Personalized Patient Risk Prediction Using Multi-modal AI on EHR and Medical Imaging Data

Patient risk prediction means understanding the probability of a particular person to be at risk of developing certain health risks in view of his or her characteristics. This capability is especially important in healthcare, which makes it possible to address the main causes of many diseases and develop unique treatment regimens for improving the quality of life. However, many previous works still have shortcomings; for example, most rely on single-modal data, resulting in inadequate evaluation and lower prediction performance. However, traditional models may fail to capture multifaceted interactions within the parameters hence degrading the performance. This paper presents a new method that combines multi-modal AI by using both Electronic Health Records (EHR) and medical imaging through one Bidirectional Long Short-Term Memory (Bi-LSTM) model. Through using Bi-LSTM which can capture the sequential dependencies in both modality of data, the method is expected to improve the accuracy. The performance of all the proposed metrics of the model shows high accuracy of 99.7% of patient risk assessment with high precision, sensitivity and F1-score making this model robust and reliable for patient risk assessment. The model is written in Python and the used libraries offer the deep learning techniques to train and evaluate the model at its best. In fact, this paper goes beyond rectifying these flaws and contributes to the development of a new methodology for applying AI techniques in healthcare among the studies. The results illuminate the promise of deeper multi-modal data integration, along with more sophisticated machine learning techniques, in enabling more accurate patient risk assessment, as well as better decision making at the point of patient care.