Differentiation between multiple sclerosis and neuromyelitis optica spectrum disorders by multiparametric...

Area of Science:

• Neuroimmunology
• Radiology
• Artificial Intelligence

Background:

• Multiple sclerosis (MS) and neuromyelitis optica spectrum disorders (NMOSD) are distinct neuroinflammatory diseases with overlapping clinical presentations.
• Accurate differentiation is crucial for appropriate treatment and management.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) model for differentiating MS from anti-aquaporin-4 antibody-positive NMOSD using magnetic resonance imaging (MRI) data.

Main Methods:

• A CNN model, based on SqueezeNet architecture, was developed to analyze R1, R2 relaxation rates, and proton density maps from 3-T MRI scans.
• Features were separated into image-specific and group-common to prevent overfitting on a small dataset.
• Leave-one-out cross-validation was employed to assess model performance.

Main Results:

• The CNN model achieved an area under the receiver operating characteristic curve of 0.859 for differentiating between MS and NMOSD.
• The model demonstrated a sensitivity of 80.0% for MS, 83.3% for NMOSD, and an overall accuracy of 81.1%.

Conclusions:

• A CNN model effectively differentiates between MS and NMOSD using MRI data, with built-in mechanisms to mitigate overfitting.
• This AI-driven approach shows potential to assist clinicians in the differential diagnosis of these challenging neurological disorders.