Explainable machine learning algorithm for classifying resting-state functional MRI in amyotrophic lateral sclerosis

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease impacting multiple brain systems.
• Resting-state functional magnetic resonance imaging (rs-fMRI) reveals altered brain function in ALS.
• Machine learning (ML) can analyze complex rs-fMRI patterns but often lacks transparency.

Purpose of the Study:

• To develop an explainable ML pipeline for classifying ALS patients and healthy controls (HCs) using rs-fMRI data.
• To enhance the transparency of ML models in neurological disease classification.

Main Methods:

• rs-fMRI data from 30 ALS patients and 30 HCs were preprocessed using independent component analysis and dual regression.
• A 3D convolutional neural network (3D-CNN) was trained for ALS/HC classification.
• Saliency maps and Grad-CAM++ were employed for model interpretability.

Main Results:

• The 3D-CNN achieved high classification accuracy: 78.3% with the sensorimotor network (SMN) and 83.3% with the visual network (VN).
• Explainability techniques highlighted key brain regions contributing to classification.
• Identified regions showed consistency with intergroup differences found in dual regression analysis.

Conclusions:

• A novel, explainable ML model was developed for rs-fMRI feature extraction and classification.
• Altered functional integrity in the SMN and VN was observed in ALS patients.
• The pipeline demonstrates potential for explainable classification of neurological diseases using rs-fMRI.