Results for “image segmentation” — Page 29

DOE OSTI2015

Lesion insertion in the projection domain: Methods and initial results

Purpose: To perform task-based image quality assessment in CT, it is desirable to have a large number of realistic patient images with known diagnostic truth. One effective way of achieving this objective is to create hybrid images that combine patient images with inserted lesions. Because conventional hybrid images generated in the image domain fails to reflect the impact of scan and reconstruction parameters on lesion appearance, this study explored a projection-domain approach. Methods: Lesions were segmented from patient images and forward projected to acquire lesion projections. The forward-projection geometry was designed according to a commercial CT scanner and accommodated both axial and helical modes with various focal spot movement patterns. The energy employed by the commercial CT scanner for beam hardening correction was measured and used for the forward projection. The lesion projections were inserted into patient projections decoded from commercial CT projection data. The combined projections were formatted to match those of commercial CT raw data, loaded onto a commercial CT scanner, and reconstructed to create the hybrid images. Two validations were performed. First, to validate the accuracy of the forward-projection geometry, images were reconstructed from the forward projections of a virtual ACR phantom and compared to physically acquired ACR phantom images in terms of CT number accuracy and high-contrast resolution. Second, to validate the realism of the lesion in hybrid images, liver lesions were segmented from patient images and inserted back into the same patients, each at a new location specified by a radiologist. The inserted lesions were compared to the original lesions and visually assessed for realism by two experienced radiologists in a blinded fashion. Results: For the validation of the forward-projection geometry, the images reconstructed from the forward projections of the virtual ACR phantom were consistent with the images physically acquired for the ACR phantom in terms of Hounsfield unit and high-contrast resolution. For the validation of the lesion realism, lesions of various types were successfully inserted, including well circumscribed and invasive lesions, homogeneous and heterogeneous lesions, high-contrast and low-contrast lesions, isolated and vessel-attached lesions, and small and large lesions. The two experienced radiologists who reviewed the original and inserted lesions could not identify the lesions that were inserted. The same lesion, when inserted into the projection domain and reconstructed with different parameters, demonstrated a parameter-dependent appearance. Conclusions: A framework has been developed for projection-domain insertion of lesions into commercial CT images, which can be potentially expanded to all geometries of CT scanners. Compared to conventional image-domain methods, the authors’ method reflected the impact of scan and reconstruction parameters on lesion appearance. Compared to prior projection-domain methods, the authors’ method has the potential to achieve higher anatomical complexity by employing clinical patient projections and real patient lesions.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2016

MO-DE-207A-09: Low-Dose CT Image Reconstruction Via Learning From Different Patient Normal-Dose Images

Purpose: To investigate a novel low-dose CT (LdCT) image reconstruction strategy for lung CT imaging in radiation therapy. Methods: The proposed approach consists of four steps: (1) use the traditional filtered back-projection (FBP) method to reconstruct the LdCT image; (2) calculate structure similarity (SSIM) index between the FBP-reconstructed LdCT image and a set of normal-dose CT (NdCT) images, and select the NdCT image with the highest SSIM as the learning source; (3) segment the NdCT source image into lung and outside tissue regions via simple thresholding, and adopt multiple linear regression to learn high-order Markov random field (MRF) pattern for each tissue region in the NdCT source image; (4) segment the FBP-reconstructed LdCT image into lung and outside regions as well, and apply the learnt MRF prior in each tissue region for statistical iterative reconstruction of the LdCT image following the penalized weighted least squares (PWLS) framework. Quantitative evaluation of the reconstructed images was based on the signal-to-noise ratio (SNR), local binary pattern (LBP) and histogram of oriented gradients (HOG) metrics. Results: It was observed that lung and outside tissue regions have different MRF patterns predicted from the NdCT. Visual inspection showed that our method obviously outperformed the traditional FBP method. Comparing with the region-smoothing PWLS method, our method has, in average, 13% increase in SNR, 15% decrease in LBP difference, and 12% decrease in HOG difference from reference standard for all regions of interest, which indicated the superior performance of the proposed method in terms of image resolution and texture preservation. Conclusion: We proposed a novel LdCT image reconstruction method by learning similar image characteristics from a set of NdCT images, and the to-be-learnt NdCT image does not need to be scans from the same subject. This approach is particularly important for enhancing image quality in radiation therapy.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2014

TU-A-9A-06: Semi-Automatic Segmentation of Skin Cancer in High-Frequency Ultrasound Images: Initial Comparison with Histology

Purpose: In skin-cancer radiotherapy, the assessment of skin lesion is challenging, particularly with important features such as the depth and width hard to determine. The aim of this study is to develop interative segmentation method to delineate tumor boundary using high-frequency ultrasound images and to correlate the segmentation results with the histopathological tumor dimensions. Methods: We analyzed 6 patients who comprised a total of 10 skin lesions involving the face, scalp, and hand. The patient’s various skin lesions were scanned using a high-frequency ultrasound system (Episcan, LONGPORT, INC., PA, U.S.A), with a 30-MHz single-element transducer. The lateral resolution was 14.6 micron and the axial resolution was 3.85 micron for the ultrasound image. Semiautomatic image segmentation was performed to extract the cancer region, using a robust statistics driven active contour algorithm. The corresponding histology images were also obtained after tumor resection and served as the reference standards in this study. Results: Eight out of the 10 lesions are successfully segmented. The ultrasound tumor delineation correlates well with the histology assessment, in all the measurements such as depth, size, and shape. The depths measured by the ultrasound have an average of 9.3% difference comparing with that in the histology images. The remaining 2 cases suffered from the situation of mismatching between pathology and ultrasound images. Conclusion: High-frequency ultrasound is a noninvasive, accurate and easy-accessible modality to image skin cancer. Our segmentation method, combined with high-frequency ultrasound technology, provides a promising tool to estimate the extent of the tumor to guide the radiotherapy procedure and monitor treatment response.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2025

Enhancing synchrotron radiation micro-CT images using deep learning: an application of Noise2Inverse on bone imaging

In bone-imaging research, in situ synchrotron radiation micro-computed tomography (SRµCT) mechanical tests are used to investigate the mechanical properties of bone in relation to its microstructure. Low-dose computed tomography (CT) is used to preserve bone's mechanical properties from radiation damage, though it increases noise. To reduce this noise, the self-supervised deep learning method Noise2Inverse was used on low-dose SRµCT images where segmentation using traditional thresholding techniques was not possible. Simulated-dose datasets were created by sampling projection data at full, one-half, one-third, one-fourth and one-sixth frequencies of an in situ SRµCT mechanical test. After convolutional neural networks were trained, Noise2Inverse performance on all dose simulations was assessed visually and by analyzing bone microstructural features. Visually, high image quality was recovered for each simulated dose. Lacunae volume, lacunae aspect ratio and mineralization distributions shifted slightly in full, one-half and one-third dose network results, but were distorted in one-fourth and one-sixth dose network results. Following this, new models were trained using a larger dataset to determine differences between full dose and one-third dose simulations. Significant changes were found for all parameters of bone microstructure, indicating that a separate validation scan may be necessary to apply this technique for microstructure quantification. Noise present during data acquisition from the testing setup was determined to be the primary source of concern for Noise2Inverse viability. While these limitations exist, incorporating dose calculations and optimal imaging parameters enables self-supervised deep learning methods such as Noise2Inverse to be integrated into existing experiments to decrease radiation dose.

Obata, Yoshihiro (ORCID:0000000303659129)↗

DOE OSTI2023

Powder Characterization Inter-Comparison

We performed qualitative and quantitative image analysis on SEM images for 5 uranium samples. Qualitative assessment was completed using the lexicon of Tamasi et al. 2017 on a subset of images from each sample to provide an overall morphological profile of each material. Quantitative analysis of the particles was done using the Morphological Analysis for Materials Attribution, or MAMA, software. We performed particle analysis primarily on samples labeled U Mo, U Si, and UO 2 . Samples labeled ADU and DU Ox were not prioritized for quantitative analysis due to staffing and time it took to segment these images. Two lab analysts worked on this effort, one focusing on the qualitative assessment and the other focusing on the quantitative assessment.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

DOE OSTI2015

WE-B-BRD-00: MRI for Radiation Oncology

The use of MRI in radiation therapy is rapidly increasing. Applications vary from the MRI simulator, to the MRI fused with CT, and to the integrated MRI+RT system. Compared with the standard MRI QA, a broader scope of QA features has to be defined in order to maximize the benefits of using MRI in radiation therapy. These QA features include geometric fidelity, image registration, motion management, cross-system alignment, and hardware interference. Advanced MRI techniques require a specific type of QA, as they are being widely used in radiation therapy planning, dose calculations, post-implant dosimetry, and prognoses. A vigorous and adaptive QA program is crucial to defining the responsibility of the entire radiation therapy group and detecting deviations from the performance of high-quality treatment. As a drastic departure from CT simulation, MRI simulation requires changes in the work flow of treatment planning and image guidance. MRI guided radiotherapy platforms are being developed and commercialized to take the advantage of the advance in knowledge, technology and clinical experience. This symposium will from an educational perspective discuss the scope and specific issues related to MRI guided radiotherapy. Learning Objectives: Understand the difference between a standard and a radiotherapy-specific MRI QA program. Understand the effects of MRI artifacts (geometric distortion and motion) on radiotherapy. Understand advanced MRI techniques (ultrashort echo, fast MRI including dynamic MRI and 4DMRI, diffusion, perfusion, and MRS) and related QA. Understand the methods to prepare MRI for treatment planning (electron density assignment, multimodality image registration, segmentation and motion management). Current status of MRI guided treatment platforms. Dr. Jihong Wang has a research grant with Elekta-MRL project. Dr. Ke Sheng receives research grants from Varian Medical systems.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2015

WE-B-BRD-03: MR QA/QC for MRgRT

The use of MRI in radiation therapy is rapidly increasing. Applications vary from the MRI simulator, to the MRI fused with CT, and to the integrated MRI+RT system. Compared with the standard MRI QA, a broader scope of QA features has to be defined in order to maximize the benefits of using MRI in radiation therapy. These QA features include geometric fidelity, image registration, motion management, cross-system alignment, and hardware interference. Advanced MRI techniques require a specific type of QA, as they are being widely used in radiation therapy planning, dose calculations, post-implant dosimetry, and prognoses. A vigorous and adaptive QA program is crucial to defining the responsibility of the entire radiation therapy group and detecting deviations from the performance of high-quality treatment. As a drastic departure from CT simulation, MRI simulation requires changes in the work flow of treatment planning and image guidance. MRI guided radiotherapy platforms are being developed and commercialized to take the advantage of the advance in knowledge, technology and clinical experience. This symposium will from an educational perspective discuss the scope and specific issues related to MRI guided radiotherapy. Learning Objectives: Understand the difference between a standard and a radiotherapy-specific MRI QA program. Understand the effects of MRI artifacts (geometric distortion and motion) on radiotherapy. Understand advanced MRI techniques (ultrashort echo, fast MRI including dynamic MRI and 4DMRI, diffusion, perfusion, and MRS) and related QA. Understand the methods to prepare MRI for treatment planning (electron density assignment, multimodality image registration, segmentation and motion management). Current status of MRI guided treatment platforms. Dr. Jihong Wang has a research grant with Elekta-MRL project. Dr. Ke Sheng receives research grants from Varian Medical systems.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2023

Image Processing Pipeline for Fluoroelastomer Crystallite Detection in Atomic Force Microscopy Images

Phase transformations in materials systems can be tracked using atomic force microscopy (AFM), enabling the examination of surface properties and macroscale morphologies. In situ measurements investigating phase transformations generate large datasets of time-lapse image sequences. The interpretation of the resulting image sequences, guided by domain-knowledge, requires manual image processing using handcrafted masks. Here this approach is time-consuming and restricts the number of images that can be processed. Her in this study, we developed an automated image processing pipeline which integrates image detection and segmentation methods. We examine five time-series AFM videos of various fluoroelastomer phase transformations. The number of image sequences per video ranges from a hundred to a thousand image sequences. The resulting image processing pipeline aims to automatically classify and analyze images to enable batch processing. Using this pipeline, the growth of each individual fluoroelastomer crystallite can be tracked through time. We incorporated statistical analysis into the pipeline to investigate trends in phase transformations between different fluoroelastomer batches. Understanding these phase transformations is crucial, as it can provide valuable insights into manufacturing processes, improve product quality, and possibly lead to the development of more advanced fluoroelastomer formulations.

36 MATERIALS SCIENCE↗

DOE OSTI2015

WE-B-BRD-01: MR-LINAC

The use of MRI in radiation therapy is rapidly increasing. Applications vary from the MRI simulator, to the MRI fused with CT, and to the integrated MRI+RT system. Compared with the standard MRI QA, a broader scope of QA features has to be defined in order to maximize the benefits of using MRI in radiation therapy. These QA features include geometric fidelity, image registration, motion management, cross-system alignment, and hardware interference. Advanced MRI techniques require a specific type of QA, as they are being widely used in radiation therapy planning, dose calculations, post-implant dosimetry, and prognoses. A vigorous and adaptive QA program is crucial to defining the responsibility of the entire radiation therapy group and detecting deviations from the performance of high-quality treatment. As a drastic departure from CT simulation, MRI simulation requires changes in the work flow of treatment planning and image guidance. MRI guided radiotherapy platforms are being developed and commercialized to take the advantage of the advance in knowledge, technology and clinical experience. This symposium will from an educational perspective discuss the scope and specific issues related to MRI guided radiotherapy. Learning Objectives: Understand the difference between a standard and a radiotherapy-specific MRI QA program. Understand the effects of MRI artifacts (geometric distortion and motion) on radiotherapy. Understand advanced MRI techniques (ultrashort echo, fast MRI including dynamic MRI and 4DMRI, diffusion, perfusion, and MRS) and related QA. Understand the methods to prepare MRI for treatment planning (electron density assignment, multimodality image registration, segmentation and motion management). Current status of MRI guided treatment platforms. Dr. Jihong Wang has a research grant with Elekta-MRL project. Dr. Ke Sheng receives research grants from Varian Medical systems.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2015

WE-B-BRD-02: MR Simulation for Radiation Therapy

The use of MRI in radiation therapy is rapidly increasing. Applications vary from the MRI simulator, to the MRI fused with CT, and to the integrated MRI+RT system. Compared with the standard MRI QA, a broader scope of QA features has to be defined in order to maximize the benefits of using MRI in radiation therapy. These QA features include geometric fidelity, image registration, motion management, cross-system alignment, and hardware interference. Advanced MRI techniques require a specific type of QA, as they are being widely used in radiation therapy planning, dose calculations, post-implant dosimetry, and prognoses. A vigorous and adaptive QA program is crucial to defining the responsibility of the entire radiation therapy group and detecting deviations from the performance of high-quality treatment. As a drastic departure from CT simulation, MRI simulation requires changes in the work flow of treatment planning and image guidance. MRI guided radiotherapy platforms are being developed and commercialized to take the advantage of the advance in knowledge, technology and clinical experience. This symposium will from an educational perspective discuss the scope and specific issues related to MRI guided radiotherapy. Learning Objectives: Understand the difference between a standard and a radiotherapy-specific MRI QA program. Understand the effects of MRI artifacts (geometric distortion and motion) on radiotherapy. Understand advanced MRI techniques (ultrashort echo, fast MRI including dynamic MRI and 4DMRI, diffusion, perfusion, and MRS) and related QA. Understand the methods to prepare MRI for treatment planning (electron density assignment, multimodality image registration, segmentation and motion management). Current status of MRI guided treatment platforms. Dr. Jihong Wang has a research grant with Elekta-MRL project. Dr. Ke Sheng receives research grants from Varian Medical systems.

60 APPLIED LIFE SCIENCES↗

NASA NTRS1991

Computations on the massively parallel processor at the Goddard Space Flight Center

Described are four significant algorithms implemented on the massively parallel processor (MPP) at the Goddard Space Flight Center. Two are in the area of image analysis. Of the other two, one is a mathematical simulation experiment and the other deals with the efficient transfer of data between distantly separated processors in the MPP array. The first algorithm presented is the automatic determination of elevations from stereo pairs. The second algorithm solves mathematical logistic equations capable of producing both ordered and chaotic (or random) solutions. This work can potentially lead to the simulation of artificial life processes. The third algorithm is the automatic segmentation of images into reasonable regions based on some similarity criterion, while the fourth is an implementation of a bitonic sort of data which significantly overcomes the nearest neighbor interconnection constraints on the MPP for transferring data between distant processors.

Strong, James P.↗

DOE OSTI2016

SU-G-BRA-04: Simulation of Errors in Maximal Intensity Projection (MIP)-Based Lung Tumor Internal Target Volumes (ITV) Using Real-Time 2D MRI and Deformable Image Registration Based Lung Tumor Tracking

Purpose: To evaluate the effect of inter- and intra-fractional tumor motion on the error in four-dimensional computed tomography (4DCT) maximal intensity projection (MIP)–based lung tumor internal target volumes (ITV), using deformable image registration of real-time 2D-sagital cine-mode MRI acquired during lung SBRT treatments. Methods: Five lung tumor patients underwent free breathing SBRT treatment on the ViewRay, with dose prescribed to PTV (4DCT MIP-based ITV+3–6mm margin). Sagittal slice cine-MR images (3.5×3.5mm pixels) were acquired through the center of the tumor at 4 frames per second throughout the treatments (3–4 fractions of 21–32 minutes duration). Tumor GTVs were contoured on the first frame of the cine and tracked throughout the treatment using off-line optical-flow based deformable registration implemented on a GPU cluster. Pseudo-4DCT MIP-based ITVs were generated from MIPs of the deformed GTV contours limited to short segments of image data. All possible pseudo-4DCT MIP-based ITV volumes were generated with 1s resolution and compared to the ITV volume of the entire treatment course. Varying pseudo-4DCT durations from 10-50s were analyzed. Results: Tumors were covered in their entirety by PTV in the patients analysed here. However, pseudo-4DCT based ITV volumes were observed that were as small as 29% of the entire treatment-ITV, depending on breathing irregularity and the duration of pseudo-4DCT. With an increase in duration of pseudo-4DCT from 10–50s the minimum volume acquired from 95% of all pseudo-4DCTs increased from 62%–81% of the treatment ITV. Conclusion: A 4DCT MIP-based ITV offers a ‘snap-shot’ of breathing motion for the brief period of time the tumor is imaged on a specific day. Real time MRI over prolonged periods of time and over multiple treatment fractions shows that the accuracy of this snap-shot varies according to inter- and intra-fractional tumor motion. Further work is required to investigate the dosimetric effect of these results.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2022

ASDFL: An adaptive super‐pixel discriminative feature‐selective learning for vehicle matching

Abstract There are a large number of cameras in modern transportation system that capture numerous vehicle images continuously. Therefore, automatic analysis of these vehicle images is helpful for traffic flow management, criminal investigations and vehicle inspections. Vehicle matching, which aims to determine whether two input images depict an identical vehicle, is one of the core tasks in vehicle analysis. Recent relevant studies have focused on local feature extraction instead of global extraction, since local details can provide crucial cues to distinguish between cars. However, these methods do not select local features; that is, they do not assign weights to local features. Therefore, in this research, we systematically study the vehicle matching task, and present a novel annotation‐free local‐based deep learning method called Adaptive super‐pixel discriminative feature‐selective learning (ASDFL) to address this issue. In ASDFL, vehicle images are segmented into clusters of super‐pixels of similar size by considering the location and colour similarities of pixels without using any component‐level annotation. These super‐pixels are deemed to be the virtual components of vehicles. Moreover, a convolutional neural network is used to extract the deep features of these virtual components. Thereafter, an instance‐specific mask generation module driven by the extracted global features is enhanced to produce a mask to select the most distinctive virtual components of each vehicle image pair in the feature space. Finally, the vehicle matching task is accomplished by classifying the selected virtual component features of each imaged vehicle pair. Extensive experiments on two popular vehicle identification benchmarks demonstrate that our method is 1.57% and 0.8% more accurate than the previous baselines in a vehicle matching task on the VeRi and VehicleID datasets, respectively, which demonstrates the effectiveness of our method.

Qin, Rong↗

DOE OSTI2023

Convolutional Neural Network for Segmenting Micro-X-ray Computed Tomography Images of Wood Cellular Structures

To further enhance the performance of wood products, improved tools are needed to study in situ cellular scale phenomena like mechanical deformations and moisture swelling. Micro-X-ray computed tomography (μXCT) using brilliant synchrotron light sources now has the spatial and temporal resolution for real-time visualization of phenomena in three-dimensional cellular structures. However, the tradeoff for speed includes the loss of intensity contrast between different types of materials within the imaged structure, such as cell wall and air in wood. This loss of contrast prevents traditional histogram-based segmentation methods from being used effectively. A new convolutional neural network (CNN) approach was therefore developed to segment fast μXCT images of wood into cell wall and air volumes. The fast μXCT and segmentation were demonstrated in the study of moisture swelling in loblolly pine (Pinus taeda) earlywood and latewood cellular structures conditioned at 0%, 33%, 75%, and 95% relative humidity (RH). The CNN segmentation results had a mean intersection over union (IoU) metric accuracy of 96%. Initial analysis of the swelling in the latewood revealed cell walls swelled about 25% when conditioned from 0% to 95% RH. Additionally, the widths of ray cell lumina in the transverse plane of latewood could be observed to increase at higher RH. The segmentation method presented here will facilitate future quantitative analyses in in situ μXCT studies of wood and other similar cellular materials.

Arzola-Villegas, Xavier (ORCID:0000000305369766)↗

DOE OSTI2014

SU-E-T-13: A Feasibility Study of the Use of Hybrid Computational Phantoms for Improved Historical Dose Reconstruction in the Study of Late Radiation Effects for Hodgkin's Lymphoma

Purpose: Radiation-induced cancer effects are well-documented following radiotherapy. Further investigation is needed to more accurately determine a dose-response relationship for late radiation effects. Recent dosimetry studies tend to use representative patients (Taylor 2009) or anthropomorphic phantoms (Wirth 2008) for estimating organ mean doses. In this study, we compare hybrid computational phantoms to patient-specific voxel phantoms to test the accuracy of University of Florida Hybrid Phantom Library (UFHP Library) for historical dose reconstructions. Methods: A cohort of 10 patients with CT images was used to reproduce the data that was collected historically for Hodgkin's lymphoma patients (i.e. caliper measurements and photographs). Four types of phantoms were generated to show a range of refinement from reference hybrid-computational phantom to patient-specific phantoms. Each patient is matched to a reference phantom from the UFHP Library based on height and weight. The reference phantom is refined in the anterior/posterior direction to create a ‘caliper-scaled phantom’. A photograph is simulated using a surface rendering from segmented CT images. Further refinement in the lateral direction is performed using ratios from a simulated-photograph to create a ‘photograph and caliper-scaled phantom’; breast size and position is visually adjusted. Patient-specific hybrid phantoms, with matched organ volumes, are generated and show the capabilities of the UF Hybrid Phantom Library. Reference, caliper-scaled, photograph and caliper-scaled, and patient-specific hybrid phantoms are compared with patient-specific voxel phantoms to determine the accuracy of the study. Results: Progression from reference phantom to patient specific hybrid shows good agreement with the patient specific voxel phantoms. Each stage of refinement shows an overall trend of improvement in dose accuracy within the study, which suggests that computational phantoms can show improved accuracy in historical dose estimates. Conclusion: Computational hybrid phantoms show promise for improved accuracy within retrospective studies when CTs and other x-ray images are not available.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2015

Factorization-based texture segmentation

This study introduces a factorization-based approach that efficiently segments textured images. We use local spectral histograms as features, and construct an M × N feature matrix using M-dimensional feature vectors in an N-pixel image. Based on the observation that each feature can be approximated by a linear combination of several representative features, we factor the feature matrix into two matrices-one consisting of the representative features and the other containing the weights of representative features at each pixel used for linear combination. The factorization method is based on singular value decomposition and nonnegative matrix factorization. The method uses local spectral histograms to discriminate region appearances in a computationally efficient way and at the same time accurately localizes region boundaries. Finally, the experiments conducted on public segmentation data sets show the promise of this simple yet powerful approach.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2013

Automated registration of multispectral MR vessel wall images of the carotid artery

Purpose: Atherosclerosis is the primary cause of heart disease and stroke. The detailed assessment of atherosclerosis of the carotid artery requires high resolution imaging of the vessel wall using multiple MR sequences with different contrast weightings. These images allow manual or automated classification of plaque components inside the vessel wall. Automated classification requires all sequences to be in alignment, which is hampered by patient motion. In clinical practice, correction of this motion is performed manually. Previous studies applied automated image registration to correct for motion using only nondeformable transformation models and did not perform a detailed quantitative validation. The purpose of this study is to develop an automated accurate 3D registration method, and to extensively validate this method on a large set of patient data. In addition, the authors quantified patient motion during scanning to investigate the need for correction. Methods: MR imaging studies (1.5T, dedicated carotid surface coil, Philips) from 55 TIA/stroke patients with ipsilateral <70% carotid artery stenosis were randomly selected from a larger cohort. Five MR pulse sequences were acquired around the carotid bifurcation, each containing nine transverse slices: T1-weighted turbo field echo, time of flight, T2-weighted turbo spin-echo, and pre- and postcontrast T1-weighted turbo spin-echo images (T1W TSE). The images were manually segmented by delineating the lumen contour in each vessel wall sequence and were manually aligned by applying throughplane and inplane translations to the images. To find the optimal automatic image registration method, different masks, choice of the fixed image, different types of the mutual information image similarity metric, and transformation models including 3D deformable transformation models, were evaluated. Evaluation of the automatic registration results was performed by comparing the lumen segmentations of the fixed image and moving image after registration. Results: The average required manual translation per image slice was 1.33 mm. Translations were larger as the patient was longer inside the scanner. Manual alignment took 187.5 s per patient resulting in a mean surface distance of 0.271 ± 0.127 mm. After minimal user interaction to generate the mask in the fixed image, the remaining sequences are automatically registered with a computation time of 52.0 s per patient. The optimal registration strategy used a circular mask with a diameter of 10 mm, a 3D B-spline transformation model with a control point spacing of 15 mm, mutual information as image similarity metric, and the precontrast T1W TSE as fixed image. A mean surface distance of 0.288 ± 0.128 mm was obtained with these settings, which is very close to the accuracy of the manual alignment procedure. The exact registration parameters and software were made publicly available. Conclusions: An automated registration method was developed and optimized, only needing two mouse clicks to mark the start and end point of the artery. Validation on a large group of patients showed that automated image registration has similar accuracy as the manual alignment procedure, substantially reducing the amount of user interactions needed, and is multiple times faster. In conclusion, the authors believe that the proposed automated method can replace the current manual procedure, thereby reducing the time to analyze the images.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

DOE OSTI2022

A machine learning pipeline for membrane segmentation of cryo-electron tomograms

We describe how to use several machine learning techniques organized in a learning pipeline to segment and identify cell membrane structures from cryo electron tomograms. These tomograms are difficult to analyze with traditional segmentation tools. The learning pipeline in our approach starts from supervised learning via a special convolutional neural network trained with simulated data. It continues with semi-supervised reinforcement learning and/or a region merging technique that tries to piece together disconnected components belonging to the same membrane structure. A parametric or non-parametric fitting procedure is then used to enhance the segmentation results and quantify uncertainties in the fitting. Domain knowledge is used in generating the training data for the neural network and in guiding the fitting procedure through the use of appropriately chosen priors and constraints. We demonstrate that the approach proposed here works well for extracting membrane surfaces in two real tomogram datasets.

97 MATHEMATICS AND COMPUTING↗