Room: Davidson Ballroom A
Purpose: The purpose of this exploratory study is to identify CT delta-radiomic features that can be used as indicators of tumor response to chemo-radiation therapy (CRT) for pancreatic cancer.
Methods: Daily non-contrast CTs (28 fractions) acquired using an in-room CT from 31 patients with pancreatic head cancer during routine CT-guided CRT deliveries were analyzed.There are 20 resectable cases with pathology response measured from surgical specimen and 11 non-resectable cases where the contrast CT based treatment responses were analyzed. The pancreatic head was delineated on daily CTs and inspected to ensure consistency. A total of 73 radiomic features were extracted from the segmented regions. Features were evaluated to determine if they changed between the daily CTs (delta radiomics feature, or DRF). Trend information was established by finding the best fit for each DRF versus response. To account for patient and fraction dependent variations, a linear mixed effects model was fitted to the DRFs.To determine if the changes in the feature were significantly associated with response and time since treatment had begun, the p-value of the log-likelihood ratio for each model was calculated and corrected for multiple comparisons using the Benjamini-Hochberg method.A t-test was also performed to determine whether the DRFs changed significantly compared to first fraction. Features that pass both tests were considered as potential features to demonstrate therapy-induced changes.
Results: The results show that 13 DRFs (including: skewness, kurtosis, contrast, complexity, coarseness, texture strength, and NESTD) passed both tests (p-value <0.05). These features exhibited similar trends for resectable and nonresectable cases and demonstrated significant changes from the first fraction to the second to fourth week of treatment.
Conclusion: CT delta-radiomic features obtained from daily CTs have the potential to predict treatment response during CRT delivery.Such DRFs, once verified with large patient datasets, may be used to guide adaptive radiation therapy.
CT, Image-guided Therapy, Modeling