Deep learning-based segmentation methods provide an effective and automated way for assessing the structure and function of the heart in cardiac magnetic resonance (CMR) images. However, despite their state-of-the-art performance on images acquired from the same source (same scanner or scanner vendor) as images used during training, their performance degrades significantly on images coming from different domains. A straightforward approach to tackle this issue consists of acquiring large quantities of multi-site and multi-vendor data, which is practically infeasible. Generative adversarial networks (GANs) for image synthesis present a promising solution for tackling data limitations in medical imaging and addressing the generalization capability of segmentation models. In this work, we explore the usability of synthesized short-axis CMR images generated using a segmentation-informed conditional GAN, to improve the robustness of heart cavity segmentation models in a variety of different settings. The GAN is trained on paired real images and corresponding segmentation maps belonging to both the heart and the surrounding tissue, reinforcing the synthesis of semantically-consistent and realistic images. First, we evaluate the segmentation performance of a model trained solely with synthetic data and show that it only slightly underperforms compared to the baseline trained with real data. By further combining real with synthetic data during training, we observe a substantial improvement in segmentation performance (up to 4% and 40% in terms of Dice score and Hausdorff distance) across multiple data-sets collected from various sites and scanner. This is additionally demonstrated across state-of-the-art 2D and 3D segmentation networks, whereby the obtained results demonstrate the potential of the proposed method in tackling the presence of the domain shift in medical data. Finally, we thoroughly analyze the quality of synthetic data and its ability to replace real MR images during training, as well as provide an insight into important aspects of utilizing synthetic images for segmentation.

One of the limiting factors for the development and adoption of novel deep-learning (DL) based medical image analysis methods is the scarcity of labeled medical images. Medical image simulation and synthesis can provide solutions by generating ample training data with corresponding ground truth labels. Despite recent advances, generated images demonstrate limited realism and diversity. In this work, we develop a flexible framework for simulating cardiac magnetic resonance (MR) images with variable anatomical and imaging characteristics for the purpose of creating a diversified virtual population. We advance previous works on both cardiac MR image simulation and anatomical modeling to increase the realism in terms of both image appearance and underlying anatomy. To diversify the generated images, we define parameters: 1)to alter the anatomy, 2) to assign MR tissue properties to various tissue types, and 3) to manipulate the image contrast via acquisition parameters. The proposed framework is optimized to generate a substantial number of cardiac MR images with ground truth labels suitable for downstream supervised tasks. A database of virtual subjects is simulated and its usefulness for aiding a DL segmentation method is evaluated. Our experiments show that training completely with simulated images can perform comparable with a model trained with real images for heart cavity segmentation in mid-ventricular slices. Moreover, such data can be used in addition to classical augmentation for boosting the performance when training data is limited, particularly by increasing the contrast and anatomical variation, leading to better regularization and generalization. The database is publicly available at https://osf.io/bkzhm/ and the simulation code will be available at https://github.com/sinaamirrajab/CMRI.

Quantification of cardiac deformation and strain with 3D ultrasound takes considerable research efforts. Nevertheless, a widespread use of these techniques in clinical practice is still held back due to the lack of a solid verification process to quantify and compare performance. In this context, the use of fully synthetic sequences has become an established tool for initial in silico evaluation. Nevertheless, the realism of existing simulation techniques is still too limited to represent reliable benchmarking data. Moreover, the fact that different centers typically make use of in-house developed simulation pipelines makes a fair comparison difficult. In this context, this paper introduces a novel pipeline for the generation of synthetic 3D cardiac ultrasound image sequences. State-of-the art solutions in the fields of electromechanical modeling and ultrasound simulation are combined within an original framework that exploits a real ultrasound recording to learn and simulate realistic speckle textures. The simulated images show typical artifacts that make motion tracking in ultrasound challenging. The ground-truth displacement field is available voxelwise and is fully controlled by the electromechanical model. By progressively modifying mechanical and ultrasound parameters, the sensitivity of 3D strain algorithms to pathology and image properties can be evaluated. The proposed pipeline is used to generate an initial library of 8 sequences including healthy and pathological cases, which is made freely accessible to the research community via our project web-page.

Knowledge of left atrial (LA) anatomy is important for atrial fibrillation ablation guidance, fibrosis quantification and biophysical modelling. Segmentation of the LA from Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) images is a complex problem. This manuscript presents a benchmark to evaluate algorithms that address LA segmentation. The datasets, ground truth and evaluation code have been made publicly available through the http://www.cardiacatlas.org website. This manuscript also reports the results of the Left Atrial Segmentation Challenge (LASC) carried out at the STACOM'13 workshop, in conjunction with MICCAI'13. Thirty CT and 30 MRI datasets were provided to participants for segmentation. Each participant segmented the LA including a short part of the LA appendage trunk and proximal sections of the pulmonary veins (PVs). We present results for nine algorithms for CT and eight algorithms for MRI. Results showed that methodologies combining statistical models with region growing approaches were the most appropriate to handle the proposed task. The ground truth and automatic segmentations were standardised to reduce the influence of inconsistently defined regions (e.g., mitral plane, PVs end points, LA appendage). This standardisation framework, which is a contribution of this work, can be used to label and further analyse anatomical regions of the LA. By performing the standardisation directly on the left atrial surface, we can process multiple input data, including meshes exported from different electroanatomical mapping systems.

Since the introduction of 3-D rotational X-ray imaging, protocols for 3-D rotational coronary artery imaging have become widely available in routine clinical practice. Intra-procedural cardiac imaging in a computed tomography (CT)-like fashion has been particularly compelling due to the reduction of clinical overhead and ability to characterize anatomy at the time of intervention. We previously introduced a clinically feasible approach for imaging the left atrium and pulmonary veins (LAPVs) with short contrast bolus injections and scan times of approximately 4 -10 s. The resulting data have sufficient image quality for intra-procedural use during electro-anatomic mapping (EAM) and interventional guidance in atrial fibrillation (AF) ablation procedures. In this paper, we present a novel technique to intra-procedural surface generation which integrates fully-automated segmentation of the LAPVs for guidance in AF ablation interventions. Contrast-enhanced rotational X-ray angiography (3-D RA) acquisitions in combination with filtered-back-projection-based reconstruction allows for volumetric interrogation of LAPV anatomy in near-real-time. An automatic model-based segmentation algorithm allows for fast and accurate LAPV mesh generation despite the challenges posed by image quality; relative to pre-procedural cardiac CT/MR, 3-D RA images suffer from more artifacts and reduced signal-to-noise. We validate our integrated method by comparing 1) automatic and manual segmentations of intra-procedural 3-D RA data, 2) automatic segmentations of intra-procedural 3-D RA and pre-procedural CT/MR data, and 3) intra-procedural EAM point cloud data with automatic segmentations of 3-D RA and CT/MR data. Our validation results for automatically segmented intra-procedural 3-D RA data show average segmentation errors of 1) approximately 1.3 mm compared with manual 3-D RA segmentations 2) approximately 2.3 mm compared with automatic segmentation of pre-procedural CT/MR data and 3) approximately 2.1 mm compared with registered intra-procedural EAM point clouds. The overall experiments indicate that LAPV surfaces can be automatically segmented intra-procedurally from 3-D RA data with comparable quality relative to meshes derived from pre-procedural CT/MR.

Automatic image processing methods are a prerequisite to efficiently analyze the large amount of image data produced by computed tomography (CT) scanners during cardiac exams. This paper introduces a model-based approach for the fully automatic segmentation of the whole heart (four chambers, myocardium, and great vessels) from 3-D CT images. Model adaptation is done by progressively increasing the degrees-of-freedom of the allowed deformations. This improves convergence as well as segmentation accuracy. The heart is first localized in the image using a 3-D implementation of the generalized Hough transform. Pose misalignment is corrected by matching the model to the image making use of a global similarity transformation. The complex initialization of the multicompartment mesh is then addressed by assigning an affine transformation to each anatomical region of the model. Finally, a deformable adaptation is performed to accurately match the boundaries of the patient's anatomy. A mean surface-to-surface error of 0.82 mm was measured in a leave-one-out quantitative validation carried out on 28 images. Moreover, the piecewise affine transformation introduced for mesh initialization and adaptation shows better interphase and interpatient shape variability characterization than commonly used principal component analysis.

Geometrical and physical modelling is of increasing importance in medical image processing. Geometric models are typically used in segmentation and registration methods to support image visualization and the extraction of measurements from images. Increasingly more, image derived geometrical models are the input for a more refined analysis using physical models. An example is the analysis of blood flow patterns in the heart or in aneurysms with CFD simulations.

Maximum intensity projections (MIPs) are an important visualization technique for angiographic data sets. Efficient data inspection requires frame rates of at least five frames per second at preserved image quality. Despite the advances in computer technology, this task remains a challenge. On the one hand, the sizes of computed tomography and magnetic resonance images are increasing rapidly. On the other hand, rendering algorithms do not automatically benefit from the advances in processor technology, especially for large data sets. This is due to the faster evolving processing power and the slower evolving memory access speed, which is bridged by hierarchical cache memory architectures. In this paper, we investigate memory access optimization methods and use them for generating MIPs on general-purpose central processing units (CPUs) and graphics processing units (GPUs), respectively. These methods can work on any level of the memory hierarchy, and we show that properly combined methods can optimize memory access on multiple levels of the hierarchy at the same time. We present performance measurements to compare different algorithm variants and illustrate the influence of the respective techniques. On current hardware, the efficient handling of the memory hierarchy for CPUs improves the rendering performance by a factor of 3 to 4. On GPUs, we observed that the effect is even larger, especially for large data sets. The methods can easily be adjusted to different hardware specifics, although their impact can vary considerably. They can also be used for other rendering techniques than MIPs, and their use for more general image processing task could be investigated in the future.

In recent years, several methods have been proposed for constructing statistical shape models to aid image analysis tasks by providing a priori knowledge. Examples include principal component analysis of manually or semiautomatically placed corresponding landmarks on the learning shapes [point distribution models (PDMs)], which is time consuming and subjective. However, automatically establishing surface correspondences continues to be a difficult problem. This paper presents a novel method for the automated construction of three-dimensional PDM from segmented images. Corresponding surface landmarks are established by adapting a triangulated learning shape to segmented volumetric images of the remaining shapes. The adaptation is based on a novel deformable model technique. We illustrate our approach using computed tomography data of the vertebra and the femur. We demonstrate that our method accurately represents and predicts shapes.

We consider elastic image registration based on a set of corresponding anatomical point landmarks and approximating thin-plate splines. This approach is an extension of the original interpolating thin-plate spline approach and allows to take into account landmark localization errors. The extension is important for clinical applications since landmark extraction is always prone to error. Our approach is based on a minimizing functional and can cope with isotropic as well as anisotropic landmark errors. In particular, in the latter case it is possible to include different types of landmarks, e.g., unique point landmarks as well as arbitrary edge points. Also, the scheme is general with respect to the image dimension and the order of smoothness of the underlying functional. Optimal affine transformations as well as interpolating thin-plate splines are special cases of this scheme. To localize landmarks we use a semi-automatic approach which is based on three-dimensional (3-D) differential operators. Experimental results are presented for two-dimensional as well as 3-D tomographic images of the human brain.

In recent years, advances in computer technology and a significant increase in the accuracy of medical imaging have made it possible to develop systems that can assist the clinician in diagnosis, planning, and treatment. This paper deals with an area that is generally referred to as computer-assisted surgery, image-directed surgery, or image-guided surgery. We report the research, development, and clinical validation performed since January 1996 in the European Applications in Surgical Interventions (EASI) project, which is funded by the European Commission in their "4th Framework Telematics Applications for Health" program. The goal of this project is the improvement of the effectiveness and quality of image-guided neurosurgery of the brain and image-guided vascular surgery of abdominal aortic aneurysms, while at the same time reducing patient risks and overall cost. We have developed advanced prototype systems for preoperative surgical planning and intraoperative surgical navigation, and we have extensively clinically validated these systems. The prototype systems and the clinical validation results are described in this paper.

A comparison of six similarity measures for use in intensity-based two-dimensional-three-dimensional (2-D-3-D) image registration is presented. The accuracy of the similarity measures are compared to a "gold-standard" registration which has been accurately calculated using fiducial markers. The similarity measures are used to register a computed tomography (CT) scan of a spine phantom to a fluoroscopy image of the phantom. The registration is carried out within a region-of-interest in the fluoroscopy image which is user defined to contain a single vertebra. Many of the problems involved in this type of registration are caused by features which were not modeled by a phantom image alone. More realistic "gold-standard" data sets were simulated using the phantom image with clinical image features overlaid. Results show that the introduction of soft-tissue structures and interventional instruments into the phantom image can have a large effect on the performance of some similarity measures previously applied to 2-D-3-D image registration. Two measures were able to register accurately and robustly even when soft-tissue structures and interventional instruments were present as differences between the images. These measures were pattern intensity and gradient difference. Their registration accuracy, for all the rigid-body parameters except for the source to film translation, was within a root-mean-square (rms) error of 0.54 mm or degrees to the "gold-standard" values. No failures occurred while registering using these measures.

Registration of intraoperative fluoroscopy images with preoperative three-dimensional (3-D) CT images can be used for several purposes in image-guided surgery. On the one hand, it can be used to display the position of surgical instruments, which are being tracked by a localizer, in the preoperative CT scan. On the other hand, the registration result can be used to project preoperative planning information or important anatomical structures visible in the CT image onto the fluoroscopy image. For this registration task, a novel voxel-based method in combination with a new similarity measure (pattern intensity) has been developed. The basic concept of the method is explained at the example of two-dimensional (2-D)/3-D registration of a vertebra in an X-ray fluoroscopy image with a 3-D CT image. The registration method is described, and the results for a spine phantom are presented and discussed. Registration has been carried out repeatedly with different starting estimates to study the capture range. Information about registration accuracy has been obtained by comparing the registration results with a highly accurate "ground-truth" registration, which has been derived from fiducial markers attached to the phantom prior to imaging. In addition, registration results for different vertebrae have been compared. The results show that the rotation parameters and the shifts parallel to the projection plane can accurately be determined from a single projection. Because of the projection geometry, the accuracy of the height above the projection plane is significantly lower.