Using deeply time-series semantics to assess depressive symptoms based on clinical interview speech

Area of Science:

• Psychiatry
• Computational Linguistics
• Artificial Intelligence

Background:

• Depression significantly burdens global health, necessitating effective assessment tools.
• Current methods like rating scales are subjective; Measurement-Based Care (MBC) emphasizes symptom assessment.
• Artificial Intelligence (AI) offers objective and consistent performance for symptom evaluation.

Purpose of the Study:

• To apply Deep Learning (DL) and Natural Language Processing (NLP) for assessing depressive symptoms during clinical interviews.
• To propose and evaluate a novel algorithm model for objective depression assessment.

Main Methods:

• A cohort of 329 patients with Major Depressive Episode was studied.
• Clinical interviews based on the Hamilton Depression Rating Scale (HAMD-17) were conducted and audio-recorded.
• A deep time-series semantics model, Multi-granularity and Multi-task joint training (MGMT), was developed for symptom assessment.

Main Results:

• The MGMT model achieved an F1 score of 0.719 for classifying four levels of depression severity.
• The model demonstrated an F1 score of 0.890 in identifying the presence of depressive symptoms.
• Results indicate the feasibility of using DL-NLP techniques for depression assessment.

Conclusions:

• Deep Learning and Natural Language Processing techniques are feasible for assessing depressive symptoms from clinical interviews.
• The proposed MGMT model shows promising performance in depression severity classification and detection.
• Future work should explore multi-dimensional models incorporating voice, facial expressions, and personalized data for enhanced assessment.