Digitalisation of the Brief Visuospatial Memory Test-Revised and Evaluation with a Machine Learning Algorithm

Area of Science:

• Neurology
• Computer Science
• Medical Technology

Background:

• Multiple sclerosis (MS) presents with diverse neurological symptoms, including cognitive dysfunction.
• Assessing cognitive impairment is crucial for managing MS progression.
• The Brief Visuospatial Memory Test-Revised (BVMT-R) is a standard tool for evaluating visual memory deficits, traditionally administered on paper.

Purpose of the Study:

• To develop and validate a novel, efficient digital tool for administering and scoring the BVMT-R.
• To assess the performance of a machine learning (ML) algorithm in rating BVMT-R drawings.
• To compare the accuracy of ML-based scoring with human expert ratings.

Main Methods:

• Digitalization of 1,525 BVMT-R drawings from patients.
• Development of a machine learning (ML) algorithm for automated scoring.
• Splitting the dataset into training and testing sets, incorporating prior data.
• Implementing a semi-automated rating system with a reliability threshold for manual review.

Main Results:

• The ML algorithm achieved 72% and 79% agreement with two neuropsychologists on the test dataset.
• A semi-automated approach, routing drawings below a 78.8% reliability threshold for manual review, improved ML agreement to 80.3% and 86.6%.
• This semi-automated method requires manual checking of only 17.4% of drawings, matching expert rater performance.

Conclusions:

• A digital administration of the BVMT-R using a tablet app is feasible and maintains quality.
• The developed ML algorithm, particularly in its semi-automated configuration, offers reliable and efficient cognitive assessment for MS.
• This technology streamlines the BVMT-R process, providing results comparable to traditional expert evaluation.