Breakthrough in Mental Health Diagnosis: Novel Spatiotemporal Spectral Feature Fusion Network Offers Enhanced Accuracy

Strategic Insights -

Research conducted at Eulji University has led to the development of a novel spatiotemporal spectral feature fusion network (STSFF-Net) for the diagnosis of depression based on facial expressions. This breakthrough aims to address the limitations of traditional methods, which often rely on self-reported questionnaires susceptible to subjective biases. The STSFF-Net model integrates AU time-series data, spatiotemporal facial representations, and a late fusion mechanism to effectively capture depressive expressions.

Key Takeaways:

  • The STSFF-Net model demonstrated superior performance over existing methods in detecting depression from facial expressions, with a mean absolute error of 3.18 and a root mean square error of 4.78.
  • The model was validated on the Chosun University Hospital Depression Risk Video dataset, comprising video recordings from 187 clinically diagnosed participants.
  • The STSFF-Net integrates a one-dimensional residual network for transforming AU time-series data into spectral features, a three-dimensional residual network enhanced with a convolutional block attention module, and a late fusion mechanism that effectively combines local and global features.
  • This research has the potential for real-world depression screening applications, reducing the risk of subjective biases associated with traditional diagnostic methods.
  • Financial supporters for this research include the Institute for Information & Communication Technology Planning & Evaluation (IITP), Republic of Korea, and the National Research Foundation of Korea.
  • The research was conducted by Jaehyo Jung and his team at Eulji University, with additional authors Daegil Choi, Jisun Hong, Jihun Lee, Zhiqiang Liu, and Gengjia Zhang.

Statistics:

  • Mean absolute error of 3.18
  • Root mean square error of 4.78
  • 187 participants in the Chosun University Hospital Depression Risk Video dataset
  • 100% of existing methods outperformed by the STSFF-Net model in detecting depression from facial expressions
  • 100% of clinically diagnosed participants in the dataset used to validate the STSFF-Net model

Sources:

  • Stsff-net: a Two-stream Network for Depression Detection From Facial Expressions. Neurocomputing, 2025;652.
  • NewsRx. Research Conducted at Eulji University Has Updated Our Knowledge about Mental Health (Stsff-net: a Two-stream Network for Depression Detection From Facial Expressions). Mental Health Weekly Digest. November 3, 2025; p 584.