Gradient Difference Research Articles

MRI-only based synthetic CT generation using dense cycle consistent generative adversarial networks.

Automated synthetic computed tomography (sCT) generation based on magnetic resonance imaging (MRI) images would allow for MRI-only based treatment planning in radiation therapy, eliminating the need for CT simulation and simplifying the patient treatment workflow. In this work, the authors propose a novel method for generation of sCT based on dense cycle-consistent generative adversarial networks (cycle GAN), a deep-learning based model that trains two transformation mappings (MRI to CT and CT to MRI) simultaneously. The cycle GAN-based model was developed to generate sCT images in a patch-based framework. Cycle GAN was applied to this problem because it includes an inverse transformation from CT to MRI, which helps constrain the model to learn a one-to-one mapping. Dense block-based networks were used to construct generator of cycle GAN. The network weights and variables were optimized via a gradient difference (GD) loss and a novel distance loss metric between sCT and original CT. Leave-one-out cross-validation was performed to validate the proposed model. The mean absolute error (MAE), peak signal-to-noise ratio (PSNR), and normalized cross correlation (NCC) indexes were used to quantify the differences between the sCT and original planning CT images. For the proposed method, the mean MAE between sCT and CT were 55.7 Hounsfield units (HU) for 24 brain cancer patients and 50.8HU for 20 prostate cancer patients. The mean PSNR and NCC were 26.6dB and 0.963 in the brain cases, and 24.5dB and 0.929 in the pelvis. We developed and validated a novel learning-based approach to generate CT images from routine MRIs based on dense cycle GAN model to effectively capture the relationship between the CT and MRIs. The proposed method can generate robust, high-quality sCT in minutes. The proposed method offers strong potential for supporting near real-time MRI-only treatment planning in the brain and pelvis.

A computational investigation of the photochemical oxaziridine and amide conversion process of open-chain conjugated nitrone with electron-withdrawing trifluoromethyl group on nitrogen

This computational study investigates the photo-excitation process and subsequent photoproduct formation steps through non-radiative deactivation channels in open-chain conjugated N-substituted nitrone systems (model compounds of corresponding retinylnitrones) having electron-withdrawing groups on nitrogen. Calculations mostly based on CASSCF/6-31G* and CASMP2/6-31G* level of theories on a representative system with N-trifluoromethyl substituent have predicted initial photo-excitation to a planar singlet state. This photochemical path is subsequently followed by a barrierless non-radiative channel towards the lowest-energy conical intersection (CI) geometry having a terminal CNO kink, and situated at 30 kcal/mol below the planar excited state. Following the direction of its gradient difference (GD) vectors, an oxaziridine-type species (R C−O=1.38 Å, R N−O=1.53 Å, < CNO =55.8∘) appears at 3–6 kcal mol −1 below the ground state nitrone system through a transition state (along its reverse direction of minimum-energy path), situated on the reaction pathway. This species with an elongated N-O bond seems to be heading towards an amide geometry. On the other hand, in the opposite GD vector direction a proper oxaziridine geometry has been obtained with a much shorter N-O bond distance (R N−O=1.42 Å). CASSCF-based photochemical studies on conjugated nitrones with N-trifluoromethyl group have revealed a non-radiative decay route of the singlet excited state through a terminally twisted conical intersection. This eventually leads to an oxaziridine-type biradical species with an elongated N-O bond and it seems to be heading towards an amide as photoproduct.

Development of fast patient position verification software using 2D-3D image registration and its clinical experience.

To improve treatment workflow, we developed a graphic processing unit (GPU)-based patient positional verification software application and integrated it into carbon-ion scanning beam treatment. Here, we evaluated the basic performance of the software. The algorithm provides 2D/3D registration matching using CT and orthogonal X-ray flat panel detector (FPD) images. The participants were 53 patients with tumors of the head and neck, prostate or lung receiving carbon-ion beam treatment. 2D/3D-ITchi-Gime (ITG) calculation accuracy was evaluated in terms of computation time and registration accuracy. Registration calculation was determined using the similarity measurement metrics gradient difference (GD), normalized mutual information (NMI), zero-mean normalized cross-correlation (ZNCC), and their combination. Registration accuracy was dependent on the particular metric used. Representative examples were determined to have target registration error (TRE) = 0.45 ± 0.23 mm and angular error (AE) = 0.35 ± 0.18° with ZNCC + GD for a head and neck tumor; TRE = 0.12 ± 0.07 mm and AE = 0.16 ± 0.07° with ZNCC for a pelvic tumor; and TRE = 1.19 ± 0.78 mm and AE = 0.83 ± 0.61° with ZNCC for lung tumor. Calculation time was less than 7.26 s.The new registration software has been successfully installed and implemented in our treatment process. We expect that it will improve both treatment workflow and treatment accuracy.

Mean Shift Segmentation을 이용한 수채화 효과 생성 기법

본 논문에서는 영상처리에서 많이 사용하는 양방향 필터링(bilateral filtering)과 평균 이동 분할(mean shift segmentation)을 이용하여 일반적인 사진을 수채화 효과가 나도록 하는 리터칭 기법에 대하여 제안한다. 먼저 양방향 필터링을 이용하여 사진의 외곽선 부분은 보존하면서 고주파 성분을 약화시키도록 한다. 그리고 양방향 필터링된 영상에서 각각 DoG(Difference of Gradient) 에지 추출과 평균 이동 분할을 수행한다. 이때 DoG 에지 추출은 원영상의 RGB 색상 공간을 CIELAB 공간으로 변환 후 휘도(luminance) 성분만 이용하여 추출하도록 하며 두 결과를 결합하여 최종 영상을 생성한다. 본 논문에서 제안하는 기법으로 다양한 사진에 대하여 실험한 결과 수채화 효과가 잘 나타나는 것을 확인하였으며 특히 주광에서 촬영한 풍경 사진들에 대하여 보다 우수한 성능을 보임을 확인하였다. We propose a retouching method that converts a general photography to a watercolor painting image using bilateral filtering and mean shift segmentation which are mostly used in image processing. The first step is to weaken high frequency components of the image, while preserving the edge of image using the bilateral filtering. And after that we perform DoG(Difference of Gradient) edge extraction and mean shift segmentation respectively from the bilateral filtered image. The DoG edge extraction is performed using luminance component of the image whose RGB color space is transformed into CIELAB space. Experimental result shows that our method can be applied to various types of image and bring better result, especially against the photo taken in daylight.

Prostate Localization on Daily Cone-Beam Computed Tomography Images: Accuracy Assessment of Similarity Metrics

To evaluate different similarity metrics (SM) using natural calcifications and observation-based measures to determine the most accurate prostate and seminal vesicle localization on daily cone-beam CT (CBCT) images. CBCT images of 29 patients were retrospectively analyzed; 14 patients with prostate calcifications (calcification data set) and 15 patients without calcifications (no-calcification data set). Three groups of test registrations were performed. Test 1: 70 CT/CBCT pairs from calcification dataset were registered using 17 SMs (6,580 registrations) and compared using the calcification mismatch error as an endpoint. Test 2: Using the four best SMs from Test 1, 75 CT/CBCT pairs in the no-calcification data set were registered (300 registrations). Accuracy of contour overlays was ranked visually. Test 3: For the best SM from Tests 1 and 2, accuracy was estimated using 356 CT/CBCT registrations. Additionally, target expansion margins were investigated for generating registration regions of interest. Test 1-Incremental sign correlation (ISC), gradient correlation (GC), gradient difference (GD), and normalized cross correlation (NCC) showed the smallest errors (mu +/- sigma: 1.6 +/- 0.9 approximately 2.9 +/- 2.1 mm). Test 2-Two of the three reviewers ranked GC higher. Test 3-Using GC, 96% of registrations showed <3-mm error when calcifications were filtered. Errors were left/right: 0.1 +/- 0.5mm, anterior/posterior: 0.8 +/- 1.0mm, and superior/inferior: 0.5 +/- 1.1 mm. The existence of calcifications increased the success rate to 97%. Expansion margins of 4-10 mm were equally successful. Gradient-based SMs were most accurate. Estimated error was found to be <3 mm (1.1 mm SD) in 96% of the registrations. Results suggest that the contour expansion margin should be no less than 4 mm.

A volumetric model‐based 2D to 3D registration method for measuring kinematics of natural knees with single‐plane fluoroscopy

Accurate measurement of the three-dimensional (3D) rigid body and surface kinematics of the natural human knee is essential for many clinical applications. Existing techniques are limited either in their accuracy or lack more realistic experimental evaluation of the measurement errors. The purposes of the study were to develop a volumetric model-based 2D to 3D registration method, called the weighted edge-matching score (WEMS) method, for measuring natural knee kinematics with single-plane fluoroscopy to determine experimentally the measurement errors and to compare its performance with that of pattern intensity (PI) and gradient difference (GD) methods. The WEMS method gives higher priority to matching of longer edges of the digitally reconstructed radiograph and fluoroscopic images. The measurement errors of the methods were evaluated based on a human cadaveric knee at 11 flexion positions. The accuracy of the WEMS method was determined experimentally to be less than 0.77 mm for the in-plane translations, 3.06 mm for out-of-plane translation, and 1.13 degrees for all rotations, which is better than that of the PI and GD methods. A new volumetric model-based 2D to 3D registration method has been developed for measuring 3D in vivo kinematics of natural knee joints with single-plane fluoroscopy. With the equipment used in the current study, the accuracy of the WEMS method is considered acceptable for the measurement of the 3D kinematics of the natural knee in clinical applications.

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

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 Skerl's similarity measure evaluation protocol. 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. 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. The authors evaluated the performance of seven similarity measures for use in 2D-3D image registration using the variation in Skerl's 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.

TH‐C‐303A‐08: Prostate Localization Using Daily Cone‐Beam Computed Tomography Imaging: Assessment of Accuracy Using Similarity Metrics and Observation‐Based Measures

Purpose: To estimate the accuracy of daily CBCT/CT prostate rigid‐body registrations using different similarity metrics (SM) and observation‐based measures, and to determine the tradeoffs of these methods. Methods and Materials: Daily CBCT images of 29 prostate cancer patients were retrospectively analyzed; 14 patients with calcifications in the prostate (calcification dataset) and 15 patients without calcifications (no‐calcification dataset). Three groups of test registrations were performed. Test1: 70 CT/CBCT pairs in the calcification dataset were registered using 17 SMs and their various parameters (6580 registrations), and compared using the calcification mismatch vector length as an end‐point. Test2: Using the 4 best SMs based on Test1, 75 CBCT/CT pairs from the no‐calcification dataset were registered using visual comparison as the endpoint (300 registrations), i.e. the accuracy of prostate contour overlays on CBCT images were visually ranked. Test3: For the top ranked SM based on the above 2 tests, the registration accuracy was estimated from 356 CBCT/CT registrations from 11 patients in the calcification dataset with and without calcification filtering from registrations. Results: Test1: Incremental‐sign‐correlation (ISC), gradient‐correlation (GC), gradient‐difference (GD), and normalized‐cross‐correlation (NCC) were the SMs with the least vector length errors (μ±σ: 1.6±0.9∼2.9±2.1 mm). Test2: Using visual criteria, 2 of the 3 reviewers ranked GC slightly higher than the other SMs. Test3: Using GC, 96% of registrations (341/356 registrations) showed less than 3 mm error when calcifications were filtered. Corresponding errors were L/R:0.1±0.5 mm, A/P:0.8±1.0 mm, and S/I: 0.5±1.1 mm. Existence of prostate calcifications increased the number of successful registrations to 97%. Conclusion: The estimated &lt;3mm error was approximately 96% (1.1 mm SD) using GC, ISC and GD similarity measures. These measures offer a relatively quick and accurate means of assessing CBCT/CT registrations in the routine clinical setting. User observation is a necessary requirement in the overall QA of the process.

SU‐GG‐I‐102: Evaluation of Nine Similarity Metrics for Intensity‐Based CT‐To‐CBCT Prostate Registrations

Purpose: The prostate can be localized automatically on daily cone‐beam CT (CBCT) images by means of intensity‐based 3D/3D image registrations. The goal of this study is to evaluate nine different similarity metrics for the best localization accuracy. Method and Materials: Seven prostate patients were retrospectively selected with natural prostate calcifications, which resulted in a dataset of 220 CBCT and 7 simulation CT (SimCT) images. On each SimCT image, a region of interest (ROI) was defined by expanding the treatment planning prostate structure by 6 mm and subtracting the rectum with 10 mm expansion. The expansions were chosen empirically. Then, all CBCT images were registered to the corresponding SimCT images by maximizing the similarity within the defined ROI. Downhill simplex method was employed for optimization with full 6D rigid‐body transformation. The pixel values corresponding to bone and air were also excluded from similarity calculations. After each registration, the calcification mismatch was recorded as a measure of prostate localization error. The registrations were repeated for each similarity metric. The nine similarity metrics investigated were: normalized cross‐correlation (NCC), entropy of difference (EOD), mutual information (MI), correlation ratio (CR), sum of absolute difference (SAD), sum of squared difference (SSD), gradient correlation (GC), gradient difference (GD), and pattern intensity (PI). Results: The absolute error means (±standard deviations) were 2.1(±1.6) mm for GD and GC, 2.6(±2.8) mm for PI, 3.9(±4.3) mm for NCC, and 6.4 (±8.0) mm for MI. The others showed over 10 mm mean errors. The directional errors of GD were −0.1(±0.5) mm, 0.1(±1.4) mm, and −0.2(±2.1) mm in the left‐right, anterior‐posterior, and superior/inferior directions respectively. Conclusion: Among the investigated similarity metrics, registrations using GD and GC showed the most accurate prostate localizations on daily CBCT images.

MO‐E‐330A‐05: Evaluation of Similarity Measures for Use in Intensity‐Based Registration in Radiotherapy

Purpose: In radiotherapy, the implementation of automated patient positioning depends on optimal selection of a similarity measure. Visual inspection of a registration is insufficient for evaluating performance of a measure, since high precision registration is usually involved. A general methodology is presented for evaluation of a measure based on convergence, precision and systematic errors for 2D and 3D imaging. Method and Materials: The following common metrics were studied: mean pixelwise product (PROD), normalized correlation coefficient (NCC), partitioned intensity uniformity (PIU), mutual information (MI), normalized mutual information (NMI), entropy of the difference image (ED), gradient correlation (GC), gradient difference (GD), and pattern intensity (PI). Imaging from an anthropomorphic phantom and clinical data were used. Portal images were acquired with a video camera EPID and a flat‐panel imager. Kilovoltage digitally reconstructed radiographs (DRR) and megavoltage DRRs (MVDRR) were used for reference images. The following shifts were carried out: translations from −10 mm to +10 mm with 0.2 mm increments, and/or rotations from −10° to +10° with 0.2° increments. The effect down‐sampling and image type were also investigated. Results: The selection of a measure depended on image type and down‐sampling schedule. PROD introduced the largest systematic error. NCC, PIU, MI, NMI, and GD typically had systematic errors less than 1 mm and 1°. Down‐sampling has less effect on NCC, PIU, and GC. MI, NMI, GC, GD, and PI have sharp “peaks” around their global optima. Asymmetry in the objective function was observed in NCC, PIU, MI, NMI, and ED. Registrations using GC, GD, and PI were less sensitive to the variation of anatomical structures. Using MVDRRs instead of DRRs increased objective function smoothness. Conclusion: A general methodology was presented for evaluation and selection of a measure. Selection was application dependent, indicating necessity of a validation study, as presented here, prior to clinical use.

DIFFERENCES IN GRADIENTS OF STIMULUS GENERALIZATION AS A FUNCTION OF "ABSTRACT" AND "CONCRETE" ATTITUDE.

DIFFERENCES IN GRADIENTS OF STIMULUS GENERALIZATION AS A FUNCTION OF "ABSTRACT" AND "CONCRETE" ATTITUDE.