Automated Rating of Multiple Sclerosis Test Results Using a Convolutional Neural Network

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Imaging

Background:

• Multiple Sclerosis (MS) frequently causes cognitive deficits, impacting patients' quality of life.
• The Brief Visuospatial Memory Test-Revised (BVMT-R) is a standard tool for assessing visual-spatial memory deficits in MS.
• Manual scoring of the BVMT-R is time-consuming and subjective, necessitating automated solutions.

Purpose of the Study:

• To digitize the BVMT-R for automated analysis.
• To develop and evaluate a machine learning (ML) algorithm for rating MS cognitive deficits.
• To assess the feasibility of ML-driven automated follow-ups for MS progression.

Main Methods:

• A convolutional neural network (CNN) was employed to analyze digitized patient drawings from the BVMT-R.
• A dataset of 624 physician-rated drawings from 135 MS patients served as the gold standard for training.
• The CNN's accuracy in determining BVMT-R point values was evaluated.

Main Results:

• The CNN achieved classification accuracy between 57% and 76% for BVMT-R scoring.
• Accuracy exceeded 80% when the training set included over 40 drawings per patient.
• The current ML model requires further refinement due to limitations in the training sample size.

Conclusions:

• Current ML classification accuracy necessitates continued physician involvement in BVMT-R analysis.
• ML-based pre-classification can support, but not replace, physician interpretation.
• Future improvements in ML algorithms hold potential for fully automated MS cognitive assessment.