A key-point based real-time tracking of lung tumor in x-ray image sequence by using difference of Gaussians filtering and optical flow

Abstract

In radiation therapy, for accurate radiation dose delivery to a target tumor and reduction of the extra exposure of normal tissues, real-time tumor tracking is typically an important technique in lung cancer treatment since lung tumors move with patients' respiration. To observe a tumor motion in real time, x-ray fluoroscopic devices can be employed, and various tracking techniques have been proposed to track tumors. However, development of a fast and accurate tracking method for clinical use is still a challenging task since the obscured image of the tumor can cause decreased tracking accuracy and can result in additional processing time for remedying the accuracy. In this study, a new key-point-based tumor tracking method, which is sufficiently fast and accurate, is presented. Given an x-ray image sequence, the proposed method employs a difference-of-Gaussians filtering technique to detect key points in the tumor region of the first frame which are robust against noise and outliers in the subsequent frames. In the subsequent frames, these key points are tracked using a fast optical flow technique, and tumor motion is estimated via their movement. To evaluate the performance, the proposed method has been tested on several clinical kV and MV x-ray image sequences. The experimental results showed that the average of the root mean square errors of tracking were [Formula: see text] and [Formula: see text] for kV and MV x-ray image sequences, respectively. This tracking performance was more accurate than previous tracking methods. In addition, the average processing times for each frame were [Formula: see text] and [Formula: see text] for kV and MV image sequences, respectively, and the proposed method was faster than previous methods as well as shorter than frame acquisition interval. Therefore, the proposed method has the potential for both highly accurate and fast tumor tracking in clinical applications.

Simultaneous object detection and segmentation for patient‐specific markerless lung tumor tracking in simulated radiographs with deep learning

“…Besides DL-based approaches, numerous other studies have investigated non-DL-based image tracking techniques for markerless lung tumor tracking, including well-established methods such as image registration, 40 template matching, 41 and optical flow 42,43 . Other less common methods like short arc tumor tracking 44 and hidden Markov model 45 have also been proposed.Some studies were exclusively conducted on digital or experimental phantoms with tumor motions that were mechanically controlled following either simulated breathing patterns 42 or patient-measured motion traces.…”

“…45,46 They were often able to achieve sub-millimeter accuracy, which might potentially be biased by the often simpler geometry of phantom anatomy and tumor shape/size. Several other studies like Rozario et al, 40 Bruin et al, 41 Shieh et al 44 and Ichiji et al 43 included retrospective studies on real patients' data such as beam's eye view (BEV) images, CBCT projection images, or clinical x-ray image sequences. Rozario et al 40 tested their image registration-based method on over 5000 frames of MV BEV images of 5 patients and reported rather unstable performance, with tumors' average 2D position deviations at 180 degrees gantry angle for a single fraction ranging from 4.6 mm up to 7.9 mm (6 fractions from 3 patients).…”

“…65% tumors were deemed successfully tracked by judging the correspondence of the predicted longitudinal tumor trajectories manually overlaid with the motion of an external Real-time Position Management (RPM) marker. Ichiji et al 43 proposed a new method that used the optical flow technique to track key points on the tumor, and tested on clinical X-ray image sequences. They achieved an average root mean square error of 2.46 mm for kV X-ray images and 1.53 mm for MV X-ray images, and discovered that the accuracy of their method linearly decreased with the motion range of the target tumor, which has not been observed in our approach.…”

“…Besides DL‐based approaches, numerous other studies have investigated non‐DL‐based image tracking techniques for markerless lung tumor tracking, including well‐established methods such as image registration, 40 template matching, 41 and optical flow 42,43 . Other less common methods like short arc tumor tracking 44 and hidden Markov model 45 have also been proposed.…”

Vector-field dynamic x-ray (VF-DXR) using optical flow method in patients with chronic obstructive pulmonary disease

“…OFM can visualize motion velocity with higher spatial resolution than the cross-correlation method [ 27 ]. Ichiji et al [ 28 ] first applied OFM to sequential x-ray images to track the real-time motion of lung tumors, although the lung parenchymal motion itself was not studied in the paper. Recently, the feasibility of VF-DXR with OFM for lung parenchyma motion analysis was reported [ 21 ].…”