Explainable AI-Driven Clinical Decision Support System for Real-Time Diagnosis in Critical Care Environments

Abstract

The issue of ensuring timely and accurate diagnosis in intensive care units (ICUs) is significant because of the constant stream of high-dimensional data about patients and the necessity to make a quick clinical decision. The process of delay or inaccuracies in diagnosis can greatly affect the patient, leaving the need to create intelligent and reliable decision support systems. This paper hypothesizes an explainable artificial intelligence (XAI)-based clinical decision support system (CDSS) that can be used in critical care settings to diagnose patients in real-time. The suggested framework involves a hybrid model of forecasting with Long Short-Term Memory (LSTM) networks to extract patterns of time-related variations, and Extreme Gradient Boosting (XGBoost) to classify organized clinical information. The system supports SHapley Additive exPlanations (SHAP) to explain model predictions on both global and patient-specific levels to improve transparency and clinical trust. Experimental analysis of benchmark ICU data points to a better and more beneficial effect on the diagnostic performance of the proposed model, where the accuracy, sensitivity, and specificity are better than the traditional ones, but the inference latency of the model remains low enough to implement it in real time. In addition, the explainability module offers valuable information on significant clinical characteristics that affect predictions, which can guide healthcare experts to make informed decisions. These findings demonstrate that the suggested XAI-based CDSS is a dependable, interpretable, and scalable tool to improve diagnostic efficiencies in critical care environments, and has a high potential to be embedded in the next-generation smart healthcare system.