Optimal threshold for PET-based autocontouring of boost volume for radiotherapy of anal carcinoma
MetadataShow full item record
- Institutt for fysikk 
For patients with anal carcinoma, local tumor control and local recurrence after radiotherapy treatment is a challenge. To improve the local control, a dose escalation could be delivered. Using FDG-PET, the most glucose-avid part of the tumor is considered an attractive boost volume. Nineteen patients with histologically verified anal carcinoma referred to chemoradiotherapy were analyzed. Three oncologists manually delineated the glucose-avid parts of the tumor for each patient. The boost volume was autocontoured within the gross tumor volume based on a percentage threshold of the maximum standardized uptake value (SUV (max)) in the tumor and the average SUV in a volume of interest of 1 cm3 (SUV (peak)). The percentage threshold was varied systematically. PET images were either used as is or filtered using a 6 mm (FWHM) Gaussian. A morphological closing of the boost volume was considered to make the contour similar to a manual delineation. To decide which thresholding procedure gave the best fit, the manual delineations and the autocontours were compared using the dice similarity coefficient (DSC). For unprocessed PET images, the maximum DSC was found at thresholds 15 % and 27.5 % based on SUV (max) and SUV (peak), respectively. At these thresholds, the median DSCs were 0.65 (range 0.18-0.87) and 0.64 (range 0.19-0.87). Applying morphological closing did not change the results dramatically. Gaussian smoothing gave maximum thresholds at 40 % and 47.5 %, respectively, with median DSCs similar to the unprocessed images. As SUV (peak) is considered a more robust parameter, the optimal threshold for autocontouring a boost volume is 27.5 % of SUV (peak). In line with literature, images with Gaussian smoothing gave much larger threhsolds. The thresholding procedure causes a variability close to the interobserver variability in manual delineation. Patients with small tumors are not expected to be candidates for dose escalation, and do not fit the thresholding procedure equally well.