Deep learning for multiple sclerosis lesion classification and stratification using MRI.

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Multiple sclerosis (MS) is a chronic neurological disease causing central nervous system damage.
• Conventional MRI struggles to detect subtle MS lesions in critical brain regions.
• Improved lesion detection is crucial for accurate diagnosis and treatment.

Purpose of the Study:

• To develop and validate a deep learning approach for precise MS lesion classification and stratification.
• To enhance the detection of small or subtle lesions, particularly in the cortical gray matter and brainstem.
• To improve the overall diagnostic accuracy of magnetic resonance imaging (MRI) in multiple sclerosis patients.

Main Methods:

• A convolutional neural network (CNN) with dual attention mechanisms was designed for high-resolution T2-weighted imaging.
• Deep learning-based reconstruction (DLR) techniques augmented the CNN model.
• A preprocessing pipeline included bias field correction, skull stripping, registration, and normalization, followed by validation on public datasets.

Main Results:

• The deep learning model achieved high performance: 96.27% precision, 95.54% sensitivity, 94.70% specificity, and 0.975 AUC.
• The method demonstrated superior performance in detecting lesions in challenging areas like the cortical gray matter and brainstem.
• Attention mechanisms improved the model's ability to identify critical MRI features for better lesion classification.

Conclusions:

• The study introduces a scalable deep learning framework for MS lesion detection and classification.
• The proposed approach offers a practical solution with superior diagnostic accuracy and generalizability for clinical applications.
• This work establishes a new benchmark for MS diagnosis and management, supporting personalized treatment strategies.