One scale to rule them all: interpretable multi-scale Deep Learning for predicting cell survival after proton and...

The relationship between the physical characteristics of the radiation field and biological damage is central to both radiotherapy and radioprotection, yet the link between spatial scales of energy deposition and biological effects remains not entirely understood.

To address this, we developed an interpretable Deep Learning (DL) model that predicts cell survival after proton and carbon ion irradiation, leveraging sequential attention to highlight relevant features and provide insight into the contributions of different energy deposition scales. Trained and tested on the Particle Irradiation Data Ensemble (PIDE) dataset, our model incorporates, in addition to linear energy transfer (LET), nanodosimetric and microdosimetric quantities simulated with MC-Startrack and Open-TOPAS, enabling multi-scale characterization.

While achieving high predictive accuracy, our approach also emphasizes transparency in decision-making. We demonstrate high accuracy in predicting relative biological effectiveness (RBE) for in vitro experiments. Multiple scales are used concurrently, with no single spatial scale predominating. Quantities defined at smaller spatial domains generally have a greater influence, whereas the LET plays a lesser role.