Depression Scale Prediction with Cross-Sample Entropy and Deep Learning.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Computational Neuroscience
• Psychiatry

Background:

• Accurate prediction of depression severity is crucial for effective treatment.
• Resting-state functional magnetic resonance imaging (rs-fMRI) offers insights into brain function.
• Deep learning models show promise in analyzing complex neuroimaging data.

Purpose of the Study:

• To develop and validate a two-stage deep learning scheme for predicting the Hamilton Depression Scale (HAM-D).
• To assess the efficacy of cross-sample entropy (CSE) derived from rs-fMRI as input for deep learning models.
• To compare the proposed model's performance against a single-stage regression model.

Main Methods:

• A two-stage deep learning approach was implemented.
• Cross-sample entropy (CSE) was calculated for 90 brain regions from rs-fMRI data.
• CSE maps were converted to 3D volumes and used as input for deep learning models for HAM-D prediction.

Main Results:

• The proposed two-stage deep learning scheme achieved lower root mean square errors (RMSE) for HAM-D prediction.
• RMSE values during training, validation, and testing were 2.73, 2.66, and 2.18, respectively.
• The two-stage model outperformed a single-stage regression model in prediction accuracy.

Conclusions:

• The developed two-stage deep learning scheme effectively predicts HAM-D scores using rs-fMRI derived CSE.
• This approach offers a promising tool for objective depression severity assessment.
• The findings highlight the potential of deep learning in psychiatric neuroimaging analysis.