Objective:

To apply a time-aware multiple instance learning model (TimeMIL) to fNIRS data for distinguishing schizophrenia (SCZ) from bipolar disorder (BD) during a verbal fluency task (VFT), addressing the challenges of misdiagnosis in clinical settings.

Approach:

Key Findings:

• TimeMIL achieved 0.928 ± 0.016 accuracy and a macro-averaged AUC of 0.984 ± 0.007 for three-class classification, indicating high diagnostic accuracy.
• TimeMIL significantly outperformed other models, including 1D-CNNs, Transformers, and TCNs, suggesting its superiority in capturing complex neural patterns.
• Attribution analyses revealed differences in the orbitofrontal cortex (OFC) across HC, SCZ, and BD, with SCZ-specific patterns in the frontal pole cortex (FPC) and ventrolateral prefrontal cortex (VLPFC), highlighting potential biomarkers.

Interpretation:

The study demonstrates the effectiveness of TimeMIL in classifying psychiatric disorders using fNIRS data, emphasizing the importance of interpretability for clinical applications and understanding model predictions.

Limitations:

• Further validation in clinical settings is required to confirm the findings and ensure generalizability.
• The study may be limited by the sample size and the specific tasks used, which could affect the robustness of the conclusions.

Conclusion:

The TimeMIL framework represents a novel approach to fNIRS-based psychiatric disease classification, offering high accuracy and interpretability, which could significantly enhance clinical diagnostic tools.

Attribution Notice

This content is an AI-generated, fully rewritten summary based on a published scholarly article. It does not reproduce the original text and is not a substitute for the original publication. Readers are encouraged to consult the source for full context, data, and methodology.