کلمات کلیدی :

Capillaroscopy،Unsupervised Learning،Systemic Lupus Erythematosus،K-Means،Diagnostic Scoring

چکیده :

Objective: Identify latent microvascular phenotypes in SLE patients via unsupervised machine learning and determine the most diagnostically representative fingers for capillaroscopy. Methods: Capillaroscopic features from eight fingers per SLE patient were analyzed using PCA and K-Means clustering. A composite scoring system ranked fingers by abnormality severity, redundancy, variability, missingness, and diagnostic weight. Results: Unsupervised clustering revealed two distinct groups: fingers with pronounced microvascular changes versus minor abnormalities. Three fingers consistently ranked highest in diagnostic utility: the left middle, right middle, and right ring fingers. These exhibited frequent, prominent abnormalities with minimal redundancy, while other digits offered overlapping or less critical data. Conclusion: Data-driven analysis identifies the left middle, right middle, and right ring fingers as optimal for SLE capillaroscopy. Focusing on these three digits may improve detection of microvascular pathology while enhancing clinical efficiency, streamlining assessments without compromising diagnostic yield.