Encrypted login | home

Program Information

Real-Time Automatic Fiducial Marker Detection in Low Contrast Cine-MV Images

W Liu

W Liu1*, W Lin2, M Ahmad1, R Nath1, (1) Yale Univ. School of Medicine, New Haven, CT, USA (2) National Chung Cheng University, Min-Hsiung, Taiwan, ROC

TU-E-BRA-4 Tuesday 2:00:00 PM - 3:50:00 PM Room: Ballroom A

Purpose: Intrafraction motion tracking using beam-line MV images have gained much attention because no additional imaging dose is introduced. Since MV images have much lower contrast than kV images, a robust marker detection algorithm is a pre-requisite. In this work, we develop a novel, fast, and robust method to detect implanted markers in low-contrast cine-MV patient images.

Methods: Several marker detection methods have been proposed in the recent years. These methods are all based on template matching or its derivatives. Template matching needs to match object shape that changes significantly for different implantation and projection angle. While these methods require a large number of templates to cover the different situations, they are often forced to use a smaller number of templates to reduce the computation load because their methods all require exhaustive search in the ROI. We solve this problem by synergetic use of modern but well-tested computer vision and AI techniques - detect implanted markers utilizing discriminant analysis for initialization and mean-shift feature space analysis for sequential tracking. This novel approach avoids exhaustive search by exploiting the temporal correlation between consecutive frames and makes it possible to perform more sophisticated detection at the beginning to improve the accuracy, followed by ultrafast sequential tracking after the initialization. The method was evaluated using 1149 cine-MV images from 2 prostate IMRT patients and compared with manual marker detection results from 6 researchers. The average of the manual detection results is considered as the ground truth.

Results: The average RMS errors of the automatic tracking from the ground truth are 1.9 and 2.1 pixels for the 2 patients (0.26mm/pixel). The standard deviations of the results from the 6 researchers are 2.3 and 2.6 pixels.

Conclusion: The proposed method can achieve similar marker detection accuracy to manual detection in low-contract cine-MV images.

Contact Email