Regression Model of Tumor and Diaphragm Position for Marker-Less Tumor Tracking in Carbon Ion Scanning Therapy for Hepatocellular Carcinoma

Abstract

In radiation therapy (RT) of esophageal cancer, CTV to PTV margins are generally isotropic and equal for all patients. However, detailed knowledge of the position variability and tumor motion caused by respiratory motion is lacking. The purpose of this study was to accurately quantify esophageal tumor position variability and respiratory motion and investigate possible surrogate structures for image guidance. Materials/Methods: The first 12 patients enrolled in a prospective cohort study (NCT02139488) were analyzed. Patients were treated with chemo-RT with a radiation dose of 23 to 28Â1.8 Gy combined with weekly carboplatin/ paclitaxel and daily 4D CBCT scans. As soft tissue contrast in CBCT is limited, gold fiducial markers (0.35 x 5 mm) were inserted during endoscopic ultrasonography preferably at the proximal border, in the middle and at the distal border of the tumor before the planning CT was made. The following regions of interest (ROI) were registered for each fraction: bony anatomy (vertebrae), carina, diaphragm, and fiducial markers using a rectangular ROI, and gross tumor volume (GTV), using a shaped ROI ("mask"). These different surrogates for setup and their implication on margins were calculated with statistics of the average residual marker displacement when using the different regions as reference (standard deviations of random [s] and systematic [ P ] errors). Breathing amplitudes also influence margins; therefore, their distribution within the cohort was determined by the fiducial ROI. Subsequently, a planning target volume margin including the average respiratory motion was determined for these scenarios according to the nonlinear van Herk formula. Since esophageal tumors border both to lung and mediastinal tissue, the parameters for this formula were conservatively chosen to be representative for water. Results: A median of 3 fiducials was placed in 12 tumors located at the mid esophagus (nZ3), lower esophagus (nZ4), and gastroesophageal junction (nZ5). The median (range) peak-to-peak respiratory tumor motion amplitude in the left-right (LR), craniocaudal (CC), and anteroposterior (AP) directions was 0.15 (0.07-0.73), 0.63 (0.39-0.95), and 0.30 (0.08-0.79) cm, respectively. The required margin, depending on surrogate used for setup correction, ranged from LR 0.69-0.88 cm and CC 0.76-1.14 cm to AP 0.56-0.7 cm (Table 1). Overall, the registration on mask results in the smallest margins. Conclusion: Substantial position variability of the GTV during RT of esophageal cancer was observed as well as interpatient variation in respiratory-induced motion. Tumor localization is considerably improved, compared to bony setup, when the GTV mask registration is used in CBCT guidance, despite low soft tissue contrast. Furthermore, patient-specific margins are required to mitigate breathing-induced motion.

Motion management in particle therapy

“…"Supervised learning IGPT" combines IGPT and machine learning. Several publications have reported on markerless tumor tracking techniques using kernel-based algorithms, 36 regression analysis, [37][38][39] and the multiple-template-matching technique. 40 Intrafractional bony structure motion, especially that of ribs, can pose a problem in markerless tumor tracking in the thorax, but may be dealt with by dual-energy subtraction.…”

Value of Three-Dimensional Imaging Systems for Image-Guided Carbon Ion Radiotherapy

“…Deviations from mismatch or tumor displacement may still occur during treatment. To decrease the frequency of those errors, many markerless tumor-tracking techniques based on data training (machine learning) have been developed [37,38,39]. Promisingly, they have been successfully implemented by clinically combining with phase-controlled-rescanning (PCR) C-ion RT technology for mobile tumors (lung and liver cancer) with an extremely high gating accuracy [40].…”