Click here to


Are you sure ?

Yes, do it No, cancel

Synthetic Fluoroscopic Image Generation for Tracking Accuracy Validation of Marker-Less Tumor Tracking in Radiotherapy

K Miyazaki12*, T Fujii2, T Umekawa2, N Miyamoto1, K Umegaki1, (1) Hokkaido University, Sapporo-shi, Hokkaido, JP, (2) Hitachi, Ltd., Hitachi-shi, Ibaraki, JP


(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

The tracking accuracy validation of marker-less tumor tracking is not established well because it is hard to define the ground truth of the tumor position in the patient’s fluoroscopic images accurately. The phantom based validation tends to be easy to track because there is no inner body motion. Therefore, we propose a new validation scheme that uses synthetic fluoroscopic images of a patient and a 3D printed tumor. The synthetic fluoroscopic images have both inner body motion and the ground truth of the 3D printed tumor position. The purpose of this study is to establish a synthetic method of separately taken two objects’ fluoroscopic images.

The 3D printed tumor model was selected from one of lung radiotherapy patients treated at Hokkaido University Hospital, whose tumor size was approximately average. We had phantom experiments to take fluoroscopic images at 3 different conditions; (a) only 3D printed tumor, (b) only lung phantom, and (c) both 3D printed tumor and lung phantom. Our goal is to minimize the differences between the image taken at (c) and the synthetic image of (a) and (b). We tried two kinds of synthetic methods, (1) a simple synthetic method (conventional) and (2) a beam hardening considered synthetic method (proposed). To evaluate these methods, we calculated MSE, PSNR and SSIM between the image taken at (c) and the synthetic image.

MSE, PSNR and SSIM were 1637, 27.7, and 0.891 for the conventional method and 327, 28.9, 0.940 for the proposed method. By considering beam hardening effect, the contrast of the 3D printed tumor in the synthetic image was well reproduced.

We evaluated two synthetic methods of separately taken two objects’ fluoroscopic images. We believe it is important to generate realistic synthetic images for the reliable validation of marker-less tumor tracking.

Funding Support, Disclosures, and Conflict of Interest: Koichi Miyazaki, Takaaki Fujii and Toru Umekawa are employees of Hitachi, Ltd., Tokyo, Japan.


Target Localization, Image Fusion, Validation


IM- X-Ray: Image processing

Contact Email