An organ deformation model using Bayesian inference to combine population and patient-specific data.

Objective.Organ deformation models have the potential to improve delivery and reduce toxicity of radiotherapy, but existing data-driven motion models are based on either patient-specific or population data. We propose to combine population and patient-specific data using a Bayesian framework. Our goal is to accurately predict individual motion patterns while using fewer scans than previous models.Approach.We have derived and evaluated two Bayesian deformation models. The models were applied retrospectively to the rectal wall from a cohort of prostate cancer patients. These patients had repeat CT scans evenly acquired throughout radiotherapy. Each model was used to create coverage probability matrices (CPMs). The spatial correlations between these estimated CPMs and the ground truth, derived from independent scans of the same patient, were calculated.Main results.Spatial correlation with ground truth were significantly higher for the Bayesian deformation models than both patient-specific and population-derived models with 1, 2 or 3 patient-specific scans as input. Statistical motion simulations indicate that this result will also hold for more than 3 scans.Significance.The improvement over previous models means that fewer scans per patient are needed to achieve accurate deformation predictions. The models have applications in robust radiotherapy planning and evaluation, among others.

Reducing systematic errors due to deformation of organs at risk in radiotherapy.

PURPOSE: In radiotherapy (RT), the planning CT (pCT) is commonly used to plan the full RT-course. Due to organ deformation and motion, the organ shapes seen at the pCT will not be identical to their shapes during RT. Any difference between the pCT organ shape and the organ's mean shape during RT will cause systematic errors. We propose to use statistical shrinkage estimation to reduce this error using only the pCT and the population mean shape computed from training data. METHODS: The method was evaluated for the rectum in a cohort of 37 prostate cancer patients that had a pCT and 7-10 treatment CTs with rectum delineations. Deformable registration was performed both within-patient and between patients, resulting in point-to-point correspondence between all rectum shapes, which enabled us to compute a population mean rectum. Shrinkage estimates were found by combining the pCTs linearly with the population mean. The method was trained and evaluated using leave-one-out cross validation. The shrinkage estimates and the patient mean shapes were compared geometrically using the Dice similarity index (DSI), Hausdorff distance (HD), and bidirectional local distance. Clinical dose/volume histograms, equivalent uniform dose (EUD) and minimum dose to the hottest 5% volume (D5%) were compared for the shrinkage estimate and the pCT. RESULTS: The method resulted in moderate but statistically significant increase in similarity to the patient mean shape over the pCT. On average, the HD was reduced from 15.6 to 13.4 mm, while the DSI was increased from 0.74 to 0.78. Significant reduction in the bias of volume estimates was found in the DVH-range of 52.5-65 Gy, where the bias was reduced from -1.3 to -0.2 percentage points, but no significant improvement was found in EUD or D5%, CONCLUSIONS: The results suggest that shrinkage estimation can reduce systematic errors due to organ deformations in RT. The method has potential to increase the accuracy in RT of deformable organs and can improve motion modeling.