S Samant

High performance computing for deformable image registration: towards a new paradigm in adaptive radiotherapy.

The advent of readily available temporal imaging or time series volumetric (4D) imaging has become an indispensable component of treatment planning and adaptive radiotherapy (ART) at many radiotherapy centers. Deformable image registration (DIR) is also used in other areas of medical imaging, including motion corrected image reconstruction. Due to long computation time, clinical applications of DIR in radiation therapy and elsewhere have been limited and consequently relegated to offline analysis. With the recent advances in hardware and software, graphics processing unit (GPU) based computing is an emerging technology for general purpose computation, including DIR, and is suitable for highly parallelized computing. However, traditional general purpose computation on the GPU is limited because the constraints of the available programming platforms. As well, compared to CPU programming, the GPU currently has reduced dedicated processor memory, which can limit the useful working data set for parallelized processing. We present an implementation of the demons algorithm using the NVIDIA 8800 GTX GPU and the new CUDA programming language. The GPU performance will be compared with single threading and multithreading CPU implementations on an Intel dual core 2.4GHz CPU using the C programming language. CUDA provides a C-like language programming interface, and allows for direct access to the highly parallel compute units in the GPU. Comparisons for volumetric clinical lung images acquired using 4DCT were carried out. Computation time for 100 iterations in the range of 1.8-13.5s was observed for the GPU with image size ranging from 2.0×106to14.2×106pixels. The GPU registration was 55-61 times faster than the CPU for the single threading implementation, and 34-39 times faster for the multithreading implementation. For CPU based computing, the computational time generally has a linear dependence on image size for medical imaging data. Computational efficiency is characterized in terms of time per megapixels per iteration (TPMI) with units of seconds per megapixels per iteration (or spmi). For the demons algorithm, our CPU implementation yielded largely invariant values of TPMI. The mean TPMIs were 0.527spmi and 0.335spmi for the single threading and multithreading cases, respectively, with <2% variation over the considered image data range. For GPU computing, we achieved TPMI=0.00916 spmi with 3.7% variation, indicating optimized memory handling under CUDA. The paradigm of GPU based real-time DIR opens up a host of clinical applications for medical imaging.

Fast Deformable Registration on the GPU: A CUDA Implementation of Demons

In the medical imaging field, we need fast deformable registration methods especially in intra-operative settings characterized by their time-critical applications. Image registration studies which are based on graphics processing units (GPUs) provide fast implementations. However, only a small number of these GPU-based studies concentrate on deformable registration. We implemented Demons, a widely used deformable image registration algorithm, on NVIDIApsilas Quadro FX 5600 GPU with the compute unified device architecture (CUDA) programming environment. Using our code, we registered 3D CT lung images of patients. Our results show that we achieved the fastest runtime among the available GPU-based Demons implementations. Additionally, regardless of the given dataset size, we provided a factor of 55 speedup over an optimized CPU-based implementation. Hence, this study addresses the need for on-line deformable registration methods in intra-operative settings by providing the fastest and most scalable Demons implementation available to date. In addition, it provides an implementation of a deformable registration algorithm on a GPU, an understudied type of registration in the general-purpose computation on graphics processors (GPGPU) community.

Characterization of a real‐time surface image‐guided stereotactic positioning system

PURPOSE The AlignRT3C system is an image-guided stereotactic positioning system (IGSPS) that provides real-time target localization. This study involves the first use of this system with three camera pods. The authors have evaluated its localization accuracy and tracking ability using a cone-beam computed tomography (CBCT) system and an optical tracking system in a clinical setting. METHODS A modified Rando head-and-neck phantom and five patients receiving intracranial stereotactic radiotherapy (SRT) were used to evaluate the calibration, registration, and position-tracking accuracies of the AlignRT3C system and to study surface reconstruction uncertainties, including the effects due to interfractional and intrafractional motion, skin tone, room light level, camera temperature, and image registration region of interest selection. System accuracy was validated through comparison with the Elekta kV CBCT system (XVI) and the Varian frameless SonArray (FSA) optical tracking system. Surface-image data sets were acquired with the AlignRT3C daily for the evaluation of pretreatment and interfractional and intrafractional motion for each patient. Results for two different reference image sets, planning CT surface contours (CTS) and previously recorded AlignRT3C optical surface images (ARTS), are reported. RESULTS The system origin displacements for the AlignRT3C and XVI systems agreed to within 1.3 mm and 0.7 degrees. Similar results were seen for AlignRT3C vs FSA. For the phantom displacements having couch angles of 0 degrees, those that utilized ART_S references resulted in a mean difference of 0.9 mm/0.4 degrees with respect to XVI and 0.3 mm/0.2 degrees with respect to FSA. For phantom displacements of more than +/- 10 mm and +/- 3 degrees, the maximum discrepancies between AlignRT and the XVI and FSA systems were 3.0 and 0.4 mm, respectively. For couch angles up to +/- 90 degrees, the mean (max.) difference between the AlignRT3C and FSA was 1.2 (2.3) mm/0.7 degrees (1.2 degrees). For all tests, the mean registration errors obtained using the CT_S references were approximately 1.3 mm/1.0 degrees larger than those obtained using the ART_S references. For the patient study, the mean differences in the pretreatment displacements were 0.3 mm/0.2 degrees between the AlignRT3C and XVI systems and 1.3 mm/1 degrees between the FSA and XVI systems. For noncoplanar treatments, interfractional motion displacements obtained using the ART_S and CT_S references resulted in 90th percentile differences within 2.1 mm/0.8 degrees and 3.3 mm/0.3 degrees, respectively, compared to the FSA system. Intrafractional displacements that were tracked for a maximum of 14 min were within 1 mm/1 degrees of those obtained with the FSA system. Uncertainties introduced by the bite-tray were as high as 3 mm/2 degrees for one patient. The combination of gantry, aSi detector panel, and x-ray tube blockage effects during the CBCT acquisition resulted in a registration error of approximately 3 mm. No skin-tone or surface deformation effects were seen with the limited patient sample. CONCLUSIONS AlignRT3C can be used as a nonionizing IGSPS with accuracy comparable to current image/marker-based systems. IGSPS and CBCT can be combined for high-precision positioning without the need for patient-attached localization devices.

Evaluation of similarity measures for use in the intensity-based rigid 2D-3D registration for patient positioning in radiotherapy.

PURPOSE Rigid 2D-3D registration is an alternative to 3D-3D registration for cases where largely bony anatomy can be used for patient positioning in external beam radiation therapy. In this article, the authors evaluated seven similarity measures for use in the intensity-based rigid 2D-3D registration using a variation in Skerls similarity measure evaluation protocol. METHODS The seven similarity measures are partitioned intensity uniformity, normalized mutual information (NMI), normalized cross correlation (NCC), entropy of the difference image, pattern intensity (PI), gradient correlation (GC), and gradient difference (GD). In contrast to traditional evaluation methods that rely on visual inspection or registration outcomes, the similarity measure evaluation protocol probes the transform parameter space and computes a number of similarity measure properties, which is objective and optimization method independent. The variation in protocol offers an improved property in the quantification of the capture range. The authors used this protocol to investigate the effects of the downsampling ratio, the region of interest, and the method of the digitally reconstructed radiograph (DRR) calculation [i.e., the incremental ray-tracing method implemented on a central processing unit (CPU) or the 3D texture rendering method implemented on a graphics processing unit (GPU)] on the performance of the similarity measures. The studies were carried out using both the kilovoltage (kV) and the megavoltage (MV) images of an anthropomorphic cranial phantom and the MV images of a head-and-neck cancer patient. RESULTS Both the phantom and the patient studies showed the 2D-3D registration using the GPU-based DRR calculation yielded better robustness, while providing similar accuracy compared to the CPU-based calculation. The phantom study using kV imaging suggested that NCC has the best accuracy and robustness, but its slow function value change near the global maximum requires a stricter termination condition for an optimization method. The phantom study using MV imaging indicated that PI, GD, and GC have the best accuracy, while NCC and NMI have the best robustness. The clinical study using MV imaging showed that NCC and NMI have the best robustness. CONCLUSIONS The authors evaluated the performance of seven similarity measures for use in 2D-3D image registration using the variation in Skerls similarity measure evaluation protocol. The generalized methodology can be used to select the best similarity measures, determine the optimal or near optimal choice of parameter, and choose the appropriate registration strategy for the end user in his specific registration applications in medical imaging.

Verification of multileaf collimator leaf positions using an electronic portal imaging device

An automated method is presented for determining individual leaf positions of the Siemens dual focus multileaf collimator (MLC) using the Siemens BEAMVIEW(PLUS) electronic portal imaging device (EPID). Leaf positions are computed with an error of 0.6 mm at one standard deviation (sigma) using separate computations of pixel dimensions, image distortion, and radiation center. The pixel dimensions are calculated by superimposing the film image of a graticule with the corresponding EPID image. A spatial correction is used to compensate for the optical distortions of the EPID, reducing the mean distortion from 3.5 pixels (uncorrected) per localized x-ray marker to 2 pixels (1 mm) for a rigid rotation and 1 pixel for a third degree polynomial warp. A correction for a nonuniform dosimetric response across the field of view of the EPID images is not necessary due to the sharp intensity gradients across leaf edges. The radiation center, calculated from the average of the geometric centers of a square field at 0 degrees and 180 degrees collimator angles, is independent of graticule placement error. Its measured location on the EPID image was stable to within 1 pixel based on 3 weeks of repeated extensions/retractions of the EPID. The MLC leaf positions determined from the EPID images agreed to within a pixel of the corresponding values measured using film and ionization chamber. Several edge detection algorithms were tested: contour, Sobel, Roberts, Prewitt, Laplace, morphological, and Canny. These agreed with each other to within < or = 1.2 pixels for the in-air EPID images. Using a test pattern, individual MLC leaves were found to be typically within 1 mm of the corresponding record-and-verify values, with a maximum difference of 1.8 mm, and standard deviations of <0.3 mm in the daily reproducibility. This method presents a fast, automatic, and accurate alternative to using film or a light field for the verification and calibration of the MLC.

Theoretical analysis and experimental evaluation of a CsI(Tl) based electronic portal imaging system

This article discusses the design and analysis of a portal imaging system based on a thick transparent scintillator. A theoretical analysis using Monte Carlo simulation was performed to calculate the x-ray quantum detection efficiency (QDE), signal to noise ratio (SNR) and the zero frequency detective quantum efficiency [DQE(0)] of the system. A prototype electronic portal imaging device (EPID) was built, using a 12.7 mm thick, 20.32 cm diameter, Csl(Tl) scintillator, coupled to a liquid nitrogen cooled CCD TV camera. The system geometry of the prototype EPID was optimized to achieve high spatial resolution. The experimental evaluation of the prototype EPID involved the determination of contrast resolution, depth of focus, light scatter and mirror glare. Images of humanoid and contrast detail phantoms were acquired using the prototype EPID and were compared with those obtained using conventional and high contrast portal film and a commercial EPID. A theoretical analysis was also carried out for a proposed full field of view system using a large area, thinned CCD camera and a 12.7 mm thick CsI(TI) crystal. Results indicate that this proposed design could achieve DQE(0) levels up to 11%, due to its order of magnitude higher QDE compared to phosphor screen-metal plate based EPID designs, as well as significantly higher light collection compared to conventional TV camera based systems.

High-precision GAFCHROMIC EBT film-based absolute clinical dosimetry using a standard flatbed scanner without the use of a scanner non-uniformity correction

To report a study of the use of GAFCHROMIC EBT radiochromic film (RCF) digitized with a commercially available flatbed document scanner for accurate and reliable all‐purpose two‐dimensional (2D) absolute dosimetry within a clinical environment. We used a simplified methodology that yields high‐precision dosimetry measurements without significant postirradiation correction. The Epson Expression 1680 Professional scanner and the Epson Expression 10000XL scanner were used to digitize the films. Both scanners were retrofitted with light‐diffusing glass to minimize the effects of Newton rings. Known doses were delivered to calibration films. Flat and wedge fields were irradiated with variable depth of solid water and 5 cm back scatter solid water. No particular scanner nonuniformity effect corrections or significant post‐scan image processing were carried out. The profiles were compared with CC04 ionization chamber profiles. The depth dose distribution was measured at a source‐to‐surface distance (SSD) of 100 cm with a field size of 10×10 cm2. Additionally, 22 IMRT fields were measured and evaluated using gamma index analysis. The overall accuracy of RCF with respect to CC04 was found to be 2%–4%. The overall accuracy of RCF was determined using the absolute mean of difference for all flat and wedge field profiles. For clinical IMRT fields, both scanners showed an overall gamma index passing rate greater than 90%. This work demonstrated that EBT films, in conjunction with a commercially available flatbed scanner, can be used as an accurate and precise absolute dosimeter. Both scanners showed that no significant scanner nonuniformity correction is necessary for accurate absolute dosimetry using the EBT films for field sizes smaller than or equal to 15×15 cm2. PACS number: 87.53.Bn

Quality Assessment of Frameless Fractionated Stereotactic Radiotherapy Using Cone Beam Computed Tomography

PURPOSE A quality assessment of intracranial stereotactic radiotherapy was performed using cone beam computed tomography (CBCT). Setup errors were analyzed for two groups of patients: (1) those who were positioned using a frameless SonArray (FSA) system and immobilized with a bite plate and thermoplastic (TP) mask (the bFSA group); and (2) those who were positioned by room laser and immobilized using a TP mask (the mLAS group). METHODS AND MATERIALS A quality assurance phantom was used to study the system differences between FSA and CBCT. The quality assessment was performed using an Elekta Synergy imager (XVI) (Elekta Oncology Systems, Norcross, GA) and an On-Board Imager (OBI) (Varian Medical Systems, Palo Alto, CA) for 25 patients. For the first three fractions, and weekly thereafter, the FSA system was used for patient positioning, after which CBCT was performed to obtain setup errors. RESULTS (1) Phantom tests: The mean differences in the isocenter displacements for the two systems was 1.2 ± 0.7 mm. No significant variances were seen between the XVI and OBI units (p~0.208). (2)Patient tests: The mean of the displacements between FSA and CBCT were independent of the CBCT system used; mean setup errors for the bFSA group were smaller (1.2 mm) than those of the mLAS group (3.2 mm) (p < 0.005). For the mLAS patients, the 90th percentile and the maximum rotational displacements were 3° and 5°, respectively. A 4-mm drift in setup accuracy occurred over the treatment course for 1 bFSA patient. CONCLUSIONS System differences of less than 1 mm between CBCT and FSA were seen. Error regression was observed for the bFSA patients, using CBCT (up to 4 mm) during the treatment course. For the mLAS group, daily CBCT imaging was needed to obtain acceptable setup accuracies.

COMPARISON OF TWO IMMOBILIZATION TECHNIQUES USING PORTAL FILM AND DIGITALLY RECONSTRUCTED RADIOGRAPHS FOR PEDIATRIC PATIENTS WITH BRAIN TUMORS

PURPOSE To compare the accuracy of two immobilization techniques for pediatric brain tumor patients. METHODS AND MATERIALS We analyzed data from 128 treatments involving 22 patients. Patients were immobilized with either a relocatable head frame (12 patients) or a vacuum bag (10 patients). Orthogonal portal films were used as verification images. Errors in patient positioning were measured by comparing verification images with digitally reconstructed radiographs generated by a three-dimensional treatment-planning system. RESULTS With the head frame, systematic errors ranged from 1.4 mm to 2.1 mm; random errors, from 1.7 mm to 2.1 mm. With the vacuum bag, systematic errors ranged from 2.1 mm to 2.5 mm; random errors, from 2.0 mm to 2.6 mm. For the head frame, the mean length of the radial displacement was 4.4 mm; 90% of the total three-dimensional deviation was less than 6.8 mm. The corresponding values for the vacuum bag were 5.0 and 6.6 mm, respectively. CONCLUSIONS The head frame and vacuum bag techniques limit the random and systematic errors in each of the three directions to within +/- 5 mm. We have used these results to determine the margin used to create the planning target volume for conformal radiation therapy.

An image quality comparison study between XVI and OBI CBCT systems

The purpose of this study is to evaluate and compare image quality characteristics for two commonly used and commercially available CBCT systems: the X‐ray Volumetric Imager and the On‐Board Imager. A commonly used CATPHAN image quality phantom was used to measure various image quality parameters, namely, pixel value stability and accuracy, noise, contrast to noise ratio (CNR), high‐contrast resolution, low contrast resolution and image uniformity. For the XVI unit, we evaluated the image quality for four manufacturer‐supplied protocols as a function of mAs. For the OBI unit, we did the same for the full‐fan and half‐fan scanning modes, which were respectively used with the full bow‐tie and half bow‐tie filters. For XVI, the mean pixel values of regions of interest were found to generally decrease with increasing mAs for all protocols, while they were relatively stable with mAs for OBI. Noise was slightly lower on XVI and was seen to decrease with increasing mAs, while CNR increased with mAs for both systems. For XVI and OBI, the high‐contrast resolution was approximately limited by the pixel resolution of the reconstructed image. On OBI images, up to 6 and 5 discs of 1% and 0.5% contrast, respectively, were visible for a high mAs setting using the full‐fan mode, while none of the discs were clearly visible on the XVI images for various mAs settings when the medium resolution reconstruction was used. In conclusion, image quality parameters for XVI and OBI have been quantified and compared for clinical protocols under various mAs settings. These results need to be viewed in the context of a recent study that reported the dose‐mAs relationship for the two systems and found that OBI generally delivered higher imaging doses than XVI. (1) PACS numbers: 85.57.C‐, 85.57.cj, 85.57.cm, 85.57.cf