Room: Track 4
Purpose: To validate a machine learning (ML) algorithm predicting the use of intracavitary (IC) versus hybrid interstitial (HIS) applicators for high-dose-rate cervical brachytherapy by assessing agreement between clinically used and ML predicted applicators in a replanning study.
Methods: A dataset of 233 fractions using IC or HIS applicators was employed to train and validate a supervised classification ML algorithm. The train/test process was repeated for 1,000 random selections of training/testing data. Fractions where the ML prediction and clinically used applicator disagreed in >5% of 1,000 iterations were selected for replanning using the ML predicted applicator. To replan HIS to IC, needle dwell positions were zeroed. To replan IC to HIS, needles were digitized parallel to the tandem in positions where there was significant spread of the high-risk clinical target volume (HR-CTV). Replan quality was assessed based on EQD2 changes of dose-volume histogram (DVH) metrics for the HR-CTV and organs-at-risk (OARs). A difference (replan – original) of 50cGy/fraction was considered a clinically significant change.
Results: 11 IC and 10 HIS fractions met the criteria for replanning. For the 11 IC to HIS replans, the mean HR-CTV D90 differences (22.7cGy) were not significant, but mean OAR differences for bladder (-118.1cGy), rectum (-68.0cGy), and sigmoid D2cc (-50.2cGy) were. For the 10 HIS to IC plans, differences in mean HR-CTV D90 (-34.0cGy), bladder (20.9cGy), rectum (37.6cGy), and sigmoid D2cc (16.8cGy) were not significant.
Conclusion: Misidentified IC plans benefited from HIS needles for OAR dose reduction while maintaining comparable HR-CTV D90, indicating an underuse of needles in some cases. Misidentified HIS plans saw no significant DVH metric changes. For these plans, the IC applicator could have delivered a comparable tumor dose with a less invasive procedure. This establishes ML’s ability to more accurately identify the use of IC versus HIS applicators, validating the algorithm’s performance.
Intracavitary Brachytherapy, Machine Learning, Model Validation