Tubular shape aware data generation for segmentation in medical imaging.

PURPOSE: Chest X-ray is one of the most widespread examinations of the human body. In interventional radiology, its use is frequently associated with the need to visualize various tube-like objects, such as puncture needles, guiding sheaths, wires, and catheters. Detection and precise localization of these tube-like objects in the X-ray images are, therefore, of utmost value, catalyzing the development of accurate target-specific segmentation algorithms. Similar to the other medical imaging tasks, the manual pixel-wise annotation of the tubes is a resource-consuming process. METHODS: In this work, we aim to alleviate the lack of annotated images by using artificial data. Specifically, we present an approach for synthetic generation of the tube-shaped objects, with a generative adversarial network being regularized with a prior-shape constraint. Namely, our model uses Frangi-based regularization to draw synthetic tubes in the predefined fake mask regions and, then, uses the adversarial component to preserve the global realistic appearance of the synthesized image. RESULTS: Our method eliminates the need for the paired image-mask data and requires only a weakly labeled dataset, with fine-tuning on a small paired sample (10-20 images) proving sufficient to reach the accuracy of the fully supervised models. CONCLUSION: We report the applicability of the approach for the task of segmenting tubes and catheters in the X-ray images, whereas the results should also hold for the other acquisition modalities and image computing applications that contain tubular objects.

A novel and rapid approach to estimate patient-specific distortions based on mDIXON MRI.

While MRI-only radiation treatment planning (RTP) is becoming more widespread, a robust clinical solution for patient-specific distortion corrections is not available. This work explores B 0 mapping based on mDIXON imaging, often performed for MR-only RTP, as an alternative to separate dual-acquisition gradient-recalled echo imaging, with the overarching goal of developing an efficient and robust approach for patient-specific distortion correction. Initial benchmarking was conducted by scanning a phantom and generating B 0 field maps with two approaches: (1) conventional B 0 mapping and (2) experimental mDIXON imaging. Distortion maps were derived from the field maps and compared. The head and neck regions, including brain, of ten healthy volunteers were then evaluated at 1.5 T and 3 T. Distortion maps were again compared between approaches, using difference maps and histogram analysis. Overall, conventional B 0 mapping was well approximated by mDIXON imaging: The distortions of 95% of the voxels in the phantom estimated by mDIXON and conventional B 0 mapping differed by <0.02 mm (1.5 T) and <0.04 mm (3 T), while the 95-percentiles of the distortions estimated by conventional B 0 mapping were <0.06 mm (1.5 T) and <0.12 mm (3 T). In head and neck the distortions of 99% of the voxels were within ±0.2 mm at 1.5 T for both approaches and within ±0.4 mm and ±0.5 mm at 3 T for mDIXON imaging and conventional B 0 mapping, respectively. The majority of differences in vivo were confined to regions with high spatial variation of the B 0 field, mostly around internal air cavities. For 1.5 T, the mDIXON imaging-based correction alone reduced the 95-percentile of distortions from 0.15 mm to 0.03 mm and within the brain from 0.06 mm to 0.02 mm. Slightly lower reductions were observed at 3 T. In conclusion, mDIXON imaging closely approximated conventional B 0 mapping for patient-specific distortion assessment. Estimates in the brain were in good agreement, and slight differences were observed near air/tissue interfaces in the head and neck. Overall, mDIXON imaging-based B 0 field maps may be advantageous for rapid patient-specific distortion correction without additional imaging.

Geometric and dosimetric impact of anatomical changes for MR-only radiation therapy for the prostate.

PURPOSE: With the move towards magnetic resonance imaging (MRI) as a primary treatment planning modality option for men with prostate cancer, it becomes critical to quantify the potential uncertainties introduced for MR-only planning. This work characterized geometric and dosimetric intra-fractional changes between the prostate, seminal vesicles (SVs), and organs at risk (OARs) in response to bladder filling conditions. MATERIALS AND METHODS: T2-weighted and mDixon sequences (3-4 time points/subject, at 1, 1.5 and 3.0 T with totally 34 evaluable time points) were acquired in nine subjects using a fixed bladder filling protocol (bladder void, 20 oz water consumed pre-imaging, 10 oz mid-session). Using mDixon images, Magnetic Resonance for Calculating Attenuation (MR-CAT) synthetic computed tomography (CT) images were generated by classifying voxels as muscle, adipose, spongy, and compact bone and by assignment of bulk Hounsfield Unit values. Organs including the prostate, SVs, bladder, and rectum were delineated on the T2 images at each time point by one physician. The displacement of the prostate and SVs was assessed based on the shift of the center of mass of the delineated organs from the reference state (fullest bladder). Changes in dose plans at different bladder states were assessed based on volumetric modulated arc radiotherapy (VMAT) plans generated for the reference state. RESULTS: Bladder volume reduction of 70 ± 14% from the final to initial time point (relative to the final volume) was observed in the subject population. In the empty bladder condition, the dose delivered to 95% of the planning target volume (PTV) (D95%) reduced significantly for all cases (11.53 ± 6.00%) likely due to anterior shifts of prostate/SVs relative to full bladder conditions. D15% to the bladder increased consistently in all subjects (42.27 ± 40.52%). Changes in D15% to the rectum were patient-specific, ranging from -23.93% to 22.28% (-0.76 ± 15.30%). CONCLUSIONS: Variations in the bladder and rectal volume can significantly dislocate the prostate and OARs, which can negatively impact the dose delivered to these organs. This warrants proper preparation of patients during treatment and imaging sessions, especially when imaging required longer scan times such as MR protocols.

Per-organ assessment of subject-induced susceptibility distortion for MR-only male pelvis treatment planning.

BACKGROUND: Patient-specific distortions, particularly near tissue/air interfaces, require assessment for magnetic resonance (MR) only radiation treatment planning (RTP). However, patients are dynamic due to changes in physiological status during imaging sessions. This work investigated changes in subject-induced susceptibility distortions to pelvic organs at different bladder states to support pelvis MR-only RTP. METHODS: Pelvises of 9 healthy male volunteers were imaged at 1.0 Tesla (T), 1.5 T, and 3.0 T. Subject-induced susceptibility distortion field maps were generated using a dual-echo gradient-recalled echo (GRE) sequence with B0 field maps obtained from the phase difference between the two echoes acquired at several bladder volume states (3-4/subject, 32 overall). T2 turbo spin echo images were also acquired at each bladder state for organ delineation. Magnet central frequency was tracked over time. Distortion map differences and boxplots were computed to characterize changes within the clinical target volume (CTV), bladder, seminal vesicles, and prostate volumes. RESULTS: The time between the initial and final B0 maps was 42.6 ± 13.9 (range: 13.2-62.1) minutes with minimal change in magnet central frequency (0.02 ± 0.05 mm (range: - 0.06 - 0.12 mm)). Subject-induced susceptibility distortion across all bladder states, field strengths, and subjects was relatively small (1.4-1.9% of all voxels in the prostate and seminal vesicles were distorted > 0.5 mm). In the bladder, no voxels exhibited distortions > 1 mm. An extreme case acquired at 3.0 T with a large volume of rectal air yielded 27.4-34.6% of voxels within the CTVs had susceptibility-induced distortions > 0.5 mm across all time points. CONCLUSIONS: Our work suggests that subject-induced susceptibility distortions caused by bladder/rectal conditions are generally small and subject-dependent. Local changes may be non-negligible within the CTV, thus proper management of filling status is warranted. Future work evaluating the impact of multiple models to accommodate for extreme status changes may be advantageous.

Organ-At-Risk Segmentation in Brain MRI using Model-Based Segmentation: Benefits of Deep Learning-Based Boundary Detectors.

Organ-at-risk (OAR) segmentation is a key step for radiotherapy treatment planning. Model-based segmentation (MBS) has been successfully used for the fully automatic segmentation of anatomical structures and it has proven to be robust to noise due to its incorporated shape prior knowledge. In this work, we investigate the advantages of combining neural networks with the prior anatomical shape knowledge of the model-based segmentation of organs-at-risk for brain radiotherapy (RT) on Magnetic Resonance Imaging (MRI). We train our boundary detectors using two different approaches: classic strong gradients as described in [4] and as a locally adaptive regression task, where for each triangle a convolutional neural network (CNN) was trained to estimate the distances between the mesh triangles and organ boundary, which were then combined into a single network, as described by [1]. We evaluate both methods using a 5-fold cross- validation on both T1w and T2w brain MRI data from sixteen primary and metastatic brain cancer patients (some post-surgical). Using CNN-based boundary detectors improved the results for all structures in both T1w and T2w data. The improvements were statistically significant (p < 0.05) for all segmented structures in the T1w images and only for the auditory system in the T2w images.

Repeatability of dose painting by numbers treatment planning in prostate cancer radiotherapy based on multiparametric magnetic resonance imaging.

Dose painting by numbers (DPBN) refers to a voxel-wise prescription of radiation dose modelled from functional image characteristics, in contrast to dose painting by contours which requires delineations to define the target for dose escalation. The direct relation between functional imaging characteristics and DPBN implies that random variations in images may propagate into the dose distribution. The stability of MR-only prostate cancer treatment planning based on DPBN with respect to these variations is as yet unknown. We conducted a test-retest study to investigate the stability of DPBN for prostate cancer in a semi-automated MR-only treatment planning workflow. Twelve patients received a multiparametric MRI on two separate days prior to prostatectomy. The tumor probability (TP) within the prostate was derived from image features with a logistic regression model. Dose mapping functions were applied to acquire a DPBN prescription map that served to generate an intensity modulated radiation therapy (IMRT) treatment plan. Dose calculations were done on a pseudo-CT derived from the MRI. The TP and DPBN map and the IMRT dose distribution were compared between both MRI sessions, using the intraclass correlation coefficient (ICC) to quantify repeatability of the planning pipeline. The quality of each treatment plan was measured with a quality factor (QF). Median ICC values for the TP and DPBN map and the IMRT dose distribution were 0.82, 0.82 and 0.88, respectively, for linear dose mapping and 0.82, 0.84 and 0.94 for square root dose mapping. A median QF of 3.4% was found among all treatment plans. We demonstrated the stability of DPBN radiotherapy treatment planning in prostate cancer, with excellent overall repeatability and acceptable treatment plan quality. Using validated tumor probability modelling and simple dose mapping techniques it was shown that despite day-to-day variations in imaging data still consistent treatment plans were obtained.

Fat-constrained 18F-FDG PET reconstruction in hybrid PET/MR imaging.

UNLABELLED: Fusion of information from PET and MR imaging can increase the diagnostic value of both modalities. This work sought to improve (18)F FDG PET image quality by using MR Dixon fat-constrained images to constrain PET image reconstruction to low-fat regions, with the working hypothesis that fatty tissue metabolism is low in glucose consumption. METHODS: A novel constrained PET reconstruction algorithm was implemented via a modification of the system matrix in list-mode time-of-flight ordered-subsets expectation maximization reconstruction, similar to the way time-of-flight weighting is incorporated. To demonstrate its use in PET/MR imaging, we modeled a constraint based on fat/water-separating Dixon MR images that shift activity away from regions of fat tissue during PET image reconstruction. PET and MR imaging scans of a modified National Electrical Manufacturers Association/International Electrotechnical Commission body phantom simulating body fat/water composition and in vivo experiments on 2 oncology patients were performed on a commercial time-of-flight PET/MR imaging system. RESULTS: Fat-constrained PET reconstruction visibly and quantitatively increased resolution and contrast between high-uptake and fatty-tissue regions without significantly affecting the images in nonfat regions. CONCLUSION: The incorporation of MR tissue information, such as fat, in image reconstruction can improve the quality of PET images. The combination of a variety of potential other MR tissue characteristics with PET represents a further justification for merging MR data with PET data in hybrid systems.