Abstract

The diagnosis of depression relies on the subjective judgment and experience of doctors, which lacks sufficient objectivity and efficiency. To address the above problems, we propose temporal convolutional network (TCN) based on the Inception architecture. This network takes RR interval sequences as input and classifies these sequences in depression patients (DP) and healthy controls (HC). The TCN modules employ different dilation rates to extract features at various scales from RR interval sequences. And the Inception model architecture effectively facilitates multi-scale feature fusion. The experimental results indicate that the Inception-TCN model demonstrates strong classification performance with an Area Under Curve (AUC) of 0.9527, accuracy of 89.95%, recall of 90.49%, precision of 91.68%, and an F1-score of 0.9108. Compared to the other four deep learning models, the performance of proposed model is significantly superior.