Functional near-infrared spectroscopy-based computer-aided diagnosis of major depressive disorder using explainable...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Limited explainable AI (XAI) models exist for functional near-infrared spectroscopy (fNIRS)-based computer-aided diagnosis (CAD) of major depressive disorder (MDD).
• Existing methods often lack transparency in identifying diagnostic biomarkers.

Purpose of the Study:

• To develop and implement an XAI model using convolutional neural networks (CNNs) for fNIRS-based CAD of MDD.
• To highlight interhemispheric asymmetry differences between MDD patients and healthy controls (HCs).
• To enhance the interpretability of deep learning models in psychiatric diagnostics.

Main Methods:

• Applied a CNN-based XAI model to fNIRS data from 48 MDD patients and 68 HCs during a verbal fluency task.
• Utilized Layer-wise Relevance Propagation (LRP) to identify input data contributions to model predictions.
• Employed ten-fold cross-validation for performance assessment.

Main Results:

• Achieved an average accuracy of 81.17%, sensitivity of 79.5%, and specificity of 82.38%.
• LRP identified the dorsolateral prefrontal cortices (DLPFC) as critical for classification.
• Revealed distinct interhemispheric asymmetry in MDD patients' brain activity patterns.

Conclusions:

• Developed a high-performing XAI model for fNIRS-based MDD diagnosis.
• Successfully visualized the deep learning model's decision-making process.
• Demonstrated the model's ability to identify neurophysiological markers of MDD.