Room: AAPM ePoster Library
Purpose: This work explores the relationship between voxel-wise dose to the heart and overall survival (OS) in the RTOG0617 trial to identify critical anatomical structures.
Methods: Dose distributions for the 398 patients in the RTOG-0617 trial were mapped to a randomly selected reference patient. Scans were deformably registered to the reference and their dose distributions were deformed accordingly using the multi-step b-splines algorithm in Plastimatch software. Only scans with dice scores above 0.85 between the reference and the deformed heart structures were used. Dataset was split into 70/30 train/test sets. Cox proportional hazards model was fit for the physical dose to each voxel within the heart on 10 bootstraps of the training set. The resulting p-values were then averaged from bootstraps. Critical regions showing significant association with OS were identified by thresholding -log10(p-values) at 50% of maximum. Kaplan-Meier analysis was performed on the test set by splitting data on the mean dose to the identified regions.
Results: Two patients were randomly selected as references for dose mapping. For both the reference patients, regions showing significant association with the OS consisted of parts where Aorta, Ventricles (right and left) and Pulmonary Artery meet. This region encompasses the four valves. Kaplan-Meier analysis showed a significant difference between the top and the bottom third riskiest groups based on the mean dose to the region identified to be critical. P-values from the log-rank tests for the two reference patients were 0.0246 and 0.0009 respectively.
Conclusion: Dose mapping was used to identify regions within Heart showing significant association with OS in the RTOG-0617 trial. The analysis presented is exploratory and can further be used to generate hypotheses based on heart function. Such analysis might uncover previously unidentified regions within Heart that contribute to OS and to build dose constraints for radiotherapy treatment planning.
Not Applicable / None Entered.
Not Applicable / None Entered.