Clinical Report: A Transparent TimeMIL Approach for fNIRS in Psychiatry
Overview
This study introduces a time-aware multiple instance learning model (TimeMIL) applied to fNIRS data, achieving high accuracy in distinguishing schizophrenia from bipolar disorder. The model's interpretability framework enhances understanding of neurophysiological differences between these disorders.
Background
Differentiating schizophrenia (SCZ) from bipolar disorder (BD) is clinically significant due to overlapping symptoms that can lead to misdiagnosis and delayed treatment. Functional near-infrared spectroscopy (fNIRS) offers a promising method for capturing brain activity during cognitive tasks, potentially providing objective neurobiological markers for these conditions. The development of deep learning models like TimeMIL aims to improve diagnostic accuracy and support personalized care in psychiatric settings.
Data Highlights
Group
Accuracy
AUC
TimeMIL
0.928 ± 0.016
0.984 ± 0.007
Key Findings
TimeMIL achieved an accuracy of 0.928 ± 0.016 in classifying SCZ, BD, and healthy controls.
The model demonstrated a macro-averaged AUC of 0.984 ± 0.007.
Attribution analyses revealed significant differences in the orbitofrontal cortex across groups.
SCZ-specific attribution patterns were noted in the frontal pole cortex and ventrolateral prefrontal cortex.
TimeMIL outperformed traditional models such as 1D-CNNs and Transformers.
Clinical Implications
The TimeMIL model provides a robust tool for enhancing the differential diagnosis of SCZ and BD, potentially reducing misdiagnosis rates. Its interpretability framework allows clinicians to understand the neurophysiological basis of predictions, fostering trust in the diagnostic process.
Conclusion
The application of TimeMIL to fNIRS data represents a significant advancement in psychiatric diagnostics, combining high accuracy with interpretability. Further validation in clinical settings is necessary to establish its utility in routine practice.
Swedish study finds two-way associations between premenstrual disorders and psychiatric conditions, with strongest links involving depression, anxiety, attention-deficit/hyperactivity disorder, bipolar disorder, and personality disorders.
Chemsex at the pharmacy counter. Gut bacteria tracking helmet impacts. PMD predicting psychiatric illness bidirectionally. This week's research keeps landing in the same uncomfortable place: medicine is improvising.
