Program Information
Pixel-Based Analysis of Stomach Dose-Surface Maps to Improve Toxicity Predictions in Gastrointestinal Radiotherapy
A Witztum*, B George , F Van den Heuvel , M Hawkins , CRUK/MRC Oxford Institute for Radiation Oncology, University of Oxford, Oxford, Oxford
Presentations
SU-K-FS2-13 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: Four Seasons 2
Purpose: To utilize dose-surface maps (DSMs) to find spatial dose-surface metrics that improve upper-GI toxicity predictions.
Methods: Stomach DSMs were produced for patients from two clinical trials treating pancreatic cancer (Cohort50, n=62; Cohort60, n=19) with follow up information available for RTOG defined upper-GI toxicity. Dose-volume metrics such as V15-55 Gy (in 5 Gy increments), and mean dose were extracted alongside equivalent dose-surface metrics (mean surface dose, SA15-55 Gy). The significance and predictive power of each metric was assessed with a Mann-Whitney test and the area under the ROC curve (AUC). To make use of the spatial information within the DSMs, a permutation test was performed on each DSM pixel to find regions significant in predicting toxicity. For each pixel, the true Kendall tau-b coefficient is calculated from the dose and toxicity grade. The grades are permuted 10,000 times and the coefficient calculated in each permutation. An uncorrected p-value (uncorrected for multiple comparisons) is calculated from the distribution of the coefficients. The mean dose in the significant region (uncorrected p<0.2) was extracted and the AUC compared to that of dose-volume and dose-surface metrics.
Results: V35 Gy (AUC = 0.80), SA35 Gy (AUC = 0.81), SA45 Gy (AUC = 0.82), and mean surface dose (AUC = 0.82) were significantly different in groups with and without grade ≥ 2 toxicity in Cohort60 but no volume or surface metric was significant in Cohort50. Mean dose in significant region (Dmean,SR) from the permutation test was significantly different in groups with and without grade ≥ 2 toxicity in Cohort60 (AUC = 0.88) and Cohort50 (AUC = 0.73).
Conclusion: Dmean,SR is a new metric that considers both surface and spatial information. In two cohorts presented here, it offered improved predictive power over dose-volume and dose-surface metrics in one, and was the only significant predictor in the other.
Contact Email: