Purpose: To analyze institutional performance irradiating IROC Houston’s SRS head phantom.
Methods: Irradiation results of 1072 SRS head phantoms between 2012-2020 were abstracted. Univariate analysis and random forest models were used to associate irradiation conditions with phantom results. Phantom results included pass/fail classification, average TLD ratio, and percent of pixels passing gamma. The following categories were evaluated in terms of how they predicted these outcomes: irradiation year, TPS algorithm, machine model, energy, and field size. Specifically, algorithm was categorized per TG-329, collimators were divided into three categories: cones, high-definition (MLC Leaf Width < 0.5 cm), and low-definition (MLC Leaf Width > 0.5 cm), and field size was calculated from the average extent of the measured 50% isodose line.
Results: Pearson chi-squared test showed that algorithm (p < 0.001) and collimator (p = 0.24) were statistically significant predictors of pass/fail. Per Tukey’s post-hoc test, pencil beam algorithm (AAA: p =0.01, GBBS: p = 0.15, Monte Carlo: p = 0.001, Measured: p < 0.001, and Superposition/Convolution: p = 0.015) and cone collimator (high-definition: p = 0.035, low-definition: p = 0.055) were both more likely to be associated with failing phantom results. Our random forest models were able to achieve an overall prediction accuracy of 90.9 ± 0.3%. For overall pass/fail, TLD result, and percent of pixels passing, the field size was the most important predictor, with poorer results associated with smaller fields. The next three most important variables were irradiation year, machine, and algorithm; however, the specific order of importance varied.
Conclusion: Field size is the most important factor in determining the outcome of the phantom. Specifically, smaller fields (≤ 3 cm), which encapsulates the majority of cone collimators (72.7%) used, tends to produce more failures. Pencil beam algorithms were inferior to accurately predicting dose as compared to all other algorithms.
Funding Support, Disclosures, and Conflict of Interest: This work received funding from the NIH/NCI grants #CA180803 and #CA214526.