Purpose: To develop a framework to automate the analysis of large cohorts of patients regarding the correlation of toxicities and patterns of failure with various dosimetric and geometric parameters.
Methods: A software package has been built to process data from patients with radiation-induced toxicities or recurrence after proton and photon radiation therapy at our institution. The package is written in C++ and allows for flexibility in processing large cohorts of patients and exploring a large parameter space. It takes as input CT, RTPLAN, RTDOSE, REG and RTSTRUCT dicom files from the TPS, as well as Monte Carlo-generated LET and dose distributions. Correlation metrics include: DVH metrics, Dice Coefficient, Jaccard Index, Hausdorff and centroid distances and volume of overlap between the toxicity structure and iso-dose and iso-LET levels. Toxicity structures and recurrence volumes were contoured by physicians.
Results: The software is used to analyze a cohort of brain patients. The metrics were computed within the TPS and compared with our calculated values to check for correctness. The code is found to be robust and can be used to analyze a large cohort of patients efficiently. For proton patients, it also facilitates the evaluation of biological dose models, allowing us to explore the relationship between their predictions with physician-contoured radiation-induced injury regions.
Conclusion: Understanding the correlation between radiation-induced toxicities and dosimetric quantities within the framework of various biologic dose models is of great interest. As a first step, we built and tested an in-house framework for automating the calculation of relevant metrics for a large cohort of patients.
Computer Software, Dose Response, LET
TH- Response Assessment: Modeling: other than machine learning