Room: Room 202
Purpose: The aim of this study was to build a system to generate virtual X-ray images of the patient at the time of radiotherapy, based on skinned mesh animation using a range imaging sensor and treatment planning computed tomography (CT) images.
Methods: The CT images were converted to three-dimensional (3D) models by the marching-cube method from contours of the body structures. The skeleton data acquired by a range imaging sensor at the same time as CT imaging was aligned with the 3D models based on metal marker positions on the CT images and the ones on the range imaging sensor. Then, the 3D models were moved by skinned mesh animation from the patient position at the time of radiotherapy, and virtual X-ray images were generated by projecting the moving 3D models. In this study, KINECTv2 sensor was used as the range imaging sensor to measure the distance to the phantom surface. The system was evaluated in terms of reproducibility of the skeleton data, alignment accuracy between the 3D models and the skeleton data, and the amount of the movement between the phantom and the skeleton data.
Results: Average reproducibility was 4.8 mm for all joints of the skeleton data, and a change exceeding 10 mm was also observed. Average alignment accuracy was 4.8 mm for all markers. This error was caused by a detection error in the depth direction of KINECTv2 sensor. The difference of the amount of movement of both the skeleton data and the actual phantom was 10 mm on average. The difference might be caused by reproducibility and alignment accuracy.
Conclusion: The present system enabled to generate virtual X-ray images for the purpose of accurate and non-invasive radiotherapy. Improvement of reproducibility of the skeleton data and detection accuracy of the depth image must be the future work.