Portable Ultra-Low Field MRI Deep-Learning Algorithms for White Matter Lesion Segmentation Improve Accuracy and...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Accurate white matter lesion (WML) identification is crucial for multiple sclerosis (MS) diagnosis and monitoring.
• Portable ultra-low field (pULF) MRI offers a mobile and cost-effective alternative for visualizing WMLs.

Purpose of the Study:

• To compare the accuracy of machine-learning (ML) and deep-learning (DL) automated segmentation tools for WMLs on pULF MRI.
• To evaluate the association of WML volumes derived from automated segmentation with clinical disability measures.

Main Methods:

• Paired pULF (64mT) and high-field (3T) MRI scans from 84 adults with MS or suspected MS were analyzed.
• Six automated WML segmentation methods (MIMoSA, WMH-SynthSeg, nnU-Net, PLAn) were applied to pULF scans and compared against manual reference segmentations.
• Dice Similarity Coefficient (DSC) was used to assess segmentation accuracy.

Main Results:

• Deep learning algorithms, specifically nnU-Net and PLAn, demonstrated superior WML segmentation accuracy on pULF MRI compared to ML methods.
• WML volumes estimated by the best-performing DL algorithms (PLAn-FL, nnU-Net-FL, PLAn-FL/T1, nnU-Net-FL/T1) were significantly associated with clinical disability scores (EDSS, SNRS).

Conclusions:

• Automated DL segmentation tools, particularly nnU-Net and PLAn, provide accurate WML quantification on pULF MRI.
• These tools can effectively reflect disease severity by correlating WML burden with clinical disability, enhancing the utility of pULF MRI in clinical trials and patient management.