Metric learning with spectral graph convolutions on brain connectivity networks.

Graph representations are often used to model structured data at an individual or population level and have numerous applications in pattern recognition problems. In the field of neuroscience, where such representations are commonly used to model structural or functional connectivity between a set of brain regions, graphs have proven to be of great importance. This is mainly due to the capability of revealing patterns related to brain development and disease, which were previously unknown. Evaluating similarity between these brain connectivity networks in a manner that accounts for the graph structure and is tailored for a particular application is, however, non-trivial. Most existing methods fail to accommodate the graph structure, discarding information that could be beneficial for further classification or regression analyses based on these similarities. We propose to learn a graph similarity metric using a siamese graph convolutional neural network (s-GCN) in a supervised setting. The proposed framework takes into consideration the graph structure for the evaluation of similarity between a pair of graphs, by employing spectral graph convolutions that allow the generalisation of traditional convolutions to irregular graphs and operates in the graph spectral domain. We apply the proposed model on two datasets: the challenging ABIDE database, which comprises functional MRI data of 403 patients with autism spectrum disorder (ASD) and 468 healthy controls aggregated from multiple acquisition sites, and a set of 2500 subjects from UK Biobank. We demonstrate the performance of the method for the tasks of classification between matching and non-matching graphs, as well as individual subject classification and manifold learning, showing that it leads to significantly improved results compared to traditional methods.

Automated cardiovascular magnetic resonance image analysis with fully convolutional networks.

BACKGROUND: Cardiovascular resonance (CMR) imaging is a standard imaging modality for assessing cardiovascular diseases (CVDs), the leading cause of death globally. CMR enables accurate quantification of the cardiac chamber volume, ejection fraction and myocardial mass, providing information for diagnosis and monitoring of CVDs. However, for years, clinicians have been relying on manual approaches for CMR image analysis, which is time consuming and prone to subjective errors. It is a major clinical challenge to automatically derive quantitative and clinically relevant information from CMR images. METHODS: Deep neural networks have shown a great potential in image pattern recognition and segmentation for a variety of tasks. Here we demonstrate an automated analysis method for CMR images, which is based on a fully convolutional network (FCN). The network is trained and evaluated on a large-scale dataset from the UK Biobank, consisting of 4,875 subjects with 93,500 pixelwise annotated images. The performance of the method has been evaluated using a number of technical metrics, including the Dice metric, mean contour distance and Hausdorff distance, as well as clinically relevant measures, including left ventricle (LV) end-diastolic volume (LVEDV) and end-systolic volume (LVESV), LV mass (LVM); right ventricle (RV) end-diastolic volume (RVEDV) and end-systolic volume (RVESV). RESULTS: By combining FCN with a large-scale annotated dataset, the proposed automated method achieves a high performance in segmenting the LV and RV on short-axis CMR images and the left atrium (LA) and right atrium (RA) on long-axis CMR images. On a short-axis image test set of 600 subjects, it achieves an average Dice metric of 0.94 for the LV cavity, 0.88 for the LV myocardium and 0.90 for the RV cavity. The mean absolute difference between automated measurement and manual measurement is 6.1 mL for LVEDV, 5.3 mL for LVESV, 6.9 gram for LVM, 8.5 mL for RVEDV and 7.2 mL for RVESV. On long-axis image test sets, the average Dice metric is 0.93 for the LA cavity (2-chamber view), 0.95 for the LA cavity (4-chamber view) and 0.96 for the RA cavity (4-chamber view). The performance is comparable to human inter-observer variability. CONCLUSIONS: We show that an automated method achieves a performance on par with human experts in analysing CMR images and deriving clinically relevant measures.

3D MR ventricle segmentation in pre-term infants with post-hemorrhagic ventricle dilatation (PHVD) using multi-phase geodesic level-sets.

Intraventricular hemorrhage (IVH) or bleed within the cerebral ventricles is a common condition among very low birth weight pre-term neonates. The prognosis for these patients is worsened should they develop progressive ventricular dilatation, i.e., post-hemorrhagic ventricle dilatation (PHVD), which occurs in 10-30% of IVH patients. Accurate measurement of ventricular volume would be valuable information and could be used to predict PHVD and determine whether that specific patient with ventricular dilatation requires treatment. While the monitoring of PHVD in infants is typically done by repeated transfontanell 2D ultrasound (US) and not MRI, once the patient's fontanels have closed around 12-18months of life, the follow-up patient scans are done by MRI. Manual segmentation of ventricles from MR images is still seen as a gold standard. However, it is extremely time- and labor-consuming, and it also has observer variability. This paper proposes an accurate multiphase geodesic level-set segmentation algorithm for the extraction of the cerebral ventricle system of pre-term PHVD neonates from 3D T1 weighted MR images. The proposed segmentation algorithm makes use of multi-region segmentation technique associated with spatial priors built from a multi-atlas registration scheme. The leave-one-out cross validation with 19 patients with mild enlargement of ventricles and 7 hydrocephalus patients shows that the proposed method is accurate, suggesting that the proposed approach could be potentially used for volumetric and morphological analysis of the ventricle system of IVH neonatal brains in clinical practice.

High-resolution 3-dimensional late gadolinium enhancement scar imaging in surgically corrected Tetralogy of Fallot: clinical feasibility of volumetric quantification and visualization.

BACKGROUND: The extent of surgical scarring in Tetralogy of Fallot (TOF) may be a marker of adverse outcomes and provide substrate for ventricular arrhythmia. In this study we evaluate the feasibility of high resolution three dimensional (3D) late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) for volumetric scar quantification in patients with surgically corrected TOF. METHODS: Fifteen consecutive patients underwent 3D LGE imaging with 3 Tesla CMR using a whole-heart, respiratory-navigated technique. A novel, signal-histogram based segmentation technique was tested for the quantification and modeling of surgical scar. Total scar volume was compared to the gold standard manual expert segmentation. The feasibility of segmented scar fusion to matched coronary CMR data for volumetric display was explored. RESULTS: Image quality sufficient for 3D scar segmentation was acquired in fourteen patients. Mean patient age was 32.2 ± 11.9 years (range 21 to 57 years) with mean right ventricle (RV) ejection fraction (EF) of 53.9 ± 9.2% and mean RV end diastolic volume of 117.0 ± 41.5 mL/m². The mean total scar volume was 11.1 ± 8.2 mL using semi-automated 3D segmentation with excellent correlation to manual expert segmentation (r = 0.99, bias = 0.89 mL, 95% CI -1.66 to 3.44). The mean segmentation time was significantly reduced using the novel semi-automated segmentation technique (10.1 ± 2.6 versus 45.8 ± 12.6 minutes). Excellent intra-observer and good inter-observer reproducibility was observed. CONCLUSION: 3D high resolution LGE imaging with semi-automated scar segmentation is clinically feasible among patients with surgically corrected TOF and shows excellent accuracy and reproducibility. This approach may offer a valuable clinical tool for risk prediction and procedural planning among this growing population.

Accuracy and reproducibility of semi-automated late gadolinium enhancement quantification techniques in patients with hypertrophic cardiomyopathy.

BACKGROUND: The presence and extent of late gadolinium enhancement (LGE) has been associated with adverse events in patients with hypertrophic cardiomyopathy (HCM). Signal intensity (SI) threshold techniques are routinely employed for quantification; Full-Width at Half-Maximum (FWHM) techniques are suggested to provide greater reproducibility than Signal Threshold versus Reference Mean (STRM) techniques, however the accuracy of these approaches versus the manual assignment of optimal SI thresholds has not been studied. In this study, we compared all known semi-automated LGE quantification techniques for accuracy and reproducibility among patients with HCM. METHODS: Seventy-six HCM patients (51 male, age 54 ± 13 years) were studied. Total LGE volume was quantified using 7 semi-automated techniques and compared to expert manual adjustment of the SI threshold to achieve optimal segmentation. Techniques tested included STRM based thresholds of >2, 3, 4, 5 and 6 SD above mean SI of reference myocardium, the FWHM technique, and the Otsu-auto-threshold (OAT) technique. The SI threshold chosen by each technique was recorded for all slices. Bland-Altman analysis and intra-class correlation coefficients (ICC) were reported for each semi-automated technique versus expert, manually adjusted LGE segmentation. Intra- and inter-observer reproducibility assessments were also performed. RESULTS: Fifty-two of 76 (68%) patients showed LGE on a total of 202 slices. For accuracy, the STRM >3SD technique showed the greatest agreement with manual segmentation (ICC = 0.97, mean difference and 95% limits of agreement = 1.6 ± 10.7 g) while STRM >6SD, >5SD, 4SD and FWHM techniques systematically underestimated total LGE volume. Slice based analysis of selected SI thresholds similarly showed the STRM >3SD threshold to most closely approximate manually adjusted SI thresholds (ICC = 0.88). For reproducibility, the intra- and inter-observer reproducibility of the >3SD threshold demonstrated an acceptable mean difference and 95% limits of agreement of -0.5 ± 6.8 g and -0.9 ± 5.6 g, respectively. CONCLUSIONS: FWHM segmentation provides superior reproducibility, however systematically underestimates total LGE volume compared to manual segmentation in patients with HCM. The STRM >3SD technique provides the greatest accuracy while retaining acceptable reproducibility and may therefore be a preferred approach for LGE quantification in this population.

Deep learning links localized digital pathology phenotypes with transcriptional subtype and patient outcome in glioblastoma.

BACKGROUND: Deep learning has revolutionized medical image analysis in cancer pathology, where it had a substantial clinical impact by supporting the diagnosis and prognostic rating of cancer. Among the first available digital resources in the field of brain cancer is glioblastoma, the most common and fatal brain cancer. At the histologic level, glioblastoma is characterized by abundant phenotypic variability that is poorly linked with patient prognosis. At the transcriptional level, 3 molecular subtypes are distinguished with mesenchymal-subtype tumors being associated with increased immune cell infiltration and worse outcome. RESULTS: We address genotype-phenotype correlations by applying an Xception convolutional neural network to a discovery set of 276 digital hematozylin and eosin (H&E) slides with molecular subtype annotation and an independent The Cancer Genome Atlas-based validation cohort of 178 cases. Using this approach, we achieve high accuracy in H&E-based mapping of molecular subtypes (area under the curve for classical, mesenchymal, and proneural = 0.84, 0.81, and 0.71, respectively; P < 0.001) and regions associated with worse outcome (univariable survival model P < 0.001, multivariable P = 0.01). The latter were characterized by higher tumor cell density (P < 0.001), phenotypic variability of tumor cells (P < 0.001), and decreased T-cell infiltration (P = 0.017). CONCLUSIONS: We modify a well-known convolutional neural network architecture for glioblastoma digital slides to accurately map the spatial distribution of transcriptional subtypes and regions predictive of worse outcome, thereby showcasing the relevance of artificial intelligence-enabled image mining in brain cancer.

Fully automated segmentation of left ventricular scar from 3D late gadolinium enhancement magnetic resonance imaging using a cascaded multi-planar U-Net (CMPU-Net).

PURPOSE: Three-dimensional (3D) late gadolinium enhancement magnetic resonance (LGE-MR) imaging enables the quantification of myocardial scar at high resolution with unprecedented volumetric visualization. Automated segmentation of myocardial scar is critical for the potential clinical translation of this technique given the number of tomographic images acquired. METHODS: In this paper, we describe the development of cascaded multi-planar U-Net (CMPU-Net) to efficiently segment the boundary of the left ventricle (LV) myocardium and scar from 3D LGE-MR images. In this approach, two subnets, each containing three U-Nets, were cascaded to first segment the LV myocardium and then segment the scar within the presegmented LV myocardium. The U-Nets were trained separately using two-dimensional (2D) slices extracted from axial, sagittal, and coronal slices of 3D LGE-MR images. We used 3D LGE-MR images from 34 subjects with chronic ischemic cardiomyopathy. The U-Nets were trained using 8430 slices, extracted in three orthogonal directions from 18 images. In the testing phase, the outputs of U-Nets of each subnet were combined using the majority voting system for final label prediction of each voxel in the image. The developed method was tested for accuracy by comparing its results to manual segmentations of LV myocardium and LV scar from 7250 slices extracted from 16 3D LGE-MR images. Our method was also compared to numerous alternative methods based on machine learning, energy minimization, and intensity-thresholds. RESULTS: Our algorithm reported a mean dice similarity coefficient (DSC), absolute volume difference (AVD), and Hausdorff distance (HD) of 85.14% ± 3.36%, 43.72 ± 27.18 cm3 , and 19.21 ± 4.74 mm for determining the boundaries of LV myocardium from LGE-MR images. Our method also yielded a mean DSC, AVD, and HD of 88.61% ± 2.54%, 9.33 ± 7.24 cm3 , and 17.04 ± 9.93 mm for LV scar segmentation on the unobserved test dataset. Our method significantly outperformed the alternative techniques in segmentation accuracy (P < 0.05). CONCLUSIONS: The CMPU-Net method provided fully automated segmentation of LV scar from 3D LGE-MR images and outperformed the alternative techniques.

Weakly Supervised Estimation of Shadow Confidence Maps in Fetal Ultrasound Imaging.

Detecting acoustic shadows in ultrasound images is important in many clinical and engineering applications. Real-time feedback of acoustic shadows can guide sonographers to a standardized diagnostic viewing plane with minimal artifacts and can provide additional information for other automatic image analysis algorithms. However, automatically detecting shadow regions using learning-based algorithms is challenging because pixel-wise ground truth annotation of acoustic shadows is subjective and time consuming. In this paper, we propose a weakly supervised method for automatic confidence estimation of acoustic shadow regions. Our method is able to generate a dense shadow-focused confidence map. In our method, a shadow-seg module is built to learn general shadow features for shadow segmentation, based on global image-level annotations as well as a small number of coarse pixel-wise shadow annotations. A transfer function is introduced to extend the obtained binary shadow segmentation to a reference confidence map. In addition, a confidence estimation network is proposed to learn the mapping between input images and the reference confidence maps. This network is able to predict shadow confidence maps directly from input images during inference. We use evaluation metrics such as DICE, inter-class correlation, and so on, to verify the effectiveness of our method. Our method is more consistent than human annotation and outperforms the state-of-the-art quantitatively in shadow segmentation and qualitatively in confidence estimation of shadow regions. Furthermore, we demonstrate the applicability of our method by integrating shadow confidence maps into tasks such as ultrasound image classification, multi-view image fusion, and automated biometric measurements.

Multi-Atlas Segmentation Using Partially Annotated Data: Methods and Annotation Strategies.

Multi-atlas segmentation is a widely used tool in medical image analysis, providing robust and accurate results by learning from annotated atlas datasets. However, the availability of fully annotated atlas images for training is limited due to the time required for the labelling task. Segmentation methods requiring only a proportion of each atlas image to be labelled could therefore reduce the workload on expert raters tasked with annotating atlas images. To address this issue, we first re-examine the labelling problem common in many existing approaches and formulate its solution in terms of a Markov Random Field energy minimisation problem on a graph connecting atlases and the target image. This provides a unifying framework for multi-atlas segmentation. We then show how modifications in the graph configuration of the proposed framework enable the use of partially annotated atlas images and investigate different partial annotation strategies. The proposed method was evaluated on two Magnetic Resonance Imaging (MRI) datasets for hippocampal and cardiac segmentation. Experiments were performed aimed at (1) recreating existing segmentation techniques with the proposed framework and (2) demonstrating the potential of employing sparsely annotated atlas data for multi-atlas segmentation.

Stratified Decision Forests for Accurate Anatomical Landmark Localization in Cardiac Images.

Accurate localization of anatomical landmarks is an important step in medical imaging, as it provides useful prior information for subsequent image analysis and acquisition methods. It is particularly useful for initialization of automatic image analysis tools (e.g. segmentation and registration) and detection of scan planes for automated image acquisition. Landmark localization has been commonly performed using learning based approaches, such as classifier and/or regressor models. However, trained models may not generalize well in heterogeneous datasets when the images contain large differences due to size, pose and shape variations of organs. To learn more data-adaptive and patient specific models, we propose a novel stratification based training model, and demonstrate its use in a decision forest. The proposed approach does not require any additional training information compared to the standard model training procedure and can be easily integrated into any decision tree framework. The proposed method is evaluated on 1080 3D high-resolution and 90 multi-stack 2D cardiac cine MR images. The experiments show that the proposed method achieves state-of-the-art landmark localization accuracy and outperforms standard regression and classification based approaches. Additionally, the proposed method is used in a multi-atlas segmentation to create a fully automatic segmentation pipeline, and the results show that it achieves state-of-the-art segmentation accuracy.

DeepCut: Object Segmentation From Bounding Box Annotations Using Convolutional Neural Networks.

In this paper, we propose DeepCut, a method to obtain pixelwise object segmentations given an image dataset labelled weak annotations, in our case bounding boxes. It extends the approach of the well-known GrabCut [1] method to include machine learning by training a neural network classifier from bounding box annotations. We formulate the problem as an energy minimisation problem over a densely-connected conditional random field and iteratively update the training targets to obtain pixelwise object segmentations. Additionally, we propose variants of the DeepCut method and compare those to a naïve approach to CNN training under weak supervision. We test its applicability to solve brain and lung segmentation problems on a challenging fetal magnetic resonance dataset and obtain encouraging results in terms of accuracy.

PVR: Patch-to-Volume Reconstruction for Large Area Motion Correction of Fetal MRI.

In this paper, we present a novel method for the correction of motion artifacts that are present in fetal magnetic resonance imaging (MRI) scans of the whole uterus. Contrary to current slice-to-volume registration (SVR) methods, requiring an inflexible anatomical enclosure of a single investigated organ, the proposed patch-to-volume reconstruction (PVR) approach is able to reconstruct a large field of view of non-rigidly deforming structures. It relaxes rigid motion assumptions by introducing a specific amount of redundant information that is exploited with parallelized patchwise optimization, super-resolution, and automatic outlier rejection. We further describe and provide an efficient parallel implementation of PVR allowing its execution within reasonable time on commercially available graphics processing units, enabling its use in the clinical practice. We evaluate PVR's computational overhead compared with standard methods and observe improved reconstruction accuracy in the presence of affine motion artifacts compared with conventional SVR in synthetic experiments. Furthermore, we have evaluated our method qualitatively and quantitatively on real fetal MRI data subject to maternal breathing and sudden fetal movements. We evaluate peak-signal-to-noise ratio, structural similarity index, and cross correlation with respect to the originally acquired data and provide a method for visual inspection of reconstruction uncertainty. We further evaluate the distance error for selected anatomical landmarks in the fetal head, as well as calculating the mean and maximum displacements resulting from automatic non-rigid registration to a motion-free ground truth image. These experiments demonstrate a successful application of PVR motion compensation to the whole fetal body, uterus, and placenta.

Contribution of mitral valve leaflet length and septal wall thickness to outflow tract obstruction in patients with hypertrophic cardiomyopathy.

We sought to examine whether elongation of the mitral valve leaflets in patients with hypertrophic cardiomyopathy (HCM) is synergistic to septal wall thickness (SWT) in the development of left ventricular outflow tract obstruction (LVOTO). HCM is a common genetic cardiac disease characterized by asymmetric septal hypertrophy and predisposition towards LVOTO. It has been reported that elongation of the mitral valve leaflets may be a primary phenotypic feature and contribute to LVOTO. However, the relative contribution of this finding versus SWT has not been studied. 152 patients (76 with HCM and 76 non-diseased age, race and BSA-matched controls) and 18 young, healthy volunteers were studied. SWT and the anterior mitral valve leaflet length (AMVLL) were measured using cine MRI. The combined contribution of these variables (SWT × AMVLL) was described as the Septal Anterior Leaflet Product (SALP). Peak LVOT pressure gradient was determined by Doppler interrogation and defined as "obstructive" if ≥ 30 mmHg. Patients with HCM were confirmed to have increased AMVLL compared with controls and volunteers (p < 0.01). Among HCM patients, both SWT and SALP were significantly higher in patients with LVOTO (N = 17) versus without. SALP showed modest improvement in predictive accuracy for LVOTO (AUC = 0.81) among the HCM population versus SWT alone (AUC = 0.77). However, in isolated patients this variable identified patients with LVOTO despite modest SWT. Elongation of the AMVLL is a primary phenotypic feature of HCM. While incremental contributions to LVOTO appear modest at a population level, specific patients may have dominant contribution to LVOTO. The combined marker of SALP allows for maintained identification of such patients despite modest increases in SWT.

Optimization-based interactive segmentation interface for multiregion problems.

Interactive segmentation is becoming of increasing interest to the medical imaging community in that it combines the positive aspects of both manual and automated segmentation. However, general-purpose tools have been lacking in terms of segmenting multiple regions simultaneously with a high degree of coupling between groups of labels. Hierarchical max-flow segmentation has taken advantage of this coupling for individual applications, but until recently, these algorithms were constrained to a particular hierarchy and could not be considered general-purpose. In a generalized form, the hierarchy for any given segmentation problem is specified in run-time, allowing different hierarchies to be quickly explored. We present an interactive segmentation interface, which uses generalized hierarchical max-flow for optimization-based multiregion segmentation guided by user-defined seeds. Applications in cardiac and neonatal brain segmentation are given as example applications of its generality.

Hierarchical max-flow segmentation framework for multi-atlas segmentation with Kohonen self-organizing map based Gaussian mixture modeling.

The incorporation of intensity, spatial, and topological information into large-scale multi-region segmentation has been a topic of ongoing research in medical image analysis. Multi-region segmentation problems, such as segmentation of brain structures, pose unique challenges in image segmentation in which regions may not have a defined intensity, spatial, or topological distinction, but rely on a combination of the three. We propose a novel framework within the Advanced segmentation tools (ASETS)(2), which combines large-scale Gaussian mixture models trained via Kohonen self-organizing maps, with deformable registration, and a convex max-flow optimization algorithm incorporating region topology as a hierarchy or tree. Our framework is validated on two publicly available neuroimaging datasets, the OASIS and MRBrainS13 databases, against the more conventional Potts model, achieving more accurate segmentations. Each component is accelerated using general-purpose programming on graphics processing Units to ensure computational feasibility.

A mitral annulus tracking approach for navigation of off-pump beating heart mitral valve repair.

PURPOSE: To develop and validate a real-time mitral valve annulus (MVA) tracking approach based on biplane transesophageal echocardiogram (TEE) data and magnetic tracking systems (MTS) to be used in minimally invasive off-pump beating heart mitral valve repair (MVR). METHODS: The authors' guidance system consists of three major components: TEE, magnetic tracking system, and an image guidance software platform. TEE provides real-time intraoperative images to show the cardiac motion and intracardiac surgical tools. The magnetic tracking system tracks the TEE probe and the surgical tools. The software platform integrates the TEE image planes and the virtual model of the tools and the MVA model on the screen. The authors' MVA tracking approach, which aims to update the MVA model in near real-time, comprises of three steps: image based gating, predictive reinitialization, and registration based MVA tracking. The image based gating step uses a small patch centered at each MVA point in the TEE images to identify images at optimal cardiac phases for updating the position of the MVA. The predictive reinitialization step uses the position and orientation of the TEE probe provided by the magnetic tracking system to predict the position of the MVA points in the TEE images and uses them for the initialization of the registration component. The registration based MVA tracking step aims to locate the MVA points in the images selected by the image based gating component by performing image based registration. RESULTS: The validation of the MVA tracking approach was performed in a phantom study and a retrospective study on porcine data. In the phantom study, controlled translations were applied to the phantom and the tracked MVA was compared to its "true" position estimated based on a magnetic sensor attached to the phantom. The MVA tracking accuracy was 1.29 ± 0.58 mm when the translation distance is about 1 cm, and increased to 2.85 ± 1.19 mm when the translation distance is about 3 cm. In the study on porcine data, the authors compared the tracked MVA to a manually segmented MVA. The overall accuracy is 2.37 ± 1.67 mm for single plane images and 2.35 ± 1.55 mm for biplane images. The interoperator variation in manual segmentation was 2.32 ± 1.24 mm for single plane images and 1.73 ± 1.18 mm for biplane images. The computational efficiency of the algorithm on a desktop computer with an Intel(®) Xeon(®) CPU @3.47 GHz and an NVIDIA GeForce 690 graphic card is such that the time required for registering four MVA points was about 60 ms. CONCLUSIONS: The authors developed a rapid MVA tracking algorithm for use in the guidance of off-pump beating heart transapical mitral valve repair. This approach uses 2D biplane TEE images and was tested on a dynamic heart phantom and interventional porcine image data. Results regarding the accuracy and efficiency of the authors' MVA tracking algorithm are promising, and fulfill the requirements for surgical navigation.

Joint segmentation of lumen and outer wall from femoral artery MR images: Towards 3D imaging measurements of peripheral arterial disease.

Three-dimensional (3D) measurements of peripheral arterial disease (PAD) plaque burden extracted from fast black-blood magnetic resonance (MR) images have shown to be more predictive of clinical outcomes than PAD stenosis measurements. To this end, accurate segmentation of the femoral artery lumen and outer wall is required for generating volumetric measurements of PAD plaque burden. Here, we propose a semi-automated algorithm to jointly segment the femoral artery lumen and outer wall surfaces from 3D black-blood MR images, which are reoriented and reconstructed along the medial axis of the femoral artery to obtain improved spatial coherence between slices of the long, thin femoral artery and to reduce computation time. The developed segmentation algorithm enforces two priors in a global optimization manner: the spatial consistency between the adjacent 2D slices and the anatomical region order between the femoral artery lumen and outer wall surfaces. The formulated combinatorial optimization problem for segmentation is solved globally and exactly by means of convex relaxation using a coupled continuous max-flow (CCMF) model, which is a dual formulation to the convex relaxed optimization problem. In addition, the CCMF model directly derives an efficient duality-based algorithm based on the modern multiplier augmented optimization scheme, which has been implemented on a GPU for fast computation. The computed segmentations from the developed algorithm were compared to manual delineations from experts using 20 black-blood MR images. The developed algorithm yielded both high accuracy (Dice similarity coefficients ≥ 87% for both the lumen and outer wall surfaces) and high reproducibility (intra-class correlation coefficient of 0.95 for generating vessel wall area), while outperforming the state-of-the-art method in terms of computational time by a factor of ≈ 20.

Multi-atlas Segmentation as a Graph Labelling Problem: Application to Partially Annotated Atlas Data.

Manually annotating images for multi-atlas segmentation is an expensive and often limiting factor in reliable automated segmentation of large databases. Segmentation methods requiring only a proportion of each atlas image to be labelled could potentially reduce the workload on expert raters tasked with labelling images. However, exploiting such a database of partially labelled atlases is not possible with state-of-the-art multi-atlas segmentation methods. In this paper we revisit the problem of multi-atlas segmentation and formulate its solution in terms of graph-labelling. Our graphical approach uses a Markov Random Field (MRF) formulation of the problem and constructs a graph connecting atlases and the target image. This provides a unifying framework for label propagation. More importantly, the proposed method can be used for segmentation using only partially labelled atlases. We furthermore provide an extension to an existing continuous MRF optimisation method to solve the proposed problem formulation. We show that the proposed method, applied to hippocampal segmentation of 202 subjects from the ADNI database, remains robust and accurate even when the proportion of manually labelled slices in the atlases is reduced to 20%.

Three-Dimensional Nonrigid MR-TRUS Registration Using Dual Optimization.

In this study, we proposed an efficient nonrigid magnetic resonance (MR) to transrectal ultrasound (TRUS) deformable registration method in order to improve the accuracy of targeting suspicious regions during a three dimensional (3-D) TRUS guided prostate biopsy. The proposed deformable registration approach employs the multi-channel modality independent neighborhood descriptor (MIND) as the local similarity feature across the two modalities of MR and TRUS, and a novel and efficient duality-based convex optimization-based algorithmic scheme was introduced to extract the deformations and align the two MIND descriptors. The registration accuracy was evaluated using 20 patient images by calculating the TRE using manually identified corresponding intrinsic fiducials in the whole gland and peripheral zone. Additional performance metrics [Dice similarity coefficient (DSC), mean absolute surface distance (MAD), and maximum absolute surface distance (MAXD)] were also calculated by comparing the MR and TRUS manually segmented prostate surfaces in the registered images. Experimental results showed that the proposed method yielded an overall median TRE of 1.76 mm. The results obtained in terms of DSC showed an average of 80.8±7.8% for the apex of the prostate, 92.0±3.4% for the mid-gland, 81.7±6.4% for the base and 85.7±4.7% for the whole gland. The surface distance calculations showed an overall average of 1.84±0.52 mm for MAD and 6.90±2.07 mm for MAXD.