← All projects

ML analysis of Rey Complex Figure performance

Interpretable machine learning explores how education, sex, and age shape Rey Complex Figure performance.

Education, Sex, and Age Shape Rey Complex Figure Performance in Cognitively Normal Adults: An Interpretable Machine Learning Study

Journal of Clinical Medicine · 14(21), 7562

Read the paper (opens in a new tab)

Study overview

  • Analyzed Complex Figure performance in 926 cognitively healthy Emory Healthy Brain Study participants, ages 45–80, with MoCA scores of at least 24.
  • Combined individual drawing-element scores, totals, completion times, and MoCA measures to classify education, sex, and age using support-vector machines, logistic regression, and random forests.

Key findings

  • Education was associated with copy, immediate-recall, and delayed-recall totals. Age and sex were associated with both recall totals; sex was not associated with the copy total.
  • Copy time, immediate-recall time, and MoCA were consistently important predictors. Logistic regression had the highest reported accuracy for education and age, while random forest performed best for sex.
  • The cross-sectional sample was predominantly White and female, and the models had no external validation cohort. The findings describe demographic associations, not causal effects or a validated diagnostic tool.

Figures

Pipeline extracting Rey Complex Figure scores for statistical and machine-learning analysis of demographic effects
Figure 2 — Feature extraction and predictive analysis pipeline for Complex Figure scores.Source: Lee et al. (2025), Figure 2 · CC BY 4.0
Bar charts comparing accuracy, precision, and recall of SVM, logistic-regression, and random-forest models for education, sex, and age classification
Figure 4 — Classifier performance by demographic target. Logistic regression achieved the highest accuracy for education and age; random forest performed best for sex.Source: Lee et al. (2025), Figure 4 · CC BY 4.0
Venn diagram showing shared and target-specific Complex Figure and MoCA predictors for education, sex, and age
Figure 5 — Overlap among the top 15 predictive features for each demographic target, highlighting shared and target-specific measures.Source: Lee et al. (2025), Figure 5 · CC BY 4.0
← All projects