Cross-Sectional Validation of an Automated Lesion Segmentation Software in Multiple Sclerosis: Comparison with...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Quantitative MRI Analysis

Background:

• Magnetic Resonance Imaging (MRI) is crucial for assessing disease burden in multiple sclerosis (MS).
• Accurate quantification of white matter lesions (WML) is essential for monitoring MS progression and treatment efficacy.
• Current WML assessment methods can be time-consuming and subjective.

Purpose of the Study:

• To evaluate the effectiveness of a commercial k-nearest neighbors (k-NN) software for quantifying WML burden in MS.
• To compare the software's WML quantification accuracy against expert radiologists' assessments.
• To determine the utility of AI-powered post-processing in MS MRI interpretation.

Main Methods:

• Retrospective review of brain MRIs from 32 MS patients and 19 non-MS controls.
• Processing MRI images using AI-powered, cloud-based k-NN software for WML quantification.
• Comparison of software-generated WML data with semi-quantitative assessments by blinded radiologists and neuroradiologists.

Main Results:

• The k-NN software achieved 94.1% accuracy for WML count and 84.3% for WML volume in differentiating MS from non-MS subjects.
• Software accuracy was comparable to radiologists' assessments (90.2%-94.1%).
• Lesion segmentation accuracy was higher in deep WM and infratentorial regions compared to juxtacortical regions (p <0.001).

Conclusions:

• k-NN derived WML volume and count are valuable quantitative metrics for MS disease burden.
• AI-powered post-processing software can enhance the interpretation of brain MRIs in MS patients.
• This technology offers potential for more objective and efficient MS assessment.