Enhancing classification of active and non-active lesions in multiple sclerosis: machine learning models and feature
Atefeh Rostami1,2, Mostafa Robatjazi3,4, Amir Dareyni5
1Department of Medical Physics and Radiological Sciences, Sabzevar University of Medical Sciences, Sabzevar, Iran.
BMC Medical Imaging
|December 20, 2024
Summary
Deep learning models show promise in classifying multiple sclerosis (MS) lesions from MRI scans. A sequential deep learning model achieved 95.60% AUC, outperforming traditional machine learning approaches for MS plaque detection.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Neurology
Background:
- Gadolinium-enhanced T1-weighted MRI is standard for active multiple sclerosis (MS) lesion detection.
- This study evaluates machine learning (ML) and deep learning (DL) for classifying MS lesions using T2-weighted MRI.
- Focuses on distinguishing active from non-active MS lesions.
Purpose of the Study:
- To assess the performance of 16 ML and 1 sequential DL model.
- To classify active versus non-active MS lesions using T2-weighted MRI.
- To identify the most effective models for MS lesion characterization.
Main Methods:
- Extracted 107 features from 75 active and 100 non-active MS lesions using 3D Slicer.
- Developed 16 ML and 1 sequential DL models with 5-fold cross-validation.
- Evaluated model performance using accuracy, precision, sensitivity, specificity, AUC, and F1 score on test data.
Main Results:
- The sequential DL model achieved the highest Area Under the Curve (AUC) of 95.60% on the test dataset.
- Hybrid Gradient Boosting Classifier (HGBC) showed a test AUC of 86.75%.
- Gradient Boosting Classifier (GBC) achieved an 87.92% AUC during cross-validation.
Conclusions:
- The study demonstrated the effectiveness of a sequential DL model for MS lesion classification.
- Ensemble methods and DL approaches show robust predictive performance.
- These findings support the potential of AI in classifying MS plaques.
