Scalable radiotherapy data curation infrastructure for deep-learning based autosegmentation of organs-at-risk: A case study in head and neck cancer

In this era of patient-centered, outcomes-driven and adaptive radiotherapy, deep learning is now being successfully applied to tackle imaging-related workflow bottlenecks such as autosegmentation and dose planning. These applications typically require supervised learning approaches enabled by relatively large, curated radiotherapy datasets which are highly reflective of the contemporary standard of care. However, little has been previously published describing technical infrastructure, recommendations, methods or standards for radiotherapy dataset curation in a holistic fashion. Our radiation oncology department has recently embarked on a large-scale project in partnership with an external partner to develop deep-learning-based tools to assist with our radiotherapy workflow, beginning with autosegmentation of organs-at-risk. This project will require thousands of carefully curated radiotherapy datasets comprising all body sites we routinely treat with radiotherapy. Given such a large project scope, we have approached the need for dataset curation rigorously, with an aim towards building infrastructure that is compatible with efficiency, automation and scalability. Focusing on our first use-case pertaining to head and neck cancer, we describe our developed infrastructure and novel methods applied to radiotherapy dataset curation, inclusive of personnel and workflow organization, dataset selection, expert organ-at-risk segmentation, quality assurance, patient de-identification, data archival and transfer. Over the course of approximately 13 months, our expert multidisciplinary team generated 490 curated head and neck radiotherapy datasets. This task required approximately 6000 human-expert hours in total (not including planning and infrastructure development time). This infrastructure continues to evolve and will support ongoing and future project efforts.

Compensators for dose and scatter management in cone-beam computed tomography

The ability of compensators (e.g., bow-tie filters) designed for kV cone-beam computed tomography (CT) to reduce both scatter reaching the detector and dose to the patient is investigated. Scattered x rays reaching the detector are widely recognized as one of the most significant challenges to cone-beam CT imaging performance. With cone-beam CT gaining popularity as a method of guiding treatments in radiation therapy, any methods that have the potential to reduce the dose to patients and/or improve image quality should be investigated. Simple compensators with a design that could realistically be implemented on a cone-beam CT imaging system have been constructed to determine the magnitude of reduction of scatter and/or dose for various cone-beam CT imaging conditions. Depending on the situation, the compensators were shown to reduce x-ray scatter at the detector and dose to the patient by more than a factor of 2. Further optimization of the compensators is a possibility to achieve greater reductions in both scatter and dose.

Single Arc Volumetric Modulated Arc Therapy for Complex Brain Gliomas: Is There an Advantage as Compared to Intensity Modulated Radiotherapy or by Adding a Partial Arc?

The objective of this study was to determine if volumetric modulated arc therapy (VMAT) offers advantages over intensity modulated radiotherapy (IMRT) for complex brain gliomas and evaluate the role of an additional partial arc. Twelve patients with glioma involving critical organs at risk (OAR) were selected [six low grade brainstem glioma (BG) and six glioblastoma (GB) cases]. BGs were prescribed 54 Gy/30 fractions (frx), and GB treated to 50 Gy/30 frx to a lower dose PTV (PTV50) with a simultaneous integrated boost delivering a total dose of 60 Gy/30 frx to a higher dose PTV (PTV60). VMAT was planned with a single arc (VMAT1) and with an additional coplanar partial arc spanning 90° (VMAT2). We observed VMATI improving the PTV equivalent uniform dose (EUD) for BG cases (p = 0.027), improving the V95 for the PTV50 in GB cases (p = 0.026) and resulting in more conformal GB plans (p = 0.008) as compare to IMRT. However, for the GB PTV60, IMRT achieved favorable V95 over VMAT1 and VMAT2 (0.0046 and 0.008, respectively). The GB total integral dose (ID) was significantly lower with VMAT1 and VMAT2 (p = 0.049 and p = 0.006, respectively). Both VMAT1 and VMAT2 reduced the ID, however, only at the 5 Gy threshold for BG cases (p = 0.011 and 0.005, respectively). VMAT achieved a lower spinal cord maximum dose and EUD for BG cases and higher optic nerve doses, otherwise no significant differences were observed. VMAT1 yielded the fastest treatment times and least MU. We conclude that VMAT offers faster treatment delivery for complex brain tumors while maintaining similar dosimetric qualities to IMRT. Selective dosimetric advantages in terms of spinal cord sparing and lowering the ID are observed favoring the use of an additional coplanar partial arc.

Intraoperative cone-beam CT for image-guided tibial plateau fracture reduction

OBJECTIVES

A mobile isocentric C-arm was modified in our laboratory in collaboration with Siemens Medical Solutions to include a large-area flat-panel detector providing multi-mode fluoroscopy and cone-beam CT (CBCT) imaging. This technology is an important advance over existing intraoperative imaging (e.g., Iso-C(3D)), offering superior image quality, increased field of view, higher spatial resolution, and soft-tissue visibility. The aim of this study was to assess the system's performance and image quality in tibial plateau (TP) fracture reconstruction.

METHODS

Three TP fractures were simulated in fresh-frozen cadaveric knees through combined axial loading and lateral impact. The fractures were reduced through a lateral approach and assessed by fluoroscopy. The reconstruction was then assessed using CBCT. If necessary, further reduction and localization of remaining displaced bone fragments was performed using CBCT images for guidance. CBCT image quality was assessed with respect to projection speed, dose and filtering technique.

RESULTS

CBCT imaging provided exquisite visualization of articular details, subtle fragment detection and localization, and confirmation of reduction and implant placement. After fluoroscopic images indicated successful initial reduction, CBCT imaging revealed areas of malalignment and displaced fragments. CBCT facilitated fragment localization and improved anatomic reduction. CBCT image noise increased gradually with reduced dose, but little difference in images resulted from increased projections. High-resolution reconstruction provided better delineation of plateau depressions.

CONCLUSION

This study demonstrated a clear advantage of intraoperative CBCT over 2D fluoroscopy and Iso-C(3D) in TP fracture fixation. CBCT imaging provided benefits in fracture type diagnosis, localization of fracture fragments, and intraoperative 3D confirmation of anatomic reduction.

Intraoperative cone-beam CT for guidance of head and neck surgery: Assessment of dose and image quality using a C-arm prototype

Cone-beam computed tomography (CBCT) with a flat-panel detector represents a promising modality for intraoperative imaging in interventional procedures, demonstrating sub-mm three-dimensional (3D) spatial resolution and soft-tissue visibility. Measurements of patient dose and in-room exposure for CBCT-guided head and neck surgery are reported, and the 3D imaging performance as a function of dose and other acquisition/reconstruction parameters is investigated. Measurements were performed on a mobile isocentric C-arm (Siemens PowerMobil) modified in collaboration with Siemens Medical Solutions (Erlangen, Germany) to provide flat-panel CBCT. Imaging dose was measured in a custom-built 16 cm cylindrical head phantom at four positions (isocenter, anterior, posterior, and lateral) as a function of kVp (80-120 kVp) and C-arm trajectory ("tube-under" and "tube-over" half-rotation orbits). At 100 kVp, for example ("tube-under" orbit), the imaging dose was 0.059 (isocenter), 0.022 (anterior), 0.10 (posterior), and 0.056 (lateral) mGy/ mAs, with scans at approximately 50 and approximately 170 mAs typical for visualization of bony and soft-tissue structures, respectively. Dose to radiosensitive structures (viz., the eyes and thyroid) were considered in particular: significant dose sparing to the eyes (a factor of 5) was achieved using a "tube-under" (rather than "tube-over") half-rotation orbit; a thyroid shield (0.5 mm Pb-equivalent) gave moderate reduction in thyroid dose due to x-ray scatter outside the primary field of view. In-room exposure was measured at positions around the operating table and up to 2 m from isocenter. A typical CBCT scan (10 mGy to isocenter) gave in-air exposure ranging from 29 mR (0.26 mSv) at 35 cm from isocenter, to <0.5 mR (<0.005 mSv) at 2 m from isocenter. Three-dimensional (3D) image quality was assessed in CBCT reconstructions of an anthropomorphic head phantom containing contrast-detail spheres (11-103 HU; 1.6-12.7 mm) and a natural human skeleton. The contrast-to-noise ratio (CNR) was evaluated across a broad range of dose (0.6-23.3 mGy). CNR increased as the square root of dose, with excellent visualization of bony and soft-tissue structures achieved at approximately 3 mGy (0.10 mSv) and approximately 10 mGy (0.35 mSv), respectively. The prototype C-arm demonstrates CBCT image quality sufficient for guidance of head and neck procedures based on soft-tissue and bony anatomy at dose levels low enough for repeat intraoperative imaging, with total dose over the course of the procedure comparable to or less than the effective dose of a typical (2 mSv) diagnostic CT of the head.

A simple, direct method for x-ray scatter estimation and correction in digital radiography and cone-beam CT

X-ray scatter poses a significant limitation to image quality in cone-beam CT (CBCT), resulting in contrast reduction, image artifacts, and lack of CT number accuracy. We report the performance of a simple scatter correction method in which scatter fluence is estimated directly in each projection from pixel values near the edge of the detector behind the collimator leaves. The algorithm operates on the simple assumption that signal in the collimator shadow is attributable to x-ray scatter, and the 2D scatter fluence is estimated by interpolating between pixel values measured along the top and bottom edges of the detector behind the collimator leaves. The resulting scatter fluence estimate is subtracted from each projection to yield an estimate of the primary-only images for CBCT reconstruction. Performance was investigated in phantom experiments on an experimental CBCT bench-top, and the effect on image quality was demonstrated in patient images (head, abdomen, and pelvis sites) obtained on a preclinical system for CBCT-guided radiation therapy. The algorithm provides significant reduction in scatter artifacts without compromise in contrast-to-noise ratio (CNR). For example, in a head phantom, cupping artifact was essentially eliminated, CT number accuracy was restored to within 3%, and CNR (breast-to-water) was improved by up to 50%. Similarly in a body phantom, cupping artifact was reduced by at least a factor of 2 without loss in CNR. Patient images demonstrate significantly increased uniformity, accuracy, and contrast, with an overall improvement in image quality in all sites investigated. Qualitative evaluation illustrates that soft-tissue structures that are otherwise undetectable are clearly delineated in scatter-corrected reconstructions. Since scatter is estimated directly in each projection, the algorithm is robust with respect to system geometry, patient size and heterogeneity, patient motion, etc. Operating without prior information, analytical modeling, or Monte Carlo, the technique is easily incorporated as a preprocessing step in CBCT reconstruction to provide significant scatter reduction.

Investigation of C-arm cone-beam CT-guided surgery of the frontal recess

OBJECTIVE/HYPOTHESIS

A cone-beam CT (CBCT) imaging system based on a mobile C-arm (Siemens PowerMobil) incorporating a high-performance flat-panel detector (Varian PaxScan) has been developed in our laboratory. We hypothesize that intraoperative C-arm CBCT provides image quality and guidance performance sufficient to assist surgical approach to the frontal recess.

STUDY DESIGN

A preclinical prospective study was conducted using six cadaver heads to assess the performance characteristics and the potential clinical utility of this imaging system.

METHODS

The mobile C-arm was employed for intraoperative CBCT guidance of the endoscopic approach to twelve frontal recesses.

RESULTS

The imaging system is capable of sub-mm 3D spatial resolution with bone and soft-tissue visibility and a field of view sufficient for guidance of head and neck surgery. The system can generate intraoperative, volumetric CT images rapidly with an acceptably low radiation exposure to the patient and with image quality sufficient for most surgical tasks. Moreover, the system is portable and compatible with the surgical setup, providing excellent access to the patient. Finally, the accuracy of the system is not bound to a registration process.

CONCLUSIONS

The ability to create updated images as surgery progresses introduces the concept of 'near-real-time' CT guidance for head and neck surgery. We found that the use of CBCT increased surgical confidence in accessing the frontal recess, resolved ambiguities with anatomical variations, and provided valuable teaching information to surgeons in training in both preoperative planning and correlation between tri-planar CT scans and intraoperative endoscopic findings.

An innovative phantom for quantitative and qualitative investigation of advanced x-ray imaging technologies

Development, characterization, and quality assurance of advanced x-ray imaging technologies require phantoms that are quantitative and well suited to such modalities. This note reports on the design, construction, and use of an innovative phantom developed for advanced imaging technologies (e.g., multi-detector CT and the numerous applications of flat-panel detectors in dual-energy imaging, tomosynthesis, and cone-beam CT) in diagnostic and image-guided procedures. The design addresses shortcomings of existing phantoms by incorporating criteria satisfied by no other single phantom: (1) inserts are fully 3D--spherically symmetric rather than cylindrical; (2) modules are quantitative, presenting objects of known size and contrast for quality assurance and image quality investigation; (3) features are incorporated in ideal and semi-realistic (anthropomorphic) contexts; and (4) the phantom allows devices to be inserted and manipulated in an accessible module (right lung). The phantom consists of five primary modules: (1) head, featuring contrast-detail spheres approximate to brain lesions; (2) left lung, featuring contrast-detail spheres approximate to lung modules; (3) right lung, an accessible hull in which devices may be placed and manipulated; (4) liver, featuring contrast-detail spheres approximate to metastases; and (5) abdomen/pelvis, featuring simulated kidneys, colon, rectum, bladder, and prostate. The phantom represents a two-fold evolution in design philosophy--from 2D (cylindrically symmetric) to fully 3D, and from exclusively qualitative or quantitative to a design accommodating quantitative study within an anatomical context. It has proven a valuable tool in investigations throughout our institution, including low-dose CT, dual-energy radiography, and cone-beam CT for image-guided radiation therapy and surgery.

Generalized DQE analysis of radiographic and dual-energy imaging using flat-panel detectors

Analysis of detective quantum efficiency (DQE) is an important component of the investigation of imaging performance for flat-panel detectors (FPDs). Conventional descriptions of DQE are limited, however, in that they take no account of anatomical noise (i.e., image fluctuations caused by overlying anatomy), even though such noise can be the most significant limitation to detectability, often outweighing quantum or electronic noise. We incorporate anatomical noise in experimental and theoretical descriptions of the "generalized DQE" by including a spatial-frequency-dependent noise-power term, S(B), corresponding to background anatomical fluctuations. Cascaded systems analysis (CSA) of the generalized DQE reveals tradeoffs between anatomical noise and the factors that govern quantum noise. We extend such analysis to dual-energy (DE) imaging, in which the overlying anatomical structure is selectively removed in image reconstructions by combining projections acquired at low and high kVp. The effectiveness of DE imaging in removing anatomical noise is quantified by measurement of S(B) in an anthropomorphic phantom. Combining the generalized DQE with an idealized task function to yield the detectability index, we show that anatomical noise dramatically influences task-based performance, system design, and optimization. For the case of radiography, the analysis resolves a fundamental and illustrative quandary: The effect of kVp on imaging performance, which is poorly described by conventional DQE analysis but is clarified by consideration of the generalized DQE. For the case of DE imaging, extension of a generalized CSA methodology reveals a potentially powerful guide to system optimization through the optimal selection of the tissue cancellation parameter. Generalized task-based analysis for DE imaging shows an improvement in the detectability index by more than a factor of 2 compared to conventional radiography for idealized detection tasks.

The influence of antiscatter grids on soft-tissue detectability in cone-beam computed tomography with flat-panel detectors

The influence of antiscatter x-ray grids on image quality in cone-beam computed tomography (CT) is evaluated through broad experimental investigation for various anatomical sites (head and body), scatter conditions (scatter-to-primary ratio (SPR) ranging from approximately 10% to 150%), patient dose, and spatial resolution in three-dimensional reconstructions. Studies involved linear grids in combination with a flat-panel imager on a system for kilovoltage cone-beam CT imaging and guidance of radiation therapy. Grids were found to be effective in reducing x-ray scatter "cupping" artifacts, with heavier grids providing increased image uniformity. The system was highly robust against ring artifacts that might arise in CT reconstructions as a result of gridline shadows in the projection data. The influence of grids on soft-tissue detectability was evaluated quantitatively in terms of absolute contrast, voxel noise, and contrast-to-noise ratio (CNR) in cone-beam CT reconstructions of 16 cm "head" and 32 cm "body" cylindrical phantoms. Imaging performance was investigated qualitatively in observer preference tests based on patient images (pelvis, abdomen, and head-and-neck sites) acquired with and without antiscatter grids. The results suggest that although grids reduce scatter artifacts and improve subject contrast, there is little strong motivation for the use of grids in cone-beam CT in terms of CNR and overall image quality under most circumstances. The results highlight the tradeoffs in contrast and noise imparted by grids, showing improved image quality with grids only under specific conditions of high x-ray scatter (SPR> 100%), high imaging dose (Dcenter> 2 cGy), and low spatial resolution (voxel size > or = 1 mm).

Volume CT with a flat-panel detector on a mobile, isocentric C-arm: Pre-clinical investigation in guidance of minimally invasive surgery

A mobile isocentric C-arm (Siemens PowerMobil) has been modified in our laboratory to include a large area flat-panel detector (in place of the x-ray image intensifier), providing multi-mode fluoroscopy and cone-beam computed tomography (CT) imaging capability. This platform represents a promising technology for minimally invasive, image-guided surgical procedures where precision in the placement of interventional tools with respect to bony and soft-tissue structures is critical. The image quality and performance in surgical guidance was investigated in pre-clinical evaluation in image-guided spinal surgery. The control, acquisition, and reconstruction system are described. The reproducibility of geometric calibration, essential to achieving high three-dimensional (3D) image quality, is tested over extended time scales (7 months) and across a broad range in C-arm angulation (up to 45 degrees), quantifying the effect of improper calibration on spatial resolution, soft-tissue visibility, and image artifacts. Phantom studies were performed to investigate the precision of 3D localization (viz., fiber optic probes within a vertebral body) and effect of lateral projection truncation (limited field of view) on soft-tissue detectability in image reconstructions. Pre-clinical investigation was undertaken in a specific spinal procedure (photodynamic therapy of spinal metastases) in five animal subjects (pigs). In each procedure, placement of fiber optic catheters in two vertebrae (L1 and L2) was guided by fluoroscopy and cone-beam CT. Experience across five procedures is reported, focusing on 3D image quality, the effects of respiratory motion, limited field of view, reconstruction filter, and imaging dose. Overall, the intraoperative cone-beam CT images were sufficient for guidance of needles and catheters with respect to bony anatomy and improved surgical performance and confidence through 3D visualization and verification of transpedicular trajectories and tool placement. Future investigation includes improvement in image quality, particularly regarding x-ray scatter, motion artifacts and field of view, and integration with optical tracking and navigation systems.

Spektr: A computational tool for x-ray spectral analysis and imaging system optimization

A set of computational tools are presented that allow convenient calculation of x-ray spectra, selection of elemental and compound filters, and calculation of beam quality characteristics, such as half-value layer, mR/mAs, and fluence per unit exposure. The TASMIP model of Boone and Seibert is adapted to a library of high-level language (Matlab) functions and shown to agree with experimental measurements across a wide range of kVp and beam filtration. Modeling of beam filtration is facilitated by a convenient, extensible database of mass and mass-energy attenuation coefficients compiled from the National Institute of Standards and Technology. The functions and database were integrated in a graphical user interface and made available online at http:// www.aip.org/epaps/epaps.html. The functionality of the toolset and potential for investigation of imaging system optimization was illustrated in theoretical calculations of imaging performance across a broad range of kVp, filter material type, and filter thickness for direct and indirect-detection flat-panel imagers. The calculations reveal a number of nontrivial effects in the energy response of such detectors that may not have been guessed from simple K-edge filter techniques, and point to a variety of compelling hypotheses regarding choice of beam filtration that warrant future investigation.

Nitrofurantoin modified release versus trimethoprim or co-trimoxazole in the treatment of uncomplicated urinary tract infection in general practice

A total of 538 patients from 45 different general practice centres across the UK was admitted to an open study and randomized to one of the following treatment groups: nitrofurantoin modified release (MR) 100 mg bd, trimethoprim 200 mg bd or co-trimoxazole 960 mg bd. Each patient received seven days of medication. Clinical cure, defined as relief from symptoms at visit 2, occurred in 87.2% of the patients treated with nitrofurantoin MR, 84.5% of the co-trimoxazole group and 86.5% of the trimethoprim group. The bacteriological cure rate for nitrofurantoin MR was comparable to co-trimoxazole at 82.3% and 83.2%, respectively, with trimethoprim the lowest at 76.8%. Whilst the cure rate for Escherichia coli infection was similar, 81.5% cured with nitrofurantoin MR, 82.5% with co-trimoxazole and 78.4% by trimethoprim, for non-E. coli pathogens nitrofurantoin MR was equivalent to co-trimoxazole with 86.7% cure but higher than trimethoprim at 72.0%. In-vitro sensitivity to all pathogens isolated at baseline was very high for nitrofurantoin at 96.1%, significantly higher than either co-trimoxazole or trimethoprim at 87.5% (P < 0.01). The test drugs were equally well tolerated with 28 patients (15.7%) reporting adverse events with nitrofurantoin MR, 28 (15.5%) with co-trimoxazole and 28 (15.6%) with trimethoprim. However, nitrofurantoin MR showed fewer patients with drug-related adverse events (5.6%) as judged by the investigator, compared to co-trimoxazole (8.8%) or trimethoprim (7.3%). (ABSTRACT TRUNCATED AT 250 WORDS)