New lesion segmentation for multiple sclerosis brain images with imaging and lesion-aware augmentation

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Multiple sclerosis (MS) is a central nervous system inflammatory disease.
• MRI-based lesion detection is vital for MS diagnosis and monitoring.
• Accurate detection of new MS lesions is crucial for assessing disease progression and treatment efficacy.

Purpose of the Study:

• To develop and evaluate a deep learning pipeline for automated detection and segmentation of new MS lesions.
• To enhance the performance of MS lesion segmentation using novel data augmentation techniques.

Main Methods:

• A deep learning pipeline built on the nnU-Net framework was proposed.
• Imaging and lesion-aware data augmentation techniques, including axial subsampling and CarveMix, were employed.
• The pipeline was evaluated on the MICCAI 2021 MS new lesion segmentation challenge (MSSEG-2) dataset.

Main Results:

• The pipeline achieved an average Dice score of 0.510 and F1 score of 0.552 on cases with new lesions.
• Low false positive rates were observed: 0.036 for lesion number and 0.192 mm³ for lesion volume.
• The proposed method demonstrated superior performance compared to other challenge participants and state-of-the-art architectures.

Conclusions:

• The developed deep learning pipeline effectively detects and segments new MS lesions.
• The use of advanced data augmentation strategies significantly improved segmentation performance.
• This approach offers a promising tool for improving MS patient management and treatment assessment.