Deep Learning-Based Automated Detection and Burden Assessment of Paramagnetic Rim Lesions on Quantitative...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Multiple Sclerosis Research

Background:

• Paramagnetic rim lesions (PRLs) are a hallmark of multiple sclerosis (MS) pathology.
• Accurate detection and quantification of PRLs are crucial for MS assessment.
• Current methods for PRL assessment can be labor-intensive and subjective.

Purpose of the Study:

• To develop and validate a deep learning (DL) framework for automated detection and burden assessment of PRLs using quantitative susceptibility mapping (QSM).
• To compare the performance of QSM-only versus QSM + fluid-attenuated inversion recovery (FLAIR) configurations.
• To evaluate the clinical relevance of the automated PRL burden in patients with MS.

Main Methods:

• Retrospective collection of brain MRI data (QSM and 3D FLAIR) from 106 patients with suspected MS.
• Manual segmentation of PRLs for ground truth establishment.
• Training a 3D nnU-Net DL framework using QSM-only and QSM + FLAIR data for lesion and patient-level classification and burden assessment.

Main Results:

• The QSM-only DL model demonstrated comparable or superior lesion-level sensitivity and precision compared to the QSM + FLAIR model in internal testing.
• Both QSM-only and QSM + FLAIR models achieved 100% patient-level sensitivity in the temporal test set.
• A higher automated PRL burden was significantly associated with poorer cognitive function (P = 0.002).

Conclusions:

• Deep learning models utilizing QSM enable automated and efficient detection and burden assessment of PRLs in MS.
• The QSM-only approach offers a simplified pipeline with performance comparable to multi-sequence methods.
• Automated PRL burden shows potential as a valuable imaging biomarker for assessing cognitive impairment in MS.