Automated versus human scoring of the Rey-Osterrieth Complex Figure Test: a rapid review

Area of Science:

• Neuropsychology
• Artificial Intelligence
• Digital Health

Background:

• Clinical neuropsychology lags in adopting automated assessment tools despite digital healthcare advances.
• Automated scoring of the Rey-Osterrieth Complex Figure Test (ROCFT) could improve efficiency and consistency.
• The clinical utility and accuracy of automated ROCFT scoring require evaluation against traditional methods.

Purpose of the Study:

• To evaluate if digital automated scoring systems for the ROCFT offer accuracy, reliability, and clinical utility equal to or superior to clinician-driven scoring.
• To assess the performance of automated ROCFT scoring systems against human raters.

Main Methods:

• A rapid review following PRISMA guidelines was conducted.
• Searches of PubMed and Web of Science (January 1, 2015 – October 12, 2025) identified studies benchmarking automated vs. human ROCFT scoring.
• Five articles, including deep-learning and rule-based algorithms, were analyzed.

Main Results:

• Five articles analyzed over 41,000 ROCFT drawings.
• Well-designed automated systems achieved expert-level performance, sometimes surpassing it.
• Deep-learning models showed high concordance with expert scoring, though performance varied with data quality.

Conclusions:

• Digital automated ROCFT scoring demonstrates comparable accuracy and reliability to traditional methods, with potential for superior performance.
• Clinical implementation requires addressing heterogeneous datasets, disorder-specific norms, and independent validation.
• Automated scoring should augment, not replace, clinical judgment, necessitating human-in-the-loop systems and governance frameworks.