Grant: Travel Awards - Institute Community Support

The mortality of lung cancer is amongst the highest of all cancer types and radiotherapy (RT) is used for more than 50% of lung cancer treatments. A concern for radiation-induced lung disease (RILD) is a major obstacle to increasing radiation prescription for better chance of tumor killing. Indeed patients show large variability in RILD under similar RT plans. There are studies that patient-specific levels of particular biomarkers (proteins and genes) are responsible for such variability, which might explain inadequate accuracy of pre-existing prediction models that are solely based on dosimetry (amount of radiation deposited in body). Therefore, it is important to model RILD as a system where interactions among biological and physical stimuli (radiation) take place. We adopt a system-based approach to build a RILD risk model which will incorporate relevant biological information with traditionally used dosimetric predictors. We will apply and further develop a statistical technique called machine learning to capture the complex relationships among biological and dosimetric variables in our prospective lung cancer patient database. The proposed project can potentially deliver a large impact on radiation oncology and radiobiology. If we can identify patients low RILD risk by using our biophysical prediction model, they might be eligible for a more aggressive RT plan leading to a better chance of cancer cure without significantly affecting iatrogenic morbidity. Our machine-learning based approach might provide a new way to analyze and understand radiation induced normal tissue toxicities, which can be used to design better therapeutic intervention.