CFRAFN: A Cross-Feature Residual Attention Fusion Network for Major Depressive Disorder Prediction Using Clinical...

Area of Science:

• Computational psychiatry
• Machine learning in healthcare
• Speech signal processing

Background:

• Major depressive disorder (MDD) is a widespread mental health condition requiring early detection.
• Voice analysis offers a non-invasive method to identify behavioral indicators of MDD.
• Existing methods for MDD detection using voice data can be improved with advanced AI techniques.

Purpose of the Study:

• To develop and evaluate a novel deep learning model, the cross-feature residual attention fusion network (CFRAFN), for detecting MDD from voice recordings.
• To construct a comprehensive Chinese depressive voice dataset for training and testing the model.
• To assess the effectiveness of integrating acoustic features and deep embeddings for enhanced MDD prediction.

Main Methods:

• Collected voice data from 221 MDD patients and 113 healthy controls to create the Chinese depressive voice dataset.
• Proposed the CFRAFN model, combining extended Geneva minimalistic acoustic parameter set features with VGGish embeddings.
• Employed residual blocks for training stability and a self-attention fusion strategy for optimal feature integration.

Main Results:

• The CFRAFN model achieved a high area under the receiver operating characteristic curve (AUC) of 0.924 on an independent test set.
• CFRAFN significantly outperformed 11 baseline models in 5-fold cross-validation.
• The model demonstrated effective capture of MDD-associated phonetic patterns through integrated feature modalities.

Conclusions:

• The proposed CFRAFN model shows excellent performance in detecting major depressive disorder using voice data.
• Integrating diverse voice features with advanced deep learning architectures can significantly improve diagnostic accuracy for MDD.
• Voice analysis holds promise as a valuable tool for objective and accessible mental health screening.