Tracking lung tumors accurately in real time in a three-dimensional (3D) space is particularly important for image-guided radiation therapy for lung cancer patients. Unless accurate tumor tracking is achieved, not only is it challenging to provide a sufficient dose escalation to the tumor, complications arise by emitting excessive doses to healthy tissue surrounding the tumor [1
]. The main cause of uncertainties in tumor tracking is movement of the internal organs due to the patient’s respiration [5
A number of techniques have been explored to reduce such tumor-tracking uncertainty. External respiratory surrogates are actively being used to indirectly track the tumor location, which include reflective landmarks placed on the abdominal surface of the patient [7
], a strain gauge-based pressure-sensing belt [10
], and optical abdominal surface imaging [12
]. These surrogates are advantageous in that they are non-invasive and require no additional radiation dose to the patients. However, these external signals may not accurately reflect the movement of the tumor within the human body [15
], and the signals may differ depending on how the external signal equipment is installed, thereby degrading the repeatability of the signals [16
]. On the other hand, a method for tracking a fiducial marker [17
], or using an electromagnetic transponder [19
] implanted in or around the tumor, has been developed, which has excellent tracking accuracy owing to its direct tumor tracking. However, patients may feel uncomfortable during such an invasive surgical procedure for implant placement. Several fiducial-less methods also have been proposed and track tumor or respiratory motion successfully; however, the techniques can be applied using a separate dedicated tracking system (e.g., multiple X-ray cameras [21
], Xsight lung tracking system [22
], 4D ultrasound imaging [23
For non-invasive and direct tumor tracking, studies using an on-board kV fluoroscopy system have been reported [6
]. When the tumors shown through the fluoroscopy have an adequately high contrast, such methods can yield good results. However, because it is difficult to obtain high-contrast tumors through fluoroscopic means, these methods have been tested on images acquired from the anterior–posterior direction, where there is relatively little overlap with other organs. These methods, which only work in a specific direction of rotation, are not suitable for rotational cone-beam systems. In addition, because the tumor location is tracked by relying on 2D fluoroscopy, there is no resolution information in the direction of the radiation direction, i.e., the depth direction, making it impossible to trace the full 3D tumor location [6
To track the complete six degrees-of-freedom (DOF) tumor movement, 2D/3D registration using megavoltage (MV) fluoroscopy [5
] or templates containing tumor movements according to the respiratory phase [28
] have been used. An MV fluoroscopy technique has a disadvantage in that separate MV images have to be acquired and synchronized with kV images. In addition, this method has been tested only at certain angles of the gantry, and it does not guarantee a good level of performance with kV images obtained in the left–right direction where the tumor overlaps with other organs. A template-based method should approximate the tumor location at the gantry angles where the tumor is not visible owing to an overlap with other objects using the surrounding gantry angle data where the tumor is visible, which results in a degraded performance at certain gantry angles.
To avoid excessive radiation uptake of healthy tissue, the radiotherapy system should accurately irradiate the tumor to the treatment beam at various angles while rotating around the freely breathing patient. In response to this need, in this study we propose a method for tracking the respiratory phase and 3D tumor position in real time during treatment using a kV image mounted on the treatment system. Using data on six lung tumor patients and a dynamic phantom, the performance of the proposed method was evaluated at 360 degrees around the patient, including the gantry angles at which the contrast of the diaphragm or tumor was low.
Radiation therapy plays an important role in the treatment of patients with lung tumors, and is usually used alone or in combination with chemotherapy or surgery. The purpose of radiation therapy is to deliver radiation to the tumor in the patient’s body. As technology develops further, it will be possible to emit radiation more precisely into the tumor [5
], but such accuracy remains very difficult to achieve.
The method presented in this study has demonstrated the feasibility of tracking the respiratory phase and 3D tumor position in real time during treatment using a kV image mounted on the treatment system. The proposed method works well even under poor conditions in which the shape of the diaphragm is not clear, and regardless of the irradiation angle, that is, the gantry angle, the method shows a good level of performance, i.e., a respiration phase tracking accuracy of (n = 6) = 97.2 ± 2.5%, and tumor tracking error in 3D of (n = 4) = 0.9 ± 0.4 mm. To the best of our knowledge, previous imaging-based tumor tracking methods have been applied only at certain angles (e.g., the anterior–posterior direction) where the tumor is relatively well visible [5
]; however, the proposed method tracked the tumor’s 3D location successfully at any angle of the gantry with sub-millimeter accuracy using a gantry-mounted on-board X-ray imaging system. Other tracking systems such as implanted fiducials [17
] or electromagnetic beacons (i.e., Calypso system) [39
] can track the tumor’s 3D position at any gantry angle because they are not limited by the gantry’s position at any given point in the treatment. However, the patient may become uncomfortable with such an invasive procedure, which also poses a risk of complications clinically, such as the occurrence of a pneumothorax [41
]. A variety of potential advantages are expected to be achieved when high-precision radiation therapy is eventually enabled using this method. In addition to improving the success rate of radiation therapy, the quality of life of the patient can be improved by reducing the risk of complications of radiation therapy. Radiation therapies, which are difficult to perform owing to the presence of important structures near the tumor, including the spinal cord, may become applicable if accurate targeting is made possible.
The performance of an elevation-based method (40.3 ± 17.0%) is much lower than that of the method proposed herein (97.2 ± 2.5%). Given the fact that the inhale/exhale identification accuracy (48.3%) is close to the accuracy achieved through a diaphragm elevation-based method (40.3%), most identification errors are due to a failure of inhale/exhale identification. The amplitude of the breathing can be analyzed well based on high- and low-diaphragmatic information, but this method alone fails to identify whether the current breathing is in an inspiratory or expiratory state. As shown in Figure 4
, the 3D trajectory of the tumor showed hysteresis, as evidenced through previous studies [44
], and thus inhale/exhale identification must be conducted to track the tumor in real time while the patient is breathing.
The proposed method identifies the respiratory phase through two steps: (1) a structural-similarity based ranking and (2) comparing the hemidiaphragm area. Some may feel that step (1) can be skipped. However, step (1) is necessary to determine whether the current breath is in an inspiratory or expiratory state, thereby determining the respiratory phase candidates to be applied during the second step. Similar to the problems of the elevation-based method mentioned above, a low performance is shown when step (1) is not applied. The elevation-based method has roughly a 50% accuracy when distinguishing between inhale and exhale phase as it only has one image at its disposal, and mid-inhale and mid-exhale phases can look very similar. Moreover, the proposed method performs robustly for projections of every 10% segment of the respiratory cycle. Given that the SSIM is sensitive to slight movements of the diaphragm phantom, as shown in Figure A1
, treatment projections with respiratory phases of finer than 10% should be identifiable.
Several major limitations of this study should be addressed. First, we did not consider changes in the tumor position that occur from the heartbeat separately from those occurring from respiration. However, when the tumor was located in the upper part of the lung, the movement of the tumor was slight (patients 5 and 6), whereas a tumor close to the diaphragm showed a large movement (patients 1 through 4). Moreover, the movement of the tumor was mainly seen in the cranial–caudal direction (Figure 4
]), which is the same as the motion pattern of the diaphragm. Based on these observations, most movements of the tumor can be explained by the motion of the diaphragm alone. Second, owing to the problem of the radiation dose administered to the patient, CBCT projections cannot be obtained by moving around the patient in all directions, and thus pseudo projections were created for every 10% segment of the respiratory phase instead. However, in the phantom study, the SSIM between the actual CBCT projections and their corresponding pseudo projections was close to 1 (0.9589 ± 0.0233). Thus, it can be said that our pseudo projection is a good surrogate for an actual projection. In addition, because a pseudo image has a lower contrast than an actual projection, the diaphragm extraction on a pseudo projection is relatively difficult to detect. Because the proposed method was evaluated under worse conditions than those found in an actual situation, the use of an actual image is not expected to lead to a deterioration in the performance. Lastly, because kV imaging is needed in addition to treatment beam, the patient is exposed to an extra radiation dose. Patients with non-small-cell lung cancer are treated with a high dose of within 63 to 103 Gy during radiation therapy [47
], compared to which, the amount of radiation delivered by the additional kV images is quite small. Thus, if a small increase in dose can reduce the therapeutic beam exposure to healthy tissue around the tumor, it will be of benefit to the patient.