Explainable prediction of recurrence after prostate cancer radiotherapy using in silico digital twin model and machine learning

Biochemical recurrence (BCR) for prostate cancer (PCa) patients treated with External Beam Radiation Therapy (RT) has an incidence rate of up to 20 %. Thus, predicting BCR after PCa RT appears crucial for personalising treatments. Current approaches, such as radiomics and deep learning, applied to clinical and in vivo imaging data, suffer from limited explainability. This paper introduces a pipeline for predicting BCR by integrating clinical data with biologically grounded features derived from in silico digital twin simulations, supported by two explainability analyses. Specifically, we leverage a previously developed in silico digital twin model to simulate tumour growth and response to radiation for 315 PCa patients retrospectively treated with RT. A logistic regression model was identified as the best predictor, integrating clinical characteristics and biologically interpretable features extracted from simulations (AUC =

" role="presentation" style="box-sizing: inherit; display: inline-block; line-height: normal; font-size-adjust: none; word-spacing: normal; overflow-wrap: normal; text-wrap-mode: nowrap; float: none; direction: ltr; max-width: none; max-height: none; min-width: 0px; min-height: 0px; border: 0px; padding: 0px; margin: 0px; scroll-margin-top: 74px; position: relative;">

 0.73). To enhance explainability, a local perturbation analysis is performed to quantify the influence of individual radiobiological parameters within the in silico model. Additionally, SHapley Additive exPlanations (SHAP) were applied to evaluate the contribution of each feature to the BCR prediction. By linking simulation-driven parameter importance with feature-level explanations, the pipeline provides coherent insights at the mechanistic and statistical levels.