A Fast 3D/3D Registration Method Based on Water Equivalent Path Length for Heavy-Ion Radiotherapy

Abstract

<h3>Purpose/Objective(s)</h3> The accuracy of patient setup needs to be high in heavy-ion radiotherapy. Accordingly, there is an increasing demand for three-dimensional (3D) image-guided systems because of the limitations of the present two-dimensional imaging registration technology. One of the limitations is the inability to directly compare with planning computed tomography (CT) images. Due to this uncertainty in conformation, high dose per fraction radiotherapy is difficult. This study proposes a fast 3D/3D registration method comparing with water equivalent path length (WEPL) that uses planning and treatment CT images to minimize range variations by using the Lucas-Kanade (LK) algorithm. <h3>Materials/Methods</h3> A six-degrees-of-freedom (6DOF) couch correction was simulated for patient positioning at our center by performing automatic rigid image registration (RIR) and using the resultant translation/rotation. Our automatic RIR algorithm was used to align the planning CT and treatment CT images, simulating 6DOF. The alignment values (Δ<i>V</i>), which represents a six-dimensional vector of the amounts of 6DOF couch movement, were calculated by minimizing the following error function: E(<i>ΔV</i>)=Σ <i>_i</i>_∈Ω [<i>T_i</i> (<i>V+ΔV</i>)-<i>P_i</i> (<i>V</i>)]², where <i>V</i> is a six-dimensional vector that denotes a CT position and rotation in a treatment room; <i>T_i</i> (<i>V</i>) and <i>P_i</i> (<i>V</i>) are treatment and planning CT images respectively, whose position in the room is represented by <i>V; i</i> is a 3D vector which denotes a point in the room; and Ω is a set of calculation points, which was assigned a planning target volume (PTV) of Ω in this study. We minimized the error function by using the LK algorithm. To minimize the range variations, each pixel value of the CT images was changed into WEPL values and the alignment values were calculated in the same manner as the above algorithm. The method was evaluated using a dataset containing eight prostate cancer patient images, including a planning CT image with PTV and treatment CT images of four fraction with a fan-beam CT. The range variations were measured by calculating WEPL variations between the planning CT and treatment CT at (1) before registration and after registration based on (2) CT pixel values and (3) WEPL values, respectively. In addition, the calculation time was measured on a workstation with a single GPU. <h3>Results</h3> The mean WEL variation across the 32 cases (eight patients of four fractions) was (1) 1.8 ± 1.16 mm, (2) 1.08 ± 0.52 mm, and (3) 0.22 ± 0.17 mm. The worst WEPL variation were (1) 6.23 mm, (2) 3.06 mm, and (3) 1.17 mm. The mean calculation time were (2) 89 ms and (3) 1066 ms. <h3>Conclusion</h3> We developed a fast 3D/3D RIR method for reducing range variations to realize more accurate heavy-ion radiotherapy. We performed experiments involving eight prostate cancer patients, and our proposed method obtained high accuracy and short computation time.