Real-Time Explainable Multimodal ML for Clinical Decision Intelligence A Hybrid Supervised–Unsupervised CDSS Framework

Abstract

The rapid growth of diverse clinical data, such as electronic health records (EHRs), lab results, and medical imaging, has opened new possibilities for data-driven decision-making in healthcare. However, issues with data quality, model interpretability, and workflow integration have hindered the safe and effective use of machine learning (ML) in clinical settings. This paper presents a clear multimodal ML framework for real-time clinical decision-making. It combines structured EHR data, lab tests, and imagingderived features to aid in diagnosis, risk prediction, and personalized treatment planning. The framework integrates supervised learning models, including Random Forest, Gradient Boosting Machines, Support Vector Machines, and Logistic Regression, along with unsupervised techniques like k-means clustering and Principal Component Analysis. This setup allows for both predictive modelling and patient grouping. We use strong preprocessing, feature selection, and cross-validation to manage missing data, reduce overfitting, and improve generalization. Model performance is measured using metrics such as accuracy, precision, recall, F1-score, and AUC-ROC, with Random Forest and Gradient Boosting showing the best results overall. To meet the important need for transparency, we include SHAP (Shapley Additive explanations) to provide explanations of model outputs, both globally and at the patient level. This helps clinicians understand the main factors behind predictions and builds trust in the system’s recommendations. The proposed framework is made for easy integration into current clinical workflows and allows for near real-time updates as new patient data come in. Overall, this work shows that a clear, multimodal ML framework can improve clinical decision-making and sets the stage for future evaluations and broader use in healthcare.

Ähnliche Arbeiten