Differentiating multiple sclerosis from non-specific white matter changes using a convolutional neural network image...

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Diagnosing multiple sclerosis (MS) often involves magnetic resonance imaging (MRI) and ruling out similar conditions.
• Non-specific white matter disease (NSWMD) frequently mimics MS on MRI, complicating diagnosis.
• Distinguishing MS from NSWMD can necessitate further tests or extended observation.

Purpose of the Study:

• To develop and assess machine learning models for differentiating MS from NSWMD.
• To improve the accuracy and efficiency of MS diagnosis using neuroimaging data.

Main Methods:

• Adult patients (2015-2019) with brain MRI using a demyelinating protocol and available medical records were included.
• Diagnoses of MS and NSWMD were confirmed via clinical documentation.
• Logistic regression and convolutional neural network (CNN) models were trained on matched MS and NSWMD cases based on T2 lesion volume (T2LV).

Main Results:

• 250 MS and 250 NSWMD MRI scans were analyzed, matched by T2LV.
• A logistic regression model achieved 68.0% accuracy in differentiating MS from NSWMD using 20 variables.
• CNN models demonstrated an average accuracy of 77% and 78% in independent validation and testing cohorts.

Conclusions:

• Automated machine learning approaches can effectively distinguish MS from NSWMD on MRI.
• These AI-driven methods offer supplementary diagnostic support for patients evaluated for MS.