Research overview
Beyond the PHQ-9 Score
CocoroLens is a research-based decision-support app that uses explainable AI to examine PHQ-9 changes during and after COVID-19 in Japan.
It goes beyond prediction by showing the psychosocial and behavioral factors that may shape changes in depressive symptom burden.
Why it matters
Why this research matters
Mental health changed during COVID-19
CocoroLens examines changes in depressive symptoms across COVID-19 and post-COVID survey phases in Japan.
Users need interpretable support
CocoroLens uses repeated surveys to examine mental health transitions beyond a single snapshot.
Longitudinal data captures change over time
CocoroLens uses repeated surveys to examine mental health transitions beyond a single snapshot.
Data flow
What CocoroLens studies
CocoroLens turns four survey phases into transition windows, connecting prior PHQ-9 history, current context, and the next observed outcome.
History Data
Past PHQ-9 scores and prior symptom patterns.
Current Status
Current psychosocial and behavioral context.
Next Changes
Observed changes linked to future PHQ-9 outcomes.
Methodology
Explainable AI Approach to Understanding Mental Health
Mental health AI must predict reliably and clearly explain why, so users can trust it.
COVID-19 Survey
White Box AI
LENS-D
Estimate depression score (PHQ-9) using Explainable Boosting Machines (EBM).
Counterfactual Explanations
- Main risk factors
- Personalized instructions
- Minimal input changes that could shift the prediction toward a more desirable outcome.
RESEARCH QUESTION
What May Happen Next? Will depressive symptoms improve, remain stable, or worsen, and which factors may help explain why?
Explainable AI
Explainable AI, not a black box
The CocoroLens app builds on research comparing interpretable models, using Explainable Boosting Machines with rule-based safety checks. Each feature's contribution stays visible, showing what shaped estimates of recovery, reassurance, or regression risk.
Research summary
Research Summary
The PHQ-9 questionnaire screens and tracks depressive symptom severity over time. This study models whether respondents recover from elevated depressive symptoms, remain reassured with low symptoms, or regress into elevated symptoms across COVID and post-COVID phases.
The regress model remains future work because regression cases were a small minority class.
In the app
What CocoroLens shows
PHQ-9 Transition
Reviews movement from current PHQ-9 status to next observed outcome.
Contributing Factors
Shows which history, context, and change features increased or lowered the model estimate.
Counterfactual Review
Allows users to explore how possible changes may affect the model's estimate.
Safety Checks
Flags predictions that may need closer clinical review before interpretation.
Use and limitations
Decision support only
CocoroLens is a research-use decision-support app. It does not diagnose, replace clinical judgment, or provide emergency mental health care. Its outputs should be reviewed together with clinical context, user history, and professional judgment.
Acknowledgments
Acknowledgments
Joshua Mendoza
- The CocoroLens mascots, Setsuko (pink) and Coco (brown), were conceptualized and visually developed with the assistance of OpenAI's ChatGPT (GPT-5.5).
- The CocoroLens application's user interface and visual design were developed based on the brand book created by Marc Cabatuan, a branding and visual identity professional.
Tetsuya Yamamoto
- Grant-in-Aid for Scientific Research of the Japan Society for the Promotion of Science (JSPS KAKENHI): Grants 18K13323, 21H00949, and 23K20771
- Project for Creative Research of the Faculty of Integrated Science, Tokushima University
- Japanese Psychological Association Research Grant for the COVID-19 Pandemic
Chigusa Uchiumi
- Grant-in-Aid for Scientific Research of the Japan Society for the Promotion of Science (JSPS KAKENHI): Grant 20K10883
- Tokushima University Research Director's Discretionary Funds
Nagisa Sugaya
- Grant-in-Aid for Scientific Research of the Japan Society for the Promotion of Science (JSPS KAKENHI): Grant 22K10586
Paul Rossener Regonia
- Ph.D. Incentive Awards, Office of the Chancellor of the University of the Philippines Diliman through the Office of the Vice Chancellor for Research and Development (UPD OVCRD), Project No. 252522 PhDIA Y2
- Engineering Research and Development for Technology for the ERDT Post-Doctoral Fellowship Grant
Project collaborators
- Joaquin Salvador
Research team