ESTRO 2021 Abstract Book

S146

ESTRO 2021

radiologist manually delineated GTV and VTV (aided by T2W, DWI, and ADC map). Automatic VTV segmentation using the tool was performed with manual input given as 1) GTV and 2) a ROI defined by a non- radiologist. ADC change between baseline and week 2 was calculated for both the manual and the semi-automatic delineation method, and their correlation was measured using Pearsons correlation coefficient. Image related ADC uncertainty was measured using test-retest data. Bootstrap of the observed non-normal distribution was used to establish the central 70% confidence interval.

Results The temporal ADC change between baseline and week 2 measured by the manual and the semi-automatic delineation method is presented in Figure 2; error bars indicate test-retest image variance (+/- 92.8 mm 2 /s). The Pearson correlation coefficient between manual and semi-automatic VTV delineation was 0.68. Between the two manual inputs (GTV vs. non-radiologist ROI), the correlation was 0.79. No association between ADC changes and ADC values were observed.

Conclusion Longitudinal ADC changes were larger than image related uncertainty, and thus potentially reflect treatment related biological changes. The presented semi-automatic segmentation for ADC calculation was not sensitive to manual input, and correlates well with manual delineation by a radiologist. The segmentation method may be useful in other targets than rectal cancer and may be well-matched for the MR-linac workflow. References