A Robust Dual-Stage Learning-Based Pipeline for Multiclass Segmentation of Multiple Sclerosis Lesions in MRI

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Accurate segmentation and classification of multiple sclerosis (MS) lesions are crucial for diagnosis and monitoring.
• Lesion heterogeneity presents significant challenges for automated analysis methods.

Purpose of the Study:

• To develop a robust dual-stage pipeline for automated MS lesion segmentation and classification.
• To evaluate the performance of deep learning (DL) and machine learning (ML) models in this task.

Main Methods:

• A dual-stage pipeline integrating DL for segmentation and ML for classification was developed.
• nnU-Net and UNETR++ were optimized for lesion segmentation; Random Forest was selected for classification.
• The pipeline combined nnU-Net for segmentation and Random Forest for classification.

Main Results:

• nnU-Net outperformed UNETR++ in lesion segmentation, with up to 12.8% improvement.
• Random Forest consistently outperformed DL models in classification, achieving over 12% higher performance.
• The hybrid pipeline integrating nnU-Net and Random Forest demonstrated the best overall performance.

Conclusions:

• The proposed pipeline effectively combines DL and ML strengths for accurate automated MS lesion analysis.
• This framework offers potential clinical utility for improving MS diagnosis and monitoring.
• Qualitative analysis suggests potential limitations in ground truth labeling affecting results.