Personalized Radiotherapy through Mathematical and AI Models: Advancing Cancer Care from Prediction to Intervention
Please login to view abstract download link
Personalized radiotherapy represents a transformative advancement in cancer care by tailoring treatment regimens to individual patients, optimizing therapeutic outcomes while minimizing treatment-related toxicity. Our research employs advanced mathematical modeling, machine learning, and deep learning techniques to develop patient-specific models that predict treatment outcomes, toxicity, and other clinical responses. These models encompass the entirety of the treatment process, from decision-making to post-treatment follow-up, seeking to enhance treatment efficiency and improve patient quality of life. To achieve this, a mathematical model was constructed to investigate the tumor response in patients undergoing combined regimens of spatially fractionated radiation therapy (SFRT) and conventional treatment. This model identified novel therapeutic strategies for immune activation and enhanced treatment effects. Additionally, an efficient proton-photon patient selection framework was proposed to jointly predict the dose distributions of these two therapies using an attention-gated U-net and compare the normal tissue complication probabilities (NTCP) to ascertain the optimal radiotherapy modality. Furthermore, a patient-specific mathematical model was constructed to estimate the severity of radiation-induced lymphopenia (RIL). This model integrates the dose to circulating lymphocytes and lymph nodes to align with the observed absolute lymphocyte count (ALC) data, which can be utilized to simulate the impact of various radiotherapy regimens on lymphocyte depletion and recovery. Finally, to evaluate the health status based on incomplete patient-reported outcomes (PRO), such as pain levels and sleep discomfort, six advanced machine learning techniques were applied to tackle a multi-class imbalance classification problem across three prevalent cancer types, facilitating more effective health management throughout the course of treatment. These innovations underscore the transformative potential of mathematical and AI models in predicting radiotherapy responses and enabling tailored clinical interventions. By integrating personalized imaging data and patient-specific clinical indicators, adaptive tools are being developed to empower clinicians to dynamically optimize treatment, thereby driving substantial improvements in cancer care.