LSTM-Based Multimodal Model for Self-Harm Prediction Combining Mental Health Case Reports and Biometric Signals
Year: 2024
Type: article
Abstract: This study focuses on developing an Artificial Intelligence (AI) model for more accurately predicting and preventing impulsive suicidal behaviors in psychiatric wards by integrating Case Report Form (CRF) data with biosignal data.The data utilized in this research consists of CRF information collected from Seoul National University Hospital, Yongin Mental Hospital, and Dongguk University Hospital, along with biosignal data gathered using wearable devices provided by HealthConnect Inc. Considering the complexity of the data, missing and outlier detections were conducted, followed by dimension reduction through SHAP Value analysis to enhance the applicability of the model in actual clinical environments.The AI model developed in this paper is based on LSTM (Long Short-Term Memory) and features a multi-input structure that can simultaneously process CRF data and biosignal data.This approach not only accounts for the patterns of changing biosignals over time but also considers the comprehensive clinical characteristics of patients to assess the risk of suicide.Notably, this model proposes a new methodology by utilizing sensor data for the early detection of impulsive suicidal actions, distinguishing it from traditional methods reliant on mental scales or voice analysis.By exploring the potential application of AI in mental health care and suicide prevention, particularly in high-risk environments like psychiatric wards, this study offers a novel approach to detecting and preventing impulsive suicidal behaviors.It is anticipated that this research will improve the quality of mental health services and provide a foundation for more effective interventions for at-risk patients, thereby enhancing the overall management and prevention of suicide.
Language: English
Subfield: Clinical Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: bronze
OpenAlex ID: W4402026550