Top Line: Is anything more dreaded on oral exams than contouring an advanced nasopharyngeal case
The Study: In practice, as this manuscript puts it, “the current manual gross tumor volume (GTV) delineation is time-consuming, laborious and error-prone.” So this Chinese group set out to create a workable auto-contouring platform for…dun dun dun…nasopharyngeal target volumes. Pretreatment MRIs from 258 patients were used as a training cohort (n=180) to create the model, a validation cohort (n=20) to optimize the model, and finally a testing cohort (n=58) to assess performance. To test the performance, nine multicenter “experts” (with over 10 years experience treating nasopharyngeal cancer) were asked to edit the auto-generated contours to their liking, and there was a low degree of deviation per the Dice similarity coefficient (a way to quantify volume overlap). So now that we know the auto-contours are good, how much can they help regular rad onc folk? Six “junior” rad oncs were asked to contour with and without the help of auto-contours, and their volumes were compared to predetermined “true” volumes. With the help of autocountors, the mean degree of revisions was reduced from 26% to 8% for gross primary tumor and from 21% to 14% for gross nodes.
TBL: Auto-contouring nasopharyngeal cancer radiation volumes can improve accuracy of target delineation for more novice physicians and reduce time burden for all physicians. | Liao, Int J Radiat Oncol Biol Phys 2022