This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Purpose To develop and evaluate a patch‐based convolutional neural network (CNN) to generate synthetic computed tomography (sCT) images for magnetic resonance (MR)‐only workflow for radiotherapy of head and neck tumors. A patch‐based deep learning method was chosen to improve robustness to abnormal anatomies caused by large tumors, surgical excisions, or dental artifacts. In this study, we evaluate whether the generated sCT images enable accurate MR‐based dose calculations in the head and neck region. Methods We conducted a retrospective study on 34 patients with head and neck cancer who underwent both CT and MR imaging for radiotherapy treatment planning. To generate the sCTs, a large field‐of‐view T2‐weighted Turbo Spin Echo MR sequence was used from the clinical protocol for multiple types of head and neck tumors. To align images as well as possible on a voxel‐wise level, CT scans were nonrigidly registered to the MR (CTreg). The CNN was based on a U‐net architecture and consisted of 14 layers with 3 × 3 × 3 filters. Patches of 48 × 48 × 48 were randomly extracted and fed into the training. sCTs were created for all patients using threefold cross validation. For each patient, the clinical CT‐based treatment plan was recalculated on sCT using Monaco TPS (Elekta). We evaluated mean absolute error (MAE) and mean error (ME) within the body contours and dice scores in air and bone mask. Also, dose differences and gamma pass rates between CT‐ and sCT‐based plans inside the body contours were calculated. Results sCT generation took 4 min per patient. The MAE over the patient population of the sCT within the intersection of body contours was 75 ± 9 Hounsfield Units (HU) (±1 SD), and the ME was 9 ± 11 HU. Dice scores of the air and bone masks (CTreg vs sCT) were 0.79 ± 0.08 and 0.70 ± 0.07, respectively. Dosimetric analysis showed mean deviations of −0.03% ± 0.05% for dose within the body contours and −0.07% ± 0.22% inside the >90% dose volume. Dental artifacts obscuring the CT could be circumvented in the sCT by the CNN‐based approach in combination with Turbo Spin Echo (TSE) magnetic resonance imaging (MRI) sequence that typically is less prone to susceptibility artifacts. Conclusions The presented CNN generated sCTs from conventional MR images without adding scan time to the acquisition. Dosimetric evaluation suggests that dose calculations performed on the sCTs are accurate, and can therefore be used for MR‐only radiotherapy treatment planning of the head and neck.

2.

BackgroundPatients with adolescent idiopathic scoliosis (AIS) are usually investigated by serial imaging studies during the course of treatment, some imaging involves ionizing radiation, and the radiation doses are cumulative. Few studies have addressed the correlation of spinal deformity captured by these different imaging modalities, for which patient positioning are different. To the best of our knowledge, this is the first study to compare the coronal, axial, and sagittal morphology of the scoliotic spine in three different body positions (upright, prone, and supine) and between three different imaging modalities (X-ray, CT, and MRI).MethodsSixty-two AIS patients scheduled for scoliosis surgery, and having undergone standard pre-operative work-up, were included. This work-up included upright full-spine radiographs, supine bending radiographs, supine MRI, and prone CT as is the routine in one of our institutions. In all three positions, Cobb angles, thoracic kyphosis (TK), lumbar lordosis (LL), and vertebral rotation were determined. The relationship among three positions (upright X-ray, prone CT, and supine MRI) was investigated according to the Bland-Altman test, whereas the correlation was described by the intraclass correlation coefficient (ICC).ResultsThoracic and lumbar Cobb angles correlated significantly between conventional radiographs (68° ± 15° and 44° ± 17°), prone CT (54° ± 15° and 33° ± 15°), and supine MRI (57° ± 14° and 35° ± 16°; ICC ≥0.96; P < 0.001). The thoracic and lumbar apical vertebral rotation showed a good correlation among three positions (upright, 22° ± 12° and 11° ± 13°; prone, 20° ± 9° and 8° ± 11°; supine, 16° ± 11° and 6° ± 14°; ICC ≥0.82; P < 0.001). The TK and LL correlated well among three different positions (TK 26° ± 11°, 22° ± 12°, and 17° ± 10°; P ≤ 0.004; LL 49° ± 12°, 45° ± 11°, and 44° ± 12°; P < 0.006; ICC 0.87 and 0.85).ConclusionsAlthough there is a generalized underestimation of morphological parameters of the scoliotic deformity in the supine and prone positions as compared to the upright position, a significant correlation of these parameters is still evident among different body positions by different imaging modalities. Findings of this study suggest that severity of scoliotic deformity in AIS patients can be largely represented by different imaging modalities despite the difference in body positioning.

Purpose Arterial transit time uncertainties and challenges during planning are potential issues for renal perfusion measurement using spatially selective arterial spin labeling techniques. To mitigate these potential issues, a spatially non‐selective technique, such as velocity‐selective arterial spin labeling (VSASL), could be an alternative. This article explores the influence of VSASL sequence parameters and respiratory induced motion on VS‐label generation. Methods VSASL data were acquired in human subjects ( n = 15), with both single and dual labeling, during paced‐breathing, while essential sequence parameters were systematically varied; (1) cutoff velocity, (2) labeling gradient orientation and (3) post‐labeling delay (PLD). Pseudo‐continuous ASL was acquired as a spatially selective reference. In an additional free‐breathing single VSASL experiment ( n = 9) we investigated respiratory motion influence on VS‐labeling. Absolute renal blood flow (RBF), perfusion weighted signal (PWS), and temporal signal‐to‐noise ratio (tSNR) were determined. Results (1) With decreasing cutoff velocity, tSNR and PWS increased. However, undesired tissue labeling occurred at low cutoff velocities (≤ 5.4 cm/s). (2) Labeling gradient orientation had little effect on tSNR and PWS. (3) For single VSASL high signal appeared in the kidney pedicle at PLD < 800 ms, and tSNR and PWS decreased with increasing PLD. For dual VSASL, maximum tSNR occurred at PLD = 1200 ms. Average cortical RBF measured with dual VSASL (264 ± 34 mL/min/100 g) at a cutoff velocity of 5.4 cm/s, and feet‐head labeling was slightly lower than with pseudo‐continuous ASL (283 ± 55 mL/min/100 g). Conclusion With well‐chosen sequence parameters, tissue labeling induced by respiratory motion can be minimized, allowing to obtain good quality RBF maps using planning‐free labeling with dual VSASL.