Machine Learning for MRI Classification of Systemic Lupus Erythematous Patients with and without Neuropsychiatric...

Area of Science:

• Neuroimaging and Machine Learning
• Computational Medicine
• Neurology

Background:

• Systemic Lupus Erythematosus (SLE) is a complex autoimmune disease.
• Neuropsychiatric SLE (NPSLE) presents significant diagnostic challenges.
• Accurate diagnosis of NPSLE is crucial for timely and effective treatment.

Purpose of the Study:

• To develop a practical Machine Learning (ML) framework for diagnosing NPSLE and SLE.
• To utilize Magnetic Resonance Imaging (MRI) derived brain features for classification.
• To evaluate the performance of different ML models in distinguishing between NPSLE, SLE, and healthy controls.

Main Methods:

• A cross-sectional study included 27 SLE patients (14 NPSLE, 13 SLE) and 20 healthy controls.
• Brain structural features, specifically regional cortical thickness, were quantitatively assessed using the VolBrain online platform.
• Four ML models (Logistic Regression, SVM, Random Forest, XGBoost) were trained and tested using fivefold cross-validation.

Main Results:

• The Random Forest (RF) model achieved the highest performance with 90% accuracy on the test set.
• RF effectively classified between NPSLE, SLE, and control groups without regularization.
• Key features for classification included Precentral gyrus right thickness norm, angular gyrus thickness asymmetry, and Parietal thickness asymmetry.

Conclusions:

• The Random Forest algorithm shows significant potential as a clinical diagnostic support tool for NPSLE.
• Machine learning applied to MRI-derived features offers a promising avenue for improving NPSLE diagnosis.
• This framework can aid clinicians in differentiating NPSLE from SLE and healthy controls.