Automated segmentation of multiple sclerosis lesions, paramagnetic rims, and central vein sign on MRI provides...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Multiple sclerosis (MS) diagnosis relies on detecting central nervous system lesions via MRI.
• Current criteria can be improved by incorporating MS-specific imaging biomarkers like paramagnetic rim lesions (PRLs) and central vein sign (CVS).
• Manual segmentation of these features is time-consuming and subjective, hindering clinical application.

Purpose of the Study:

• To develop and validate a fully-automated method for simultaneous segmentation of MS lesions, PRLs, and CVS.
• To improve the accuracy and efficiency of MS diagnosis using advanced neuroimaging analysis.
• To overcome the limitations of manual segmentation in clinical practice.

Main Methods:

• Developed Automated Lesion, PRL, and CVS Analysis (ALPaCA), a joint segmentation method.
• Trained ALPaCA using subject-level cross-validation on 47 adults with MS and 50 with MS mimics.
• Employed a 3D convolutional neural network with an attention mechanism for lesion, PRL, and CVS prediction from multi-contrast MRI scans.

Main Results:

• ALPaCA achieved high cross-validation AUROCs: 0.95 for lesions, 0.91 for PRLs, and 0.87 for CVS, outperforming previous methods (p < 0.001).
• Subject-level ALPaCA scores showed higher correlations with manual counts for lesions and PRLs compared to existing methods (p < 0.001; p = 0.03).
• ALPaCA's PRL and CVS scores were highly associated with MS diagnosis in logistic regressions (p < 0.001).

Conclusions:

• ALPaCA enables fully-automated, simultaneous segmentation of MS lesions, PRLs, and CVS using clinically feasible MRI scans.
• The automated segmentations demonstrate superior performance at both lesion and subject levels compared to current methods.
• This technology has the potential to enhance the accuracy and efficiency of MS diagnosis and monitoring.