A deep learning approach for automated scoring of the Rey-Osterrieth complex figure.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Memory deficits are common in neurological and psychiatric disorders.
• The Rey-Osterrieth complex figure (ROCF) test is a standard tool for assessing non-verbal visual memory.
• Manual ROCF scoring is subjective, time-consuming, and prone to variability.

Purpose of the Study:

• To develop and validate a deep learning model for automated ROCF scoring.
• To improve the objectivity, reliability, and efficiency of ROCF assessment.
• To provide a digital tool for healthcare institutions to evaluate memory performance.

Main Methods:

• Collected over 20,000 hand-drawn ROCF images from diverse participants.
• Utilized crowdsourced human intelligence for unbiased ground truth scoring.
• Trained and evaluated a multihead convolutional neural network (CNN) for automated scoring.

Main Results:

• The AI model demonstrated highly unbiased predictions, closely matching ground truth scores.
• The CNN outperformed both online raters and human clinicians in scoring accuracy.
• The system reliably identified and scored individual figure elements, enhancing explainability.
• Performance was validated in a large, independent, pre-registered prospective study.

Conclusions:

• An AI-powered system can objectively, reliably, and efficiently score ROCF drawings.
• This technology offers a valuable digital tool for assessing memory deficits globally.
• Automated ROCF scoring has the potential to standardize and improve neuropsychological assessments.