Automatic classification of IgA endomysial antibody test for celiac disease: a new method deploying machine learning

Area of Science:

• Immunology
• Medical Diagnostics
• Artificial Intelligence

Background:

• The endomysial autoantibody (EmA) test is crucial for celiac disease diagnosis but is limited by subjective interpretation and the need for expert evaluation.
• Observer variability in EmA testing can impact diagnostic accuracy and consistency.

Purpose of the Study:

• To investigate the feasibility of applying machine learning for an observer-independent, automated assessment and classification of IgA-class EmA tests.
• To develop a novel computational method for celiac disease diagnostics using EmA imaging.

Main Methods:

• A dataset of 2597 high-quality IgA-class EmA images was analyzed using machine learning algorithms.
• A classification model was trained on images pre-classified by expert evaluators into four categories: positive, negative, IgA deficient, and equivocal.

Main Results:

• The machine learning model demonstrated high performance with a sensitivity of 82.84% and specificity of 99.40%.
• Overall accuracy reached 96.80% with a classification error of 3.20%.
• The model achieved excellent area under the curve values across all classes, indicating robust classification capabilities. Mean assessment time per image was 16.11 seconds.

Conclusions:

• Machine learning provides a rapid, precise, and objective method for analyzing IgA-class EmA tests.
• This AI-driven approach has the potential to significantly simplify EmA test interpretation and improve celiac disease diagnostics.