PULSAR (personalized, ultra-fractionated stereotactic adaptive radiotherapy) is the adaptation of stereotactic ablative radiotherapy towards personalized cancer management. It has potential to harness the synergy between radiation therapy and immunotherapy, such as immune checkpoint inhibitors to amplify the anti-tumor immune response. For the first time, we applied a transformer-based attention mechanism to investigate the underlying interactions between combined PULSAR and PD-L1 blockade immunotherapy, based on the preliminary experimental results of a murine cancer model (Lewis Lung Carcinoma, LLC). The radiation and administration of α-PD-L1 were viewed as two external stimulation signals occurring in a temporal sequence. Our study demonstrates the utility of a transformer model in 1) predicting tumor changes in response to specific treatment schemes, and 2) generating self-attention and cross-attention maps. The cross-attention maps serve as a biological representation of the semantic similarity between source and target sentences in neural translation, offering insights into the causal relationships of the PULSAR effect. Our model offers a unique perspective with the potential to enhance the understanding of the temporal dependencies of the PULSAR effect on time, dose, and T cell dynamics. In a broader context, our proposed framework offers the potential to explore varying intervals and doses for subsequent treatments while monitoring the biological parameters impacted by these perturbations. This approach can lead to more personalized and rational radiation or drug interactions.
Read full abstract