Graph Learning-Based Suicidal Ideation Detection Via Tree-Drawing Test

Strategic Insights -

Researchers at the South China University of Technology have developed a novel graph learning approach to detect suicidal ideation in adolescents. The Tree-Drawing Test (TDT) is an effective tool for suicidal ideation detection, constructed by annotating psychological features automatically from tree-drawing images. A Graph Convolutional Network (GCN) model is used to realize individual suicidal ideation detection. The proposed method has demonstrated significantly improved performance compared to traditional machine learning and convolutional neural network approaches. The ablation study highlights the effectiveness of feature 'leaves and fruits' in detecting suicidal ideation. The research aims to achieve a balance between accuracy and scalability in suicidal ideation detection.

Key Takeaways:

  • The Tree-Drawing Test (TDT) is an effective tool for suicidal ideation detection, utilizing a novel graph learning approach to enable automatic application.
  • The proposed method constructs a semantic graph based on psychological features annotated automatically from tree-drawing images.
  • A Graph Convolutional Network (GCN) model is used to realize individual suicidal ideation detection, demonstrating improved performance compared to traditional approaches.
  • The ablation study shows that feature 'leaves and fruits' is effective in detecting suicidal ideation in adolescents.
  • The research aims to balance accuracy and scalability in suicidal ideation detection, highlighting its effectiveness in large-scale screening.
  • The proposed method remains stable even when the model is disturbed, maintaining model stability.
  • The results demonstrate the effectiveness of the proposed method in detecting suicidal ideation in a real-world dataset of 806 students.

Statistics:

  • 806 students from primary and secondary school in Shaanxi Province, China, were included in the real-world dataset for evaluation.
  • The proposed method achieved a macro-F1 score of, G-mean of, and false positive rate of in the ablation study.
  • The ablation study demonstrated improvements in feature 'leaves and fruits' in detecting suicidal ideation compared to traditional approaches.
  • The proposed method remained stable with a model stability rate of even when a tree-drawing image could not be fully represented.

Sources:

  • "Graph learning based suicidal ideation detection via tree-drawing test." Frontiers in Psychiatry, 2025;16:1617650.
  • "Researchers at South China University of Technology Discuss Findings in Technology (Graph learning based suicidal ideation detection via tree-drawing test)." Journal of Engineering. August 18, 2025; p 2396.