3D Attention M-net for Short-axis Left Ventricular Myocardium Segmentation in Mice MR cardiac Images.

Small rodent cardiac magnetic resonance imaging (MRI) plays an important role in preclinical models of cardiac disease. Accurate myocardial boundaries delineation is crucial to most morphological and functional analysis in rodent cardiac MRIs. However, rodent cardiac MRIs, due to animal's small cardiac volume and high heart rate, are usually acquired with sub-optimal resolution and low signal-to-noise ratio (SNR). These rodent cardiac MRIs can also suffer from signal loss due to the intra-voxel dephasing. These factors make automatic myocardial segmentation challenging. Manual contouring could be applied to label myocardial boundaries but it is usually laborious, time consuming, and not systematically objective. In this study, we present a deep learning approach based on 3D attention M-net to perform automatic segmentation of left ventricular myocardium. In the deep learning architecture, we use dual spatial-channel attention gates between encoder and decoder along with multi-scale feature fusion path after decoder. Attention gates enable networks to focus on relevant spatial information and channel features to improve segmentation performance. A distance derived loss term, besides general dice loss and binary cross entropy loss, was also introduced to our hybrid loss functions to refine segmentation contours. The proposed model outperforms other generic models, like U-Net and FCN, in major segmentation metrics including the dice score (0.9072), Jaccard index (0.8307) and Hausdorff distance (3.1754 pixels), which are comparable to the results achieved by state-of-the-art models on human cardiac ACDC17 datasets.Clinical relevance Small rodent cardiac MRI is routinely used to probe the effect of individual genes or groups of genes on the etiology of a large number of cardiovascular diseases. An automatic myocardium segmentation algorithm specifically designed for these data can enhance accuracy and reproducibility of cardiac structure and function analysis.

Diffeomorphic Upsampling of Serially Acquired Sparse 2D Cross-Sections in Cardiac MRI.

In this paper, we propose a new technique for interpolating shapes in order to upsample a sparsely acquired serial-section image stack. The method is based on a maximum a posteriori estimation strategy which models neighboring sections as observations of random deformations of an image to be estimated. We show the computation of diffeomorphic trajectories between observed sections and define estimated upsampled image sections as a Jacobian-weighted sum of flowing images at corresponding distances along those trajectories. We apply this methodology to upsample stacks of sparse 2D magnetic resonance cross-sections through live mouse hearts. We show that the proposed method results in smoother and more accurate reconstructions over linear interpolation, and report a Dice coefficient of 0.8727 against ground truth segmentations in our dataset and statistically significant improvements in both left ventricular segmentation accuracy and image intensity estimates.

Shape analysis of hypertrophic and hypertensive heart disease using MRI-based 3D surface models of left ventricular geometry.

The focus of this study was to develop advanced mathematical tools to construct high-resolution 3D models of left-ventricular (LV) geometry to evaluate focal geometric differences between patients with hypertrophic cardiomyopathy (HCM) and hypertensive heart disease (HHD) using cardiac magnetic resonance (MR) cross-sectional images. A limiting factor in 3D analysis of cardiac MR cross-sections is the low out-of-plane resolution of the acquired images. To overcome this problem, we have developed a mathematical framework to construct a population-based high-resolution 3D LV triangulated surface (template) in which an iterative matching algorithm maps a surface mesh of a normal heart to a set of cross-sectional contours that were extracted from short-axis cine cardiac MR images of patients who were diagnosed with either HCM or HHD. A statistical analysis was conducted on deformations that were estimated at each surface node to identify shape differences at end-diastole (ED), end-systole (ES), and motion-related shape variation from ED to ES. Some significant shape difference in radial thickness was detected at ES. Differences of LV 3D surface geometry were identified focally on the basal anterior septum wall. Further research is needed to relate these findings to the HCM morphological substrate and to design a classifier to discriminate among different etiologies of LV hypertrophy.

Diffusion tensor imaging of deep gray matter in children treated for brain malignancies.

PURPOSE: Previous DTI studies reported microstructural changes in white matter of patients receiving treatment for brain malignancies. The primary aim of this prospective pilot longitudinal study was to examine if DTI can detect microstructural changes in deep gray matter (as evaluated by the apparent diffusion coefficient, ADC) between pediatric patients treated with cranial radiation therapy and typically developing healthy children. The relationship between ADC and neurobehavioral performance was also examined. METHODS: ADC was measured at 1.5 T in the caudate, putamen, globus pallidus, thalamus, and hippocampus in nine patients (mean age 11.8 years) and nine age-matched healthy controls. The study was designed with four visits: baseline, 6-month, 15-month, and 27-month follow-ups. RESULTS: Patients had 24 % higher overall mean ADC in the hippocampus compared with controls (p = 0.003). Post hoc analyses revealed significantly elevated ADC at baseline (p = 0.003) and at the 27-month follow-up (p = 0.006). Nevertheless, patients performed normally on a verbal memory test considered to be a hippocampus-related function. Relative to controls, patients' performance on the tests of the visual-spatial working memory decreased over time (group by visit, p = 0.036). Both patients and controls showed a decline in motor speed with increasing ADC in the globus pallidus and putamen. CONCLUSIONS: Childhood brain malignancies and their treatment may affect gray matter microstructure as measured by water diffusion. Significant findings in the hippocampus but not other regions suggest that differences in tissue sensitivity to disease- and treatment-related injury among gray matter regions may exist. ADC in basal ganglia may be associated with motor performance.

Estimating dense cardiac 3D motion using sparse 2D tagged MRI cross-sections.

In this work, we describe a new method, an extension of the Large Deformation Diffeomorphic Metric Mapping to estimate three-dimensional deformation of tagged Magnetic Resonance Imaging Data. Our approach relies on performing non-rigid registration of tag planes that were constructed from set of initial reference short axis tag grids to a set of deformed tag curves. We validated our algorithm using in-vivo tagged images of normal mice. The mapping allows us to compute root mean square distance error between simulated tag curves in a set of long axis image planes and the acquired tag curves in the same plane. Average RMS error was 0.31 ± 0.36(SD) mm, which is approximately 2.5 voxels, indicating good matching accuracy.

Ontological labels for automated location of anatomical shape differences.

A method for automated location of shape differences in diseased anatomical structures via high resolution biomedical atlases annotated with labels from formal ontologies is described. In particular, a high resolution magnetic resonance image of the myocardium of the human left ventricle was segmented and annotated with structural terms from an extracted subset of the Foundational Model of Anatomy ontology. The atlas was registered to the end systole template of a previous study of left ventricular remodeling in cardiomyopathy using a diffeomorphic registration algorithm. The previous study used thresholding and visual inspection to locate a region of statistical significance which distinguished patients with ischemic cardiomyopathy from those with nonischemic cardiomyopathy. Using semantic technologies and the deformed annotated atlas, this location was more precisely found. Although this study used only a cardiac atlas, it provides a proof-of-concept that ontologically labeled biomedical atlases of any anatomical structure can be used to automate location-based inferences.

Initial results on development and application of statistical atlas of femoral cartilage in osteoarthritis to determine sex differences in structure: data from the Osteoarthritis Initiative.

PURPOSE: To create an average atlas of knee femoral cartilage morphology, to apply the atlas for quantitative assessment of osteoarthritis (OA), and to study localized sex differences. MATERIALS AND METHODS: High-resolution 3D magnetic resonance imaging (MRI) data of the knee cartilage collected at 3 T as part of the Osteoarthritis Initiative (OAI) were used. An atlas was created based on images from 30 male Caucasian high-risk subjects with no symptomatic OA at baseline. A female cohort of age- and disease-matched Caucasian subjects was also selected from the OAI database. The Jacobian determinant was calculated from the deformation vector fields that nonlinearly registered each subject to the atlas. Statistical analysis based on the general linear model was used to test for regions of significant differences in the Jacobian values between the two cohorts. RESULTS: The average Jacobian was larger in women (1.2 ± 0.078) than in men (1.08 ± 0.097), showing that after global scaling to the male template, the female cartilage was thicker in most regions. Regions showing significant structural differences include the medial weight bearing region, the trochlear (femoral) side of the patellofemoral compartment, and the lateral posterior condyle. CONCLUSION: Sex-based differences in cartilage structure were localized using tensor based morphometry in a cohort of high-risk subjects.

Computational method for identifying and quantifying shape features of human left ventricular remodeling.

Left ventricular remodeling during the development of heart failure is a strong predictor of cardiovascular mortality. However, methods to objectively quantify remodeling-associated shape changes are not routinely available but may be possible with new computational anatomy tools. In this study, we analyzed and compared multi-detector computed tomographic (MDCT) images of ventricular shape at end-systole (ES) and end-diastole (ED) to determine whether regional structural characteristics could be identified and, as a proof of principle, whether differences in hearts of patients with anterior myocardial infarction (MI) and ischemic cardiomyopathy (ICM) could be distinguished from those with global nonischemic cardiomyopathy (NICM). MDCT images of hearts from 11 patients (5 with ICM) with ejection fractions (EF) < 35% were analyzed. An average ventricular shape model (template) was constructed for each cardiac phase by bringing heart shapes into correspondence using linear and nonlinear image matching algorithms. Next, transformation fields were computed between the template image and individual heart images in the population. Principal component analysis (PCA) method was used to quantify ventricular shape differences described by the transformation vector fields. Statistical analysis of PCA coefficients revealed significant ventricular shape differences at ED (p = 0.03) and ES (p = 0.03). For validation, a second set of 14 EF-matched patients (8 with ICM) were evaluated. The discrimination rule learned from the training data set was able to differentiate ICM from NICM patients (p = 0.008). Application of a novel shape analysis method to in vivo human cardiac images acquired on a clinical scanner is feasible and can quantify regional shape differences at end-systole in remodeled myopathic human myocardium. This approach may be useful in identifying differences in the remodeling process between ICM and NICM populations and possibly in differentiating the populations.

Cardiac Motion Analysis in Ischemic and Non-Ischemic Cardiomyopathy Using Parallel Transport.

In this study, we used the multi-detector computed tomographic (MDCT) images of heart left ventricles at end-diastole and end-systole to perform quantitative analysis and comparison of heart motion in patients with anterior wall myocardial infarction and ischemic cardiomyopathy (ICM) versus those with global non-ischemic cardiomyopathy (NICM). MDCT ventricular images of 25 subjects (13 with ICM) with ejection fraction (EF)< 35% were analyzed. We used parallel transport in diffeomorphism under the large deformation diffeomorphic metric mapping framework to translate within subject motion related deformation in a global template coordinate system. We then performed a hypothesis testing on the ventricular motion variation in the global template coordinate. Statistical analysis indicates that there are meaningful ventricular motion differences between ICM and NICM groups. Additionally, subjects with ICM demonstrated less wall thickening at ES in the anterior wall where the pathology is located.

Exploratory voxel-based analysis of diffusion indices and hemispheric asymmetry in normal aging.

Age-related microstructural changes in brain white matter can be studied by utilizing indices derived from diffusion tensor imaging (DTI): apparent diffusion coefficient (ADC) and fractional anisotropy (FA). The objective of this study is to examine alterations in FA and ADC by employing exploratory voxel-based analysis (VBA) and region(s) of interest (ROI)-based analysis. A highly nonlinear registration algorithm was used to align the ADC and FA image volumes of different subjects to perform accurate voxel-level statistics for two age groups, as well as for hemispheric asymmetry for both age groups. VBA shows significant age-related decline in FA with frontal predominance (frontal white matter, and genu and anterior body of the corpus callosum), superior portions of a splenium and highly oriented fibers of the posterior limb of the internal capsule and the anterior and posterior limbs of the external capsule. Hemispheric asymmetry of FA, as assessed by VBA, showed that for the young-age group, significant right-greater-than-left asymmetry exists in the genu, splenium and body of the corpus callosum and that left-greater-than-right asymmetry exists in the anterior limb of the external capsule and in the posterior limb of the internal capsule, thalamus, cerebral peduncle and temporal-parietal regions. VBA of the hemispheric asymmetry of the middle-age group revealed much less asymmetry. Regions showing age-related changes and hemispheric asymmetry from VBA were, for a majority of the findings, in conformance with ROI analysis and with the known pattern of development and age-related degradation of fiber tracks. The study shows the feasibility of the VBA of DTI indices for exploratory investigations of subtle differences in population cohorts, especially when findings are not localized and/or known a priori.

OpenSourcePACS: an extensible infrastructure for medical image management.

The development of comprehensive picture archive and communication systems (PACS) has mainly been limited to proprietary developments by vendors, though a number of freely available software projects have addressed specific image management tasks. The openSourcePACS project aims to provide an open source, common foundation upon which not only can a basic PACS be readily implemented, but to also support the evolution of new PACS functionality through the development of novel imaging applications and services. openSourcePACS consists of four main software modules: 1) image order entry, which enables the ordering and tracking of structured image requisitions; 2) an agent-based image server framework that coordinates distributed image services including routing, image processing, and querying beyond the present digital image and communications in medicine (DICOM) capabilities; 3) an image viewer, supporting standard display and image manipulation tools, DICOM presentation states, and structured reporting; and 4) reporting and result dissemination, supplying web-based widgets for creating integrated reports. All components are implemented using Java to encourage cross-platform deployment. To demonstrate the usage of openSourcePACS, a preliminary application supporting primary care/specialist communication was developed and is described herein. Ultimately, the goal of openSourcePACS is to promote the wide-scale development and usage of PACS and imaging applications within academic and research communities.

Statistical representation of mean diffusivity and fractional anisotropy brain maps of normal subjects.

PURPOSE: To create diffusion tensor atlases from echo planar imaging (EPI) images acquired at 3 T in 10 normal subjects. MATERIALS AND METHODS: Data from 10 right-handed healthy adult volunteers (mean age of 31 +/- 3 years; eight males) were acquired using a 3.0-T scanner. Geometric distortion artifacts correction was accomplished by combining parallel acquisition to reduce the distortion as well as postprocessing by registration to a geometrically accurate T2-weighted fast-spin-echo image. This reduced distortions to within a voxel for most of the internal structures of the brain. The apparent diffusion coefficient (ADC) and fractional anisotropy (FA) atlases were created by warping images using an iterative optical-flow-based local deformation algorithm that used two channels of data: ADC and FA. RESULTS: A three-dimensional distance measure was used to evaluate the accuracy of the registration algorithm with contours defined on two structures: the corpus callosum and cerebellum. The average three-dimensional distance value for the nine subjects (with the 10th as the reference) was 0.2 mm for the corpus callosum and 1.2 mm for the cerebellum. CONCLUSION: A high-resolution, diffusion MR atlas with full brain coverage was developed. Additionally, maps of the SD of the diffusion indices were also generated to provide an estimate of the variance within a normal population. Active shape and texture models were also generated for the corpus callosum as an alternate method of representing the variance in morphology and diffusion indices.

Quantitative metrics for evaluating parallel acquisition techniques in diffusion tensor imaging at 3 Tesla.

OBJECTIVES: Single-shot echo-planar based diffusion tensor imaging is prone to geometric and intensity distortions. Parallel imaging is a means of reducing these distortions while preserving spatial resolution. A quantitative comparison at 3 T of parallel imaging for diffusion tensor images (DTI) using k-space (generalized auto-calibrating partially parallel acquisitions; GRAPPA) and image domain (sensitivity encoding; SENSE) reconstructions at different acceleration factors, R, is reported here. MATERIALS AND METHODS: Images were evaluated using 8 human subjects with repeated scans for 2 subjects to estimate reproducibility. Mutual information (MI) was used to assess the global changes in geometric distortions. The effects of parallel imaging techniques on random noise and reconstruction artifacts were evaluated by placing 26 regions of interest and computing the standard deviation of apparent diffusion coefficient and fractional anisotropy along with the error of fitting the data to the diffusion model (residual error). RESULTS: The larger positive values in mutual information index with increasing R values confirmed the anticipated decrease in distortions. Further, the MI index of GRAPPA sequences for a given R factor was larger than the corresponding mSENSE images. The residual error was lowest in the images acquired without parallel imaging and among the parallel reconstruction methods, the R = 2 acquisitions had the least error. The standard deviation, accuracy, and reproducibility of the apparent diffusion coefficient and fractional anisotropy in homogenous tissue regions showed that GRAPPA acquired with R = 2 had the least amount of systematic and random noise and of these, significant differences with mSENSE, R = 2 were found only for the fractional anisotropy index. CONCLUSION: Evaluation of the current implementation of parallel reconstruction algorithms identified GRAPPA acquired with R = 2 as optimal for diffusion tensor imaging.

Geometric distortion correction of high-resolution 3 T diffusion tensor brain images.

Diffusion-weighted images based on echo planar sequences suffer from distortions due to field inhomogeneities from susceptibility differences as well as from eddy currents arising from diffusion gradients. In this paper, a novel approach using nonlinear warping based on optic flow to correct distortions of baseline and diffusion weighted echo planar images (EPI) acquired at 3 T is presented. The distortion correction was estimated by warping the echo planar images to the anatomically correct T2-weighted fast spin echo images (T2-FSE). A global histogram intensity matching of the T2-FSE precedes the base line EPI image distortion correction. A local intensity-matching algorithm was used to transform labeled T2-FSE regions to match intensities of diffusion-weighted EPI images prior to distortion correction of these images. Evaluation was performed using three methods: (i) visual comparison of overlaid contours, (ii) a global mutual information index, and (iii) a local distance measure between homologous points. Visual assessment and the global index demonstrated a decrease in geometrical distortion and the distance measure showed that distortions are reduced to a subvoxel level. In conclusion, the warping algorithm is effective in reducing geometric distortions, enabling generation of anatomically correct diffusion tensor images at 3 T.

Parametric brain MR atlases: standardization for imaging informatics.

This paper is focused on the development of normal MR brain atlases of intrinsic MR parameters. These parameters permit quantitative comparisons across imaging studies (as opposed to raw image intensity values) and are important markers of neurological diseases. The development includes fast sequences to generate three parameters (T1: spin-lattice relaxation, T2: spin-spin relaxation, and Diffusion Tensor) covering the whole brain with isotropic and high-resolution images. The analysis of raw data to generate the parametric images is followed by registration algorithms to bring the image studies acquired on normal subjects aligned to a common frame of reference. The registration method includes both linear and non-linear algorithms. Two atlas schemes are discussed: an average atlas and a probabilistic atlas. Initial results on sequence development and registration are presented. The atlases are envisaged as an integral part of an imaging informatics infrastructure that enables image analysis across imaging studies to perform automated image data mining.