Statistical deformation reconstruction using multi-organ shape features for pancreatic cancer localization

Respiratory motion and the associated deformations of abdominal organs tumors are essential information in clinical applications. However, inter- intra-patient multi-organ complex have not been statistically formulated, whereas single organ widely studied. In this paper, we introduce a deformation library its application to reconstruction based on shape features multiple organs. Statistical motion/deformation models stomach, liver, left right kidneys, duodenum were generated by matching their region labels defined four-dimensional computed tomography images. A total 250 volumes measured from 25 pancreatic cancer patients. This paper also proposes per-region-based learning using reproducing kernel predict displacement for adaptive radiotherapy. The experimental results show that proposed concept estimates better than general per-patient-based achieves clinically acceptable estimation error with mean distance 1.2 $\pm$ 0.7 mm Hausdorff 4.2 2.3 throughout respiratory motion.