Search

Purpose: Localization prior to delivery of SBRT to free-breathing patients is performed by aligning the planning internal target volume (ITV) from 4DCT with an on-board free-breathing cone-beam CT (FB-CBCT) image. The FB-CBCT image is assumed to also generate an ITV that captures the full range of motion, due to the acquisition spanning multiple respiratory cycles. However, the ITV could potentially be underestimated when the ratio of time spent in inspiration versus time spent in expiration (I/E ratio) deviates from unity. Therefore, the aim of this study was to investigate the effect of variable I/E ratios on the FB ITV generated from a FB-CBCT scan. Methods: This study employed both phantom and patient imaging data. For the phantom study, five periodic respiratory cycles were simulated with different I/E ratios. Six patient respiratory cycles with variable I/E ratios were also selected. All profiles were then programmed into a motion phantom for imaging and modified to exhibit three peak-to-peak motion amplitudes (0.5, 1.0, and 2.0 cm). Each profile was imaged using two spherical targets with 1.0 and 3.0 cm diameters. 2D projections were acquired with full gantry rotation of a kiloVoltage (kV) imager mounted onto the gantry of a modern linear accelerator. CBCT images were reconstructed from 2D projections using a standard filtered back-projection reconstruction algorithm. Quantitative analyses for the phantom study included computing the change in contrast along the direction of target motion as well as determining the area (which is proportional to the target volume) inside of the contour extracted using a Canny edge detector. For the patient study, projection data that were previously acquired under an investigational 4D CBCT slow-gantry imaging protocol were used to generate both FB-CBCT and 4D CBCT images. Volumes were then manually contoured from both datasets (using the same window and level) for quantitative comparison. Results: The phantom study indicated a reduction in contrast at the inferior edge of the ITV (corresponding to inspiration) as the ratio decreased, for both simulated and patient respiratory cycles. For the simulated phantom respiratory cycles, the contrast reduction of the smallest I/E ratio was 27.6% for the largest target with the smallest amplitude and 89.7% for the smallest target with the largest amplitude. For patient respiratory cycles, these numbers were 22.3% and 94.0%, respectively. The extracted area from inside of the target contours showed a decreasing trend as the I/E ratio decreased. In the patient study, the FB-CBCT ITVs of both lung tumors studied were underestimated when compared with their corresponding 4D CBCT ITV. The underestimations found were 40.1% for the smaller tumor and 24.2% for the larger tumor. Conclusions: The ITV may be underestimated in a FB-CBCT image when a patient’s respiratory pattern is characterized by a disparate length of time spent in inspiration versus expiration. Missing the full target motion information during on-board verification imaging may result in localization errors.

Purpose: To develop a computer-aided detection (CADe) scheme for nodules in chest radiographs (CXRs) with a high sensitivity and a low false-positive (FP) rate. Methods: The authors developed a CADe scheme consisting of five major steps, which were developed for improving the overall performance of CADe schemes. First, to segment the lung fields accurately, the authors developed a multisegment active shape model. Then, a two-stage nodule-enhancement technique was developed for improving the conspicuity of nodules. Initial nodule candidates were detected and segmented by using the clustering watershed algorithm. Thirty-one shape-, gray-level-, surface-, and gradient-based features were extracted from each segmented candidate for determining the feature space, including one of the new features based on the Canny edge detector to eliminate a major FP source caused by rib crossings. Finally, a nonlinear support vector machine (SVM) with a Gaussian kernel was employed for classification of the nodule candidates. Results: To evaluate and compare the scheme to other published CADe schemes, the authors used a publicly available database containing 140 nodules in 140 CXRs and 93 normal CXRs. The CADe scheme based on the SVM classifier achieved sensitivities of 78.6% (110/140) and 71.4% (100/140) with averages of 5.0 (1165/233) FPs/image and 2.0more » (466/233) FPs/image, respectively, in a leave-one-out cross-validation test, whereas the CADe scheme based on a linear discriminant analysis classifier had a sensitivity of 60.7% (85/140) at an FP rate of 5.0 FPs/image. For nodules classified as ''very subtle'' and ''extremely subtle,'' a sensitivity of 57.1% (24/42) was achieved at an FP rate of 5.0 FPs/image. When the authors used a database developed at the University of Chicago, the sensitivities was 83.3% (40/48) and 77.1% (37/48) at an FP rate of 5.0 (240/48) FPs/image and 2.0 (96/48) FPs /image, respectively. Conclusions: These results compare favorably to those described for other commercial and noncommercial CADe nodule detection systems.« less

Accurate arterial centerline extraction is essential for comprehensive visualization in CT Angiography. Time consuming manual tracking is needed when automated methods fail to track centerlines through severely diseased and occluded vessels. A previously described algorithm, Partial Vector Space Projection (PVSP), which uses vessel shape information from a database to bridge occlusions of the femoropopliteal artery, has a limited accuracy in long (>100 mm) occlusions. In this article we introduce a new algorithm, Intermediate Point Detection (IPD), which uses calcifications in the occluded artery to provide additional information about the location of the centerline to facilitate improvement in PVSP performance. It identifies calcified plaque in image space to find the most useful point within the occlusion to improve the estimate from PVSP. In this algorithm candidates for calcified plaque are automatically identified on axial CT slices in a restricted region around the estimate obtained from PVSP. A modified Canny edge detector identifies the edge of the calcified plaque and a convex polygon fit is used to find the edge of the calcification bordering the wall of the vessel. The Hough transform for circles estimates the center of the vessel on the slice, which serves as a candidate intermediate point. Each candidate is characterized by two scores based on radius and relative position within the occluded segment, and a polynomial function is constructed to define a net score representing the potential benefit of using this candidate for improving the centerline. We tested our approach in 44 femoropopliteal artery occlusions of lengths up to 398 mm in 30 patients with peripheral arterial occlusive disease. Centerlines were tracked manually by four-experts, twice each, with their mean serving as the reference standard. All occlusions were first interpolated with PVSP using a database of femoropopliteal arterial shapes obtained from a total of 60 subjects. Occlusions longer than 80 mm (N=20) were then processed with the IPD algorithm, provided calcifications were found (N=14). We used the maximum point-wise distance of an interpolated curve from the reference standard as our error metric. The IPD algorithm significantly reduced the average error of the initial PVSP from 2.76 to 1.86 mm (p

Ultrasonography has been used for breast cancer screening in Japan. Screening using a conventional hand-held probe is operator dependent and thus it is possible that some areas of the breast may not be scanned. To overcome such problems, a mechanical whole breast ultrasound (US) scanner has been proposed and developed for screening purposes. However, another issue is that radiologists might tire while interpreting all images in a large-volume screening; this increases the likelihood that masses may remain undetected. Therefore, the aim of this study is to develop a fully automatic scheme for the detection of masses in whole breast US images in order to assist the interpretations of radiologists and potentially improve the screening accuracy. The authors database comprised 109 whole breast US imagoes, which include 36 masses (16 malignant masses, 5 fibroadenomas, and 15 cysts). A whole breast US image with 84 slice images (interval between two slice images: 2 mm) was obtained by the ASU-1004 US scanner (ALOKA Co., Ltd., Japan). The feature based on the edge directions in each slice and a method for subtracting between the slice images were used for the detection of masses in the authors proposed scheme. The Canny edge detector was applied to detect edges in US images; these edges were classified as near-vertical edges or near-horizontal edges using a morphological method. The positions of mass candidates were located using the near-vertical edges as a cue. Then, the located positions were segmented by the watershed algorithm and mass candidate regions were detected using the segmented regions and the low-density regions extracted by the slice subtraction method. For the removal of false positives (FPs), rule-based schemes and a quadratic discriminant analysis were applied for the distribution between masses and FPs. As a result, the sensitivity of the authors scheme for the detection of masses was 80.6% (29/36) with 3.8 FPs per whole breast image. The authors scheme for a computer-aided detection may be useful in improving the screening performance and efficiency.

Segmentation of lymphoma containing lymph nodes is a difficult task because of multiple variables associated with the tumor's location, intensity distribution, and contrast to its surrounding tissues. In this paper, we present a reliable and practical marker-controlled watershed algorithm for semi-automated segmentation of lymphoma in sequential CT images. Robust determination of internal and external markers is the key to successful use of the marker-controlled watershed transform in the segmentation of lymphoma and is the focus of this work. The external marker in our algorithm is the circle enclosing the lymphoma in a single slice. The internal marker, however, is determined automatically by combining techniques including Canny edge detection, thresholding, morphological operation, and distance map estimation. To obtain tumor volume, the segmented lymphoma in the current slice needs to be propagated to the adjacent slice to help determine the external and internal markers for delineation of the lymphoma in that slice. The algorithm was applied to 29 lymphomas (size range, 9-53 mm in diameter; mean, 23 mm) in nine patients. A blinded radiologist manually delineated all lymphomas on all slices. The manual result served as the "gold standard" for comparison. Several quantitative methods were applied to objectively evaluate the performance of the segmentation algorithm. The algorithm received a mean overlap, overestimation, and underestimation ratios of 83.2%, 13.5%, and 5.5%, respectively. The mean average boundary distance and Hausdorff boundary distance were 0.7 and 3.7 mm. Preliminary results have shown the potential of this computer algorithm to allow reliable segmentation and quantification of lymphomas on sequential CT images.

Accurate liver segmentation on computed tomography(CT)images is a challenging task especially at sites where surrounding tissues (e.g., stomach, kidney) have densities similar to that of the liver and lesions reside at the liver edges. We have developed a method for semiautomatic delineation of the liver contours on contrast-enhanced CTimages. The method utilizes a snake algorithm with a gradient vector flow (GVF) field as its external force. To improve the performance of the GVF snake in the segmentation of the liver contour, an edge map was obtained with a Canny edge detector, followed by modifications using a liver template and a concavity removal algorithm. With the modified edge map, for which unwanted edges inside the liver were eliminated, the GVF field was computed and an initial liver contour was formed. The snake algorithm was then applied to obtain the actual liver contour. This algorithm was extended to segment the liver volume in a slice-by-slice fashion, where the result of the preceding slice constrained the segmentation of the adjacent slice. 551 two-dimensional liverimages from 20 volumetric images with colorectal metastases spreading throughout the livers were delineated using this method, and also manually by a radiologist for evaluation. The difference ratio, which is defined as the percentage ratio of mismatching volume between the computer and the radiologist’s results, ranged from 2.9% to 7.6% with a median value of 5.3%.

A new technique based on normalized binary image correlation between two edge images has been proposed for positioning proton-beam radiotherapy patients. A Canny edge detector was used to extract two edge images from a reference x-ray image and a test x-ray image of a patient before positioning. While translating and rotating the edged test image, the absolute value of the normalized binary image correlation between the two edge images is iteratively maximized. Each time before rotation, dilation is applied to the edged test image to avoid a steep reduction of the image correlation. To evaluate robustness of the proposed method, a simulation has been carried out using 240 simulated edged head front-view images extracted from a reference image by varying parameters of the Canny algorithm with a given range of rotation angles and translation amounts in x and y directions. It was shown that resulting registration errors have an accuracy of one pixel in x and y directions and zero degrees in rotation, even when the number of edge pixels significantly differs between the edged reference image and the edged simulation image. Subsequently, positioning experiments using several sets of head, lung, and hip data have been performed. We have observedmore » that the differences of translation and rotation between manual positioning and the proposed method were within one pixel in translation and one degree in rotation. From the results of the validation study, it can be concluded that a significant reduction in workload for the physicians and technicians can be achieved with this method.« less

Purpose: Evaluate commonly used segmentation algorithms on a commercially available real-time MR image guided radiotherapy (MR-IGRT) system (ViewRay), compare the strengths and weaknesses of each method, with the purpose of improving motion tracking for more accurate radiotherapy. Methods: MR motion images of bladder, kidney, duodenum, and liver tumor were acquired for three patients using a commercial on-board MR imaging system and an imaging protocol used during MR-IGRT. A series of 40 frames were selected for each case to cover at least 3 respiratory cycles. Thresholding, Canny edge detection, fuzzy k-means (FKM), k-harmonic means (KHM), and reaction-diffusion level set evolution (RD-LSE), along with the ViewRay treatment planning and delivery system (TPDS) were included in the comparisons. To evaluate the segmentation results, an expert manual contouring of the organs or tumor from a physician was used as a ground-truth. Metrics value of sensitivity, specificity, Jaccard similarity, and Dice coefficient were computed for comparison. Results: In the segmentation of single image frame, all methods successfully segmented the bladder and kidney, but only FKM, KHM and TPDS were able to segment the liver tumor and the duodenum. For segmenting motion image series, the TPDS method had the highest sensitivity, Jarccard, and Dice coefficients inmore » segmenting bladder and kidney, while FKM and KHM had a slightly higher specificity. A similar pattern was observed when segmenting the liver tumor and the duodenum. The Canny method is not suitable for consistently segmenting motion frames in an automated process, while thresholding and RD-LSE cannot consistently segment a liver tumor and the duodenum. Conclusion: The study compared six different segmentation methods and showed the effectiveness of the ViewRay TPDS algorithm in segmenting motion images during MR-IGRT. Future studies include a selection of conformal segmentation methods based on image/organ-specific information, different filtering methods and their influences on the segmentation results. Parag Parikh receives research grant from ViewRay. Sasa Mutic has consulting and research agreements with ViewRay. Yanle Hu receives travel reimbursement from ViewRay. Iwan Kawrakow and James Dempsey are ViewRay employees.« less

Purpose: Investigating a new approach in MRI based treatment planning using the combination of (Ultrashort Echo Time) UTE and T1 weighted spin echo pulse sequences to delineate air, bone and water (soft tissues) in generating pseudo CT images comparable with CT. Methods: A gel phantom containing chicken bones, ping pang balls filled with distilled water and air bubbles, was made. It scanned with MRI using UTE and 2D T1W SE pulse sequences with (in plane resolution= 0.53mm, slice thickness= 2 mm) and CT with (in plane resolution= 0.5 mm and slice thickness= 0.75mm) as a ground truth for geometrical accuracy. The UTE and T1W SE images were registered with CT using mutual information registration algorithm provided by Philips Pinnacle treatment planning system. The phantom boundaries were detected using Canny edge detection algorithm for CT, and MR images. The bone, air bubbles and water in ping pong balls were segmented from CT images using threshold 300HU, - 950HU and 0HU, respectively. These tissue inserts were automatically segmented from combined UTE and T1W SE images using edge detection and relative intensity histograms of the phantom. The obtained segmentations of air, bone and water inserts were evaluated with those obtained from CT. Results: Bone and air can be clearly differentiated in UTE images comparable to CT. Combining UTE and T1W SE images successfully segmented the air, bone and water. The maximum segmentation differences from combine MRI images (UTE and T1W SE) and CT are within 1.3 mm, 1.1mm for bone, air, respectively. The geometric distortion of UTE sequence is small less than 1 pixel (0.53 mm) of MR image resolution. Conclusion: Our approach indicates that MRI can be used solely for treatment planning and its quality is comparable with CT.

Purpose: To investigate the positional accuracy possible with real time Cherenkov video imaging based patient positioning and movement tracking during EBRT. Methods: In a phase 1 clinical trial, including 12 patients undergoing post-lumpectomy whole breast irradiation, Cherenkov emission was imaged with a time-gated ICCD camera synchronized to the LINAC pulse output, during different fractions of the treatment. Patients were positioned with the aid of the AlignRT system. Inter-fraction setup variation was studied by rigid image registrations between images acquired at individual treatments to the average image from all imaged treatment fractions. The amplitude of respiratory motion was calculated from the registration of each frame of Cherenkov images to the reference. A Canny edge detection algorithm was utilized to highlight the beam field edges and biological features provided by major blood vessels apparent in the images. Results: Real-time Cherenkoscopy could monitor the treatment delivery, patient motion and alignment of the beam edge to the treatment region simultaneously. For all the imaged fractions, the patient positioning errors were within our clinical tolerances (3 mm in shifts and 3 degree in pitch angle rotation), with 4.60% exceeding no more than 1 mm in shifts. The averaged error of repetitive patient positioning was 1.2 mm in linear shift and 0.34 degrees in rotational pitch, consistent with the accuracy reported by the AlignRT system. The edge detection algorithm enhanced features such as field edges and blood vessels. Patient positioning errors and respiratory motion retrieved from rigid image registration were consistent with the edge enhanced images. Besides positioning errors caused by global inaccurate setups, edge enhanced blood vessels indicate the existence of deformations Conclusion: Real-time Cherenkoscopy imaging during EBRT is a novel imaging tool that can be used for treatment monitoring, patient positioning and motion tracking with approximately 1.2mm displacement and 0.34 degree rotational accuracy.

Purpose: The anatomical changes that occur between the simulation CT and daily cone-beam CT (CBCT) are investigated using an automated landmark-guided deformable image registration (LDIR) algorithm with simultaneous intensity correction. LDIR was designed to be accurate in the presence of tissue intensity mismatch and heavy noise contamination. Method: An auto-landmark generation algorithm was used in conjunction with a local small volume (LSV) gradient matching search engine to map corresponding landmarks between the CBCT and planning CT. The LSVs offsets were used to perform an initial deformation, generate landmarks, and correct local intensity mismatch. The landmarks act as stabilizing controlpoints in the Demons objective function. The accuracy of the LDIR algorithm was evaluated on one synthetic case with ground truth and data of ten head and neck cancer patients. The deformation vector field (DVF) accuracy was accessed using a synthetic case. The Root mean square error of the 3D canny edge (RMSECE), mutual information (MI), and feature similarity index metric (FSIM) were used to access the accuracy of LDIR on the patient data. The quality of the corresponding deformed contours was verified by an attending physician. Results: The resulting 90 percentile DVF error for the synthetic case was within 5.63mm formore » the original demons algorithm, 2.84mm for intensity correction alone, 2.45mm using controlpoints without intensity correction, and 1.48 mm for the LDIR algorithm. For the five patients the mean RMSECE of the original CT, Demons deformed CT, intensity corrected Demons CT, control-point stabilized deformed CT, and LDIR CT was 0.24, 0.26, 0.20, 0.20, and 0.16 respectively. Conclusion: LDIR is accurate in the presence of multimodal intensity mismatch and CBCT noise contamination. Since LDIR is GPU based it can be implemented with minimal additional strain on clinical resources. This project has been supported by a CPRIT individual investigator award RP11032.« less

Purpose: A region identification algorithm has been developed to facilitate the automatic analysis of Module 1 of the ACR CT accreditation phantom. Method and Materials: The ACR CT accreditation phantom was scanned using a GE Lightspeed QX/i scanner, and a preliminary algorithm was developed to identify the centers of the air and bone‐mimicking areas in Module 1 of the phantom. The air and the “bone” regions in the CTimage were first isolated by binarizing the image with a threshold of −500 and 800 HU, respectively. Next, a Canny edge detector was applied to the air‐only binary images to detect the edges, and the center of the phantom was then identified using a Hough‐based circular object detection algorithm. Lastly, the edge of the phantom was removed and the center of the air‐only region is subsequently detected. A similar procedure was used to identify the center of the “bone” region. Results: To assess the algorithm, the CTimage from Module 1 of the ACR CT phantom was computationally rotated to various orientations. The centers of the air and “bone” regions were then located using the preliminary algorithm. The resulting averages for both regions were very reproducible (within 1 HU). Conclusion: The preliminary results above have demonstrated that the algorithm can robustly identify the centers of the air and bone regions. Once the centers of these two regions are located, the required ROIs can be extracted from the original CTimage, and the required statistics computed. Since the phantom is a rigid object, this implies that the other areas of interest in Module 1 can be easily identified. This will be demonstrated in the presentation.

Purpose: The intermediate goal of this work is to develop a novel 3D image boundary detection algorithm in order to reliably detect the soft tissue boundaries in patient CT images. This is the very critical first step towards a potential solution for the currently unsolved problem ‐ automatic verification of deformable image registration (DIR) on patient images. Our hypothetic procedure of DIR verification is 1) tissue boundary detection, 2) boundary matching, and 3) verification of DIR results using the matched tissue boundaries. Methods: A novel 3D image boundary detection algorithm was developed. Multiple steps were used in the tissue boundary procedure, including 3D gradient computation supporting anisotropic voxels, hysteresis thresholding, non‐maxima suppression directly based on the boundary normal directions, boundary thinning, tissue boundary type selection based on the image intensity values on both sides of the boundaries, boundary separation and voxel reconnection based on boundary probability tests. These steps are innovative and critical to enable reliable detection of separated tissue boundaries that would be used in the following boundary matching step. Preprocessing includes automatic couch table removal and non‐local‐mean noise reduction. Post‐processing includes small piece removal and boundary voxels to point coordinate conversion. Results: The algorithm was implemented in MATLAB and tested using patients' head‐neck and pelvis CT images. Results were visually assessed. Compared to the results of the 3D extension of the standard Canny edge detection algorithm, results of the proposed algorithm are significantly more accurate and reliable. Conclusion: The proposed algorithm is a new approach of medical image processing. Combined with image feature matching methods from computer vision fields, it could lead to a potential solution for DIR verification, which is the current roadblock for many clinical applications including treatment adaptation, automatic segmentation and treatment response evaluation.

Purpose: We have developed two automated methods for real‐time detection of intra‐fractional target motion during treatment with the aid of metallic markers based on ‘in‐line’ acquired x‐ray images. Fast detection algorithms are desirable to facilitate tumor tracking applications. Methods: A Siemens linac was equipped with a 16×16″ a‐Si flat panel detector(FPD) underneath the treatment head. A diagnostic x‐ray tube was positioned opposite to the treatment head; geometrically separating MV treatment and kV imaging fields on the FPD. We analyzed x‐ray images of a lung phantom equipped with a marker for better contrast. For the automated marker detection we developed two algorithms as part of our image acquisition application. Both algorithms are based on gray‐value template matching with normalized cross‐correlation (NCC). One of the algorithms pre‐ calculates suitable candidate regions by employing a canny edge detection filter. In this study, we wanted to a) study the quality of template matching under the influence of MV induced image artifacts (e.g. stripe artifacts), and b) investigate if pre‐selection of candidate regions leads to reliable marker detection.Results: Our lung‐phantom study has demonstrated that matching a 21×21 pixel template over an 81×81 image with NCC takes about 9.9+−4.5 ms. This method is very reliable and works for a wide dose range. Using canny pre‐filtering reduces the latency to 1.1+−0.7 ms, but marker detection becomes less reliable at our lowest dose level (0.6 mAs). In the presence of an 8×8 cm2 MV field, the latencies remain unchanged but the correlation factors decrease sharply by up to 22%. We observed that MV induced stripe artifacts overlapping with the marker are the greatest hurdle to successful detection.Conclusions: While template‐matching over the entire image reliably finds the correct marker position, it is possible to accelerate the search for high contrastimages with the help of canny filtering. Supported by Siemens Healthcare/CR.

Purpose: To modify the digital XCAT phantom to include patient‐specific 3D lungtumor trajectories and demonstrate potential applications. Methods: An algorithm is developed to program the existing XCAT phantom to move according to recorded patient 3D lung trajectory data. The algorithm works in a two‐step manner. First, we incorporate a tumor into the existing digital phantom. Introduction of tumor trajectory at this stage would result in an asynchronous motion between tumor and the chest‐wall and diaphragm. In the second step, an image processing based methodology is developed to synchronize tumor trajectory with lung motion. Tumor location is calculated and a scale factor is derived based on the distance of the tumor from chest wall and diaphragm. The distances are calculated by extracting the contours of lungs using the Canny edge detection method and then fitting the edges to a quadratic polynomial. This step synchronizes tumor trajectory with lung motion. The internal motion data from 10 patients are used to verify the accuracy between actual tumor location and the location obtained from a modified XCAT phantom. The modified XCAT phantom is then used to simulate a 4DCT data acquisition and rebinning with a phase sorting algorithm. Results: The accuracy of the tumor locations in the modified XCAT phantom are found to be 0.29, 0.54, and 0.39 mm (RMS) with corresponding standard deviation values of 0.04, 0.17, and 0.06. The sorted 4DCT simulation algorithm produces volume data comparable to ground‐truth data. The image artifacts in the simulated 4DCT are similar to those that are seen clinically. Conclusion: We have developed an algorithm to modify the XCAT digital phantom to realistically account for independent tumor motion. The modified phantom can be used for studies of CT reconstruction, tumor tracking and dose reconstruction.

A novel algorithm, histogram shifting (HS) is presented, which performs edge detection or edge enhancement with the choice of two parameters. The histogram of a region surrounding each pixel is found and translated toward the origin, resulting in the new pixel value. Images from a variety of medical imaging modalities were processed with HS to perform detection and enhancement of edges. Comparison with results obtained from conventional edge detection (e.g., Sobel) and with conventional edge‐enhancement algorithms is discussed. HS appears to perform the edge‐detection operation without introducing ‘‘double‐edge’’ effects often obtained with conventional edge‐detection algorithms. HS also appears to perform edge enhancement without introducing extensive noise artifacts, which may be noticeable with many edge‐enhancement algorithms.

Purpose: To clinically evaluate the accuracy of an automatic patient positioning system based on the image correlation of two edge images in radiotherapy.Method and Materials: Ninety‐six head & neck images from eight patients undergoing proton therapy were compared with a digitally reconstructedradiograph(DRR) of planning CT. Two edge images, a reference image and a test image, were extracted by applying a Canny edge detector algorithm to a DRR and a 2D x‐ray image, respectively, of each patient before positioning. The value for patient positioning was determined using an optimization algorithm based on an image correlation factor, which recognizes shifts between two edge images by translating and rotating the test image with respect to the reference image.Results: In a simulation using a humanoid phantom, performed to verify the effectiveness of the proposed method, no registration errors were observed for given ranges of rotation, pitch and translation in the x, y and z directions. For real patients, however, there were discrepancies between the automatic positioning method and manual positioning by physicians or technicians. Using edged head coronal‐ and sagittal‐view images, the average differences in registration between these two methods for the x, y and z directions were 0.10 cm, 0.09 cm and 0.11 cm, respectively, whereas the maximum discrepancies were 0.34 cm, 0.38 cm, and 0.50 cm, respectively. For rotation and pitch, the average registration errors were 0.96° and 1.25°, respectively, and the maximum errors were 3.6° and 2.3°, respectively. Conclusion: The proposed automatic patient positioning system based on edge image comparison was relatively accurate for head & neck patients. However, image deformation during treatment may render the automatic method less accurate, since the test image many differ significantly from the reference image.

A set of megavoltage (MV) and kilovoltage (kV) images are obtained during a patient's image‐guidedradiotherapytreatment(IGRT). The kV images are acquired prior to treating the fields using an on‐board x‐ray imager. However, The MV images can be acquired both prior and during the actual treatments. Fan‐beam reconstruction of back‐projected kV images obtained at four or more gantry angles can be used to orient the planning CT to match the treatment geometry. Registering the planning CT to the treatment geometry is necessary if one is to perform dosereconstruction using the multi‐planar images and the planning CT, in the absence of a cone beam CT. The MV and kV images provide the exit fluence information for 3D dosereconstruction. Using these 2D treatmentimages along with the 3D planning CT volume provides adequate information about the treatmentanatomy that is needed for the dosereconstruction process. This system might alleviate the need for cone beam CT for patient positioning and/or dosereconstruction. The planar treatmentimages provide adequately sampled sinogram planes in the spatial dimension, but not in the radial direction. Nonetheless, there is enough data in the sinogram space for registering the treatment geometry to the planning CT. Canny edge finding method is used to decipher anatomical structure in the limited reconstructed volume as well as in the CT volume. This technique can detect both strong and weak edges; and it is less likely than the other edge finding techniques to be fooled by noise. The resulting edge maps are registered to each other by affine transformations. The transformation matrices are then applied to the CT volume in order to register it to the treatment geometry.

Purpose: To reduce the workload of manual contouring by developing a semiautomatic method that can successfully perform segmentation of liver,kidney, bladder, lung, and trachea. Method and Materials: A Canny edge detector was used to extract edges. The obtained edge image was used by a gradient vector flow (GVF) snake algorithm to deform an initial surface onto the edge. For the GVF snake to work, the initial surface has to be reasonably close to the edge. This problem was solved by using a fast marching algorithm on the Canny edge image. Several measures were adopted to prevent the initial surface to leak through the places where the edge was broken. Results: This segmentation method was tested on clinical CTimages and found to be very successful in liver,kidney, bladder, lung, and trachea. There are cases when the organ to be segmented contacted with other organs and there is no imagecontrast observed between them. Small error may exist in those areas. Those errors can be easily identified and corrected by some simple human interactions. Conclusions: Our study has shown that the algorithm we developed can successfully perform CTimage segmentation of several important structures. This algorithm can greatly reduce the workload of contouring the patient.