Defining the loss function is an important part of neural network design and critically determines the success of deep learning modeling. A significant shortcoming of the conventional loss functions is that they weight all regions in the input image volume equally, despite the fact that the system is known to be heterogeneous (i.e., some regions can achieve high prediction performance more easily than others). Here, we introduce a region-specific loss to lift the implicit assumption of homogeneous weighting for better learning. We divide the entire volume into multiple sub-regions, each with an individualized loss constructed for optimal local performance. Effectively, this scheme imposes higher weightings on the sub-regions that are more difficult to segment, and vice versa. Furthermore, the regional false positive and false negative errors are computed for each input image during a training step and the regional penalty is adjusted accordingly to enhance the overall accuracy of the prediction. Using different public and in-house medical image datasets, we demonstrate that the proposed regionally adaptive loss paradigm outperforms conventional methods in the multi-organ segmentations, without any modification to the neural network architecture or additional data preparation.

Magnetic resonance imaging (MRI) is gaining popularity in guiding radiation treatment for intrahepatic cancers due to its superior soft tissue contrast and potential of monitoring individual motion and liver function. This study investigates a deep learning-based method that generates synthetic CT volumes from T1-weighted MR Dixon images in support of MRI-based intrahepatic radiotherapy treatment planning. Training deep neutral networks for this purpose has been challenged by mismatches between CT and MR images due to motion and different organ filling status. This work proposes to resolve such challenge by generating ‘semi-synthetic’ CT images from rigidly aligned CT and MR image pairs. Contrasts within skeletal elements of the ‘semi-synthetic’ CT images were determined from CT images, while contrasts of soft tissue and air volumes were determined from voxel-wise intensity classification results on MR images. The resulting ‘semi-synthetic’ CT images were paired with their corresponding MR images and used to train a simple U-net model without adversarial components. MR and CT scans of 46 patients were investigated and the proposed method was evaluated for 31 patients with clinical radiotherapy plans, using 3-fold cross validation. The averaged mean absolute errors between synthetic CT and CT images across patients were 24.10 HU for liver, 28.62 HU for spleen, 47.05 HU for kidneys, 29.79 HU for spinal cord, 105.68 HU for lungs and 110.09 HU for vertebral bodies. VMAT and IMRT plans were optimized using CT-derived electron densities, and doses were recalculated using corresponding synthetic CT-derived density grids. Resulting dose differences to planning target volumes and various organs at risk were small, with the average difference less than 0.15 Gy for all dose metrics evaluated. The similarities in both image intensity and radiation dose distributions between CT and synthetic CT volumes demonstrate the accuracy of the method and its potential in supporting MRI-only radiotherapy treatment planning.

A technique for generating MRI-derived synthetic CT volumes (MRCTs) is demonstrated in support of adaptive liver stereotactic body radiation therapy (SBRT). Under IRB approval, 16 subjects with hepatocellular carcinoma were scanned using a single MR pulse sequence (T1 Dixon). Air-containing voxels were identified by intensity thresholding on T1-weighted, water and fat images. The envelope of the anterior vertebral bodies was segmented from the fat image and fuzzy-C-means (FCM) was used to classify each non-air voxel as mid-density, lower-density, bone, or marrow in the abdomen, with only bone and marrow classified within the vertebral body envelope. MRCT volumes were created by integrating the product of the FCM class probability with its assigned class density for each voxel. MRCTs were deformably aligned with corresponding planning CTs and 2-ARC-SBRT-VMAT plans were optimized on MRCTs. Fluence was copied onto the CT density grids, dose recalculated, and compared. The liver, vertebral bodies, kidneys, spleen and cord had median Hounsfield unit differences of less than 60. Median target dose metrics were all within 0.1 Gy with maximum differences less than 0.5 Gy. OAR dose differences were similarly small (median: 0.03 Gy, std:0.26 Gy). Results demonstrate that MRCTs derived from a single abdominal imaging sequence are promising for use in SBRT dose calculation.

Separating bone from air in MR data is one of the major challenges in using MR images to derive synthetic CT. The problem is further complicated when the anatomic regions filled with air are altered across scans due to air mobility, for instance, in pelvic regions, thereby the air regions estimated using an ultrashort echo time (UTE) sequence are invalid in other image series acquired for multispectral classification. This study aims to develop and investigate a female pelvic bone shape model to identify low intensity regions in MRI where air is unlikely to be present in support of synthetic CT generation without UTE imaging. CT scans of 30 patients were collected for the study, 17 of them also have corresponding MR scans. The shape model was built from the CT dataset, where the reference image was aligned to each of the training images using B-spline deformable registration. Principal component analysis was performed on B-spline coefficients for a compact model where shape variance was described by linear combination of principal modes. The model was applied to identify pelvic bone in MR images by deforming the corresponding MR data of the reference image to target MR images, where the search space of the deformation process was constrained within the subspace spanned by principal modes. The local minima in the search space were removed effectively by the shape model, thus supporting an efficient binary search for the optimal solution. We evaluated the model by its efficacy in identifying bone voxels and excluding air regions. The model was tested across the 17 patients that have corresponding MR scans using a leave-one-out cross validation. A simple model using the first leading principal mode only was found to achieve reasonable accuracy, where an averaged 87% of bone voxels were correctly identified. Finally dilation of the optimally fit bone mask by 5 mm was found to cover 96% of bone voxels while minimally impacting the overlap with air (below 0.4%).

Abdominal organ motions introduce geometric uncertainties to gastrointestinal radiotherapy. This study investigated slow drifting motion induced by changes of internal anatomic organ arrangements using a 3D radial MRI sequence with a scan length of 20 min. Breathing motion and cyclic GI motion were first removed through multi-temporal resolution image reconstruction. Slow drifting motion analysis was performed using an image time series consisting of 72 image volumes with a temporal sampling rate of 17 s. B-spline deformable registration was performed to align image volumes of the time series to a reference volume. The resulting deformation fields were used for motion velocity evaluation and patient-specific motion model construction through principal component analysis (PCA). Geometric uncertainties introduced by slow drifting motion were assessed by Hausdorff distances between unions of organs at risk (OARs) at different motion states and reference OAR contours as well as probabilistic distributions of OARs predicted using the PCA model. Thirteen examinations from 11 patients were included in this study. The averaged motion velocities ranged from 0.8 to 1.9 mm min ⁻¹ , 0.7 to 1.6 mm min ⁻¹ , 0.6 to 2.0 mm min ⁻¹ and 0.7 to 1.4 mm min ⁻¹ for the small bowel, colon, duodenum and stomach respectively; the averaged Hausdorff distances were 5.6 mm, 5.3 mm, 5.1 mm and 4.6 mm. On average, a margin larger than 4.5 mm was needed to cover a space with OAR occupancy probability higher than 55%. Temporal variations of geometric uncertainties were evaluated by comparing across four 5 min sub-scans extracted from the full scan. Standard deviations of Hausdorff distances across sub-scans were less than 1 mm for most examinations, indicating stability of relative margin estimates from separate time windows. These results suggested slow drifting motion of GI organs is significant and geometric uncertainties introduced by such motion should be accounted for during radiotherapy planning and delivery.

Abdominal organ motions introduce geometric uncertainties to radiotherapy. This study investigates a multi-temporal resolution 3D motion prediction scheme that accounts for both breathing and slow drifting motion in the abdomen in support of MRI-guided radiotherapy. Ten-minute MRI scans were acquired for 8 patients using a volumetric golden-angle stack-of-stars sequence. The first five-minutes was used for patient-specific motion modeling. Fast breathing motion was modeled from high temporal resolution radial k-space samples, which served as a navigator signal to sort k-space data into different bins for high spatial resolution reconstruction of breathing motion states. Slow drifting motion was modeled from a lower temporal resolution image time series which was reconstructed by sequentially combining a large number of breathing-corrected k-space samples. Principal components analysis (PCA) was performed on deformation fields between different motion states. Gaussian kernel regression and linear extrapolation were used to predict PCA coefficients of future motion states for breathing motion (340 ms ahead of acquisition) and slow drifting motion (8.5 s ahead of acquisition) respectively. k-space data from the remaining five-minutes was used to compare ground truth motions states obtained from retrospective reconstruction/deformation with predictions. Median distances between predicted and ground truth centroid positions of gross tumor volume (GTV) and organs at risk (OARs) were less than 1 mm on average. 95- percentile Hausdorff distances between predicted and ground truth GTV contours of various breathing motions states were 2 mm on average, which was smaller than the imaging resolution and 95-percentile Hausdorff distances between predicted and ground truth OAR contours of different slow drifting motion states were less than 0.2 mm. These results suggest that multi-temporal resolution motion models are capable of volumetric predictions of breathing and slow drifting motion with sufficient accuracy and temporal resolution for MRI-based tracking, and thus have potential for supporting MRI-guided abdominal radiotherapy.