Digital Twin of Cancer Patients Using Multimodal Data for Personalized Treatment and Monitoring

Abstract

Recent advances in healthcare data collection have made it possible to gather various multimodal datasets. These include clinical records, molecular profiles, physiological sensors, posture measurements, and environmental factors. Traditional methods mainly depend on static analysis techniques like PCA and t-SNE, applied to offline datasets. This limits their use in real-time monitoring and adaptable interventions. In this study, we introduce a new framework for dynamic Human Digital Twin (HDT) modeling, aimed at personalized cancer care and ergonomic risk assessment. Our framework uses effective feature selection methods such as Chi-Square tests, PCA, t-SNE, and linear discriminant analysis (LDA), specifically designed for streaming multimodal physiological data. We utilize surrogate learning models to predict patient or operator performance by incorporating physiology, posture, and environmental signals. These models support reinforcement learning (RL) agents, which optimize treatment strategies or task assignments in real-time. We improve explainability through SHAP and LIME frameworks, alongside interactive dashboards and AR/VR-style user interfaces. This provides clear insights into workload, performance, and treatment outcomes. Our performance metrics go beyond usual accuracy and regression scores, including uncertainty quantification, calibration, and time-series metrics like dynamic time warping (DTW). Continuous integration of multimodal data from wearable and biosensor streams allows for real-time monitoring and responsive actions within the digital twin environment. This scalable and understandable HDT platform connects past clinical analysis with real-time sensor data. It shows promise for enhancing personalized healthcare delivery, precision cancer management, and ergonomic risk reduction. The proposed system offers a practical and adaptable solution for next-generation smart healthcare applications.