An Exploratory Study on the Prediction of Recurrence Risk of Bipolar Disorder Using Sentiment Analysis Technology Based on Multi-modal Feature Fusion

Bipolar disorder (BD) has become a significant public health problem with complex clinical manifestations, difficult treatment, and poor prognosis. However, there is still a lack of effective biological markers for diagnosing and predicting recurrence. Sentiment analysis computing usually refers to using machine equipment to classify, identify, interpret, and imitate human emotions. However, current multi-modal emotion analysis research is mainly based on one or two modalities. Due to the diversity and complexity of patients' emotional expressions, this single- and dual-modal information analysis is far from enough for accurate discrimination of emotional symptoms. Only emotion analysis technology based on multi-modal feature fusion can make more precise and effective judgments. The current project is based on our previous research on cognitive neuroimaging and big data analysis of bipolar disorder. The investigators plan to enroll 200 BD patients who meet DSM-5 diagnostic criteria and 200 healthy controls. The investigators will use sentiment analysis technology with multi-modal feature fusion (text data, audio and visual modalities, eye movements, and electrophysiology) to identify BD recurrence. Biological markers for risk prediction and an algorithm model for joint judgment of multi-source information will be established to analyze the characterization data. The effectiveness of this recurrence prediction model will be further verified and optimized through a large-sample, prospective cohort study design. It is hoped that it can provide a new method for predicting the recurrence risk of BD patients. In the near future, clinical decision-making aids based on this auxiliary method can be developed, and the translational application value of clinical diagnosis and treatment can be explored.