Machine learning-based multiclass model for autoimmune disease diagnosis and classification through nailfold...

Area of Science:

• Rheumatology
• Medical Imaging
• Computational Biology

Background:

• Autoimmune diseases like rheumatoid arthritis (RA) and systemic lupus erythematosus (SLE) involve microvascular damage.
• Nailfold videocapillaroscopy (NVC) visualizes capillary changes, offering potential diagnostic insights.
• Distinguishing between these conditions and healthy controls can be challenging using traditional methods.

Purpose of the Study:

• To develop and validate a predictive model for differentiating controls, RA, and SLE using NVC image features.
• To identify key NVC features that characterize each group.
• To assess the model's performance in classifying these distinct populations.

Main Methods:

• Collected 600 NVC images from 396 participants (controls, RA, SLE) and split into training/test sets (7:3 ratio).
• Extracted nine NVC features and employed an eXtreme Gradient Boosting multiclassification model.
• Utilized SHapley Additive exPlanations (SHAP) for feature importance and model interpretation.

Main Results:

• Seven NVC features significantly differed among groups; the model achieved AUCs of 0.96 (training) and 0.80 (test).
• Key features included papilla shape, red blood cell aggregation, capillary loop number, crossed capillary loops, and subpapillary venous plexus (SVP).
• Distinct NVC patterns emerged: papilla shape for controls, crossed capillary loops for RA, and SVP for SLE.

Conclusions:

• A robust multiclassification model was developed for differentiating autoimmune diseases using NVC features.
• The findings deepen the understanding of microvascular alterations in RA and SLE.
• This model presents a potential tool for clinical diagnosis and monitoring of autoimmune conditions.