ESTRO 2021 Abstract Book

S269

ESTRO 2021

Conclusion We developed a method that highlights regions where anatomical changes are particularly large and are not accurately modelled using PCA based on only scans from the first week of treatment. The oropharyngeal region was particularly challenging for PCA models in H&N. OC-0364 A statistical method for reducing systematic errors in the presence of organ at risk deformations Ø. Rørtveit 1,3 , L.B. Hysing 1,3 , A.S. Stordal 2,4 , S. Pilskog 1,3 1 Haukeland University Hospital, Dept. of cancer treatment and medical physics, Bergen, Norway; 2 NORCE, Energy, Bergen, Norway; 3 University of Bergen, IFT, Bergen, Norway; 4 University of Bergen, Dept. of Mathematics, Bergen, Norway Purpose or Objective Organ at risk (OAR) deformation is challenging for radiotherapy (RT), as deformations between planning and treatment can cause systematic errors in position, shape and dose of the OAR. Knowledge about organ shapes from previous patients can potentially be exploited to improve prediction of the shape changes occurring throughout the course of RT. We focus on the rectum, an important OAR in prostate and cervical RT. Statistical shrinkage estimation is introduced as a method to combine a population mean rectum shape obtained from previously treated patients with the planning CT (pCT) rectum shape of a new patient. Our aim is to investigate if this method leads to reduced systematic errors caused by rectal deformations. Materials and Methods The rectum shapes of 37 prostate cancer patients were delineated on the pCT and on each of their 7-10 repeat CTs taken during the course of RT. Deformable registration was performed both within and between patients to enable computation of a population mean rectum accounting for deformations. Shrinkage estimates were found as weighted linear combinations of the pCT rectum and the population mean shape. The weight given to the population mean shape, the shrinkage factor , was optimized on training data. The method was trained and evaluated using leave-one-out cross validation. Similarity between the shrinkage estimated shapes and the individual patient mean shape over the repeat CTs was evaluated using the Dice similarity index (DSI). Dosimetric analysis was performed with clinical VMAT simultaneous integrated boost for prostate cancer, including treatment to the prostate, seminal vesicles and the pelvic lymph nodes (50-67.5 Gy/25 fr). DVHs and two dosimetric parameters correlated with increased rectal toxicity – generalized equivalent uniform dose (gEUD) and the minimum dose to the hottest 5% volume (D5%) – were compared for the shrinkage estimate and the pCT. Results Figure 1a shows the similarity (DSI) to the patient mean shape for the pCT shape versus the shrinkage estimated shape for all patients. Shrinkage estimation resulted in moderate but significant improvement in similarity to the patients mean rectum as compared to the pCT. The median DSI increased from 0.73 for the