MRI-based detection of multiple sclerosis using an optimized attention-based deep learning framework

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Neurology

Background:

• Multiple Sclerosis (MS) is a chronic neurological disorder impacting millions globally.
• Early detection of MS is critical to mitigate long-term disability.
• Distinguishing MS lesions from other brain anomalies in MRI scans presents a diagnostic challenge.

Purpose of the Study:

• To develop and validate a novel deep learning framework, 2DRK-MSCAN, for early and accurate detection of MS lesions using MRI data.
• To enhance the precision and robustness of MS lesion identification through advanced AI techniques.

Main Methods:

• Utilized Gradient Domain Guided Filtering (GDGF) for pre-processing and image quality enhancement.
• Employed an EfficientNetV2L backbone within a U-shaped encoder-decoder architecture for segmentation and feature extraction.
• Integrated deep diffusion residual kernels and multiscale snake convolutional attention mechanisms in the 2DRK-MSCAN model for classification.

Main Results:

• The 2DRK-MSCAN framework achieved an outstanding 99.9% accuracy in cross-validation experiments.
• Demonstrated high precision in differentiating MS lesions from other brain anomalies.
• Validated on three publicly available MRI-based brain tumor datasets.

Conclusions:

• The 2DRK-MSCAN framework provides a reliable and effective solution for early MS detection via MRI.
• Shows significant potential to support timely intervention and improve patient care for Multiple Sclerosis.
• Clinical validation is currently underway to further confirm its efficacy.