Impact of training data composition on the generalizability of convolutional neural network aortic cross-section segmentation in four-dimensional magnetic resonance flow imaging.

BACKGROUND: Four-dimensional cardiovascular magnetic resonance flow imaging (4D flow CMR) plays an important role in assessing cardiovascular diseases. However, the manual or semi-automatic segmentation of aortic vessel boundaries in 4D flow data introduces variability and limits the reproducibility of aortic hemodynamics visualization and quantitative flow-related parameter computation. This paper explores the potential of deep learning to improve 4D flow CMR segmentation by developing models for automatic segmentation and analyzes the impact of the training data on the generalization of the model across different sites, scanner vendors, sequences, and pathologies. METHODS: The study population consists of 260 4D flow CMR datasets, including subjects without known aortic pathology, healthy volunteers, and patients with bicuspid aortic valve (BAV) examined at different hospitals. The dataset was split to train segmentation models on subsets with different representations of characteristics, such as pathology, gender, age, scanner model, vendor, and field strength. An enhanced three-dimensional U-net convolutional neural network (CNN) architecture with residual units was trained for time-resolved two-dimensional aortic cross-sectional segmentation. Model performance was evaluated using Dice score, Hausdorff distance, and average symmetric surface distance on test data, datasets with characteristics not represented in the training set (model-specific), and an overall evaluation set. Standard diagnostic flow parameters were computed and compared with manual segmentation results using Bland-Altman analysis and interclass correlation. RESULTS: The representation of technical factors, such as scanner vendor and field strength, in the training dataset had the strongest influence on the overall segmentation performance. Age had a greater impact than gender. Models solely trained on BAV patients' datasets performed well on datasets of healthy subjects but not vice versa. CONCLUSION: This study highlights the importance of considering a heterogeneous dataset for the training of widely applicable automatic CNN segmentations in 4D flow CMR, with a particular focus on the inclusion of different pathologies and technical aspects of data acquisition.

Spatiotemporal phase unwrapping for real-time phase-contrast flow MRI.

PURPOSE: To develop and evaluate a practical phase unwrapping method for real-time phase-contrast flow MRI using temporal and spatial continuity. METHODS: Real-time phase-contrast MRI of through-plane flow was performed using highly undersampled radial FLASH with phase-sensitive reconstructions by regularized nonlinear inversion. Experiments involved flow in a phantom and the human aorta (10 healthy subjects) with and without phase wrapping for velocity encodings of 100 cm·s(-1) and 200 cm·s(-1) . Phase unwrapping was performed for each individual cardiac cycle and restricted to a region of interest automatically propagated to all time frames. The algorithm exploited temporal continuity in forward and backward direction for all pixels with a "continuous" representation of blood throughout the entire cardiac cycle (inner vessel lumen). Phase inconsistencies were corrected by a comparison with values from direct spatial neighbors. The latter approach was also applied to pixels exhibiting a discontinuous signal intensity time course due to movement-induced spatial displacements (peripheral vessel zone). RESULTS: Phantom and human flow MRI data were successfully unwrapped. When halving the velocity encoding, the velocity-to-noise ratio (VNR) increased by a factor of two. CONCLUSION: The proposed phase unwrapping method for real-time flow MRI allows for measurements with reduced velocity encoding and increased VNR.

Learning carotid vessel wall segmentation in black-blood MRI using sparsely sampled cross-sections from 3D data.

Purpose: Atherosclerosis of the carotid artery is a major risk factor for stroke. Quantitative assessment of the carotid vessel wall can be based on cross-sections of three-dimensional (3D) black-blood magnetic resonance imaging (MRI). To increase reproducibility, a reliable automatic segmentation in these cross-sections is essential. Approach: We propose an automatic segmentation of the carotid artery in cross-sections perpendicular to the centerline to make the segmentation invariant to the image plane orientation and allow a correct assessment of the vessel wall thickness (VWT). We trained a residual U-Net on eight sparsely sampled cross-sections per carotid artery and evaluated if the model can segment areas that are not represented in the training data. We used 218 MRI datasets of 121 subjects that show hypertension and plaque in the ICA or CCA measuring ≥ 1.5 mm in ultrasound. Results: The model achieves a high mean Dice coefficient of 0.948/0.859 for the vessel's lumen/wall, a low mean Hausdorff distance of 0.417 / 0.660 mm , and a low mean average contour distance of 0.094 / 0.119 mm on the test set. The model reaches similar results for regions of the carotid artery that are not incorporated in the training set and on MRI of young, healthy subjects. The model also achieves a low median Hausdorff distance of 0.437 / 0.552 mm on the 2021 Carotid Artery Vessel Wall Segmentation Challenge test set. Conclusions: The proposed method can reduce the effort for carotid artery vessel wall assessment. Together with human supervision, it can be used for clinical applications, as it allows a reliable measurement of the VWT for different patient demographics and MRI acquisition settings.

Perfect Match: Radiomics and Artificial Intelligence in Cardiac Imaging.

Cardiovascular diseases remain a significant health burden, with imaging modalities like echocardiography, cardiac computed tomography, and cardiac magnetic resonance imaging playing a crucial role in diagnosis and prognosis. However, the inherent heterogeneity of these diseases poses challenges, necessitating advanced analytical methods like radiomics and artificial intelligence. Radiomics extracts quantitative features from medical images, capturing intricate patterns and subtle variations that may elude visual inspection. Artificial intelligence techniques, including deep learning, can analyze these features to generate knowledge, define novel imaging biomarkers, and support diagnostic decision-making and outcome prediction. Radiomics and artificial intelligence thus hold promise for significantly enhancing diagnostic and prognostic capabilities in cardiac imaging, paving the way for more personalized and effective patient care. This review explores the synergies between radiomics and artificial intelligence in cardiac imaging, following the radiomics workflow and introducing concepts from both domains. Potential clinical applications, challenges, and limitations are discussed, along with solutions to overcome them.

Population-based reference values for 4D flow MRI derived aortic blood flow parameters.

Objective. This study assesses age-related differences of thoracic aorta blood flow profiles and provides age- and sex-specific reference values using 4D flow cardiovascular magnetic resonance (CMR) data.Approach. 126 volunteers (age 20-80 years, female 51%) underwent 4D flow CMR and 12 perpendicular analysis planes in the thoracic aorta were specified. For these planes the following parameters were evaluated: body surface area-adjusted aortic area (A'), normalized flow displacement (NFD), the degree of wall parallelism (WPD), the minimal relative cross-sectional area through which 80% of the volume flow passes (A80) and the angle between flow direction and centerline (α).Main results. Age-related differences in blood flow parameters were seen in the ascending aorta with higher values for NFD and angle and lower values for WPD and A80 in older subjects. All parameters describing blood flow patterns correlated with the cross-sectional area in the ascending aorta. No relevant sex-differences regarding blood flow profiles were found.Significance. These age- and sex-specific reference values for quantitative parameters describing blood flow within the aorta might help to study the clinical relevance of flow profiles in the future.

Characterization of aortic aging using 3D multi-parametric MRI-long-term follow-up in a population study.

We comprehensively studied morphological and functional aortic aging in a population study using modern three-dimensional MR imaging to allow future comparison in patients with diseases of the aortic valve or aorta. We followed 80 of 126 subjects of a population study (20 to 80 years of age at baseline) using the identical methodology 6.0 ± 0.5 years later. All underwent 3 T MRI of the thoracic aorta including 3D T1 weighted MRI (spatial resolution 1 mm3) for measuring aortic diameter and plaque thickness and 4D flow MRI (spatial/temporal resolution = 2 mm3/20 ms) for calculating global and regional aortic pulse wave velocity (PWV) and helicity of aortic blood flow. Mean diameter of the ascending aorta (AAo) decreased and plaque thickness increased significantly in the aortic arch (AA) and descending aorta (DAo) in females. PWV of the thoracic aorta increased (6.4 ± 1.5 to 7.0 ± 1.7 m/s and 6.8 ± 1.5 to 7.3 ± 1.8 m/s in females and males, respectively) over time. Local normalized helicity volumes (LNHV) decreased significantly in the AAo and AA (0.33 to 0.31 and 0.34 to 0.32 in females and 0.34 to 0.32 and 0.32 to 0.28 in males). By contrast, helicity increased significantly in the DAo in both genders (0.28 to 0.29 and 0.29 to 0.30, respectively). 3D MRI was able to characterize changes in aortic diameter, plaque thickness, PWV and helicity during six years in our population. Aortic aging determined by 3D multi-parametric MRI is now available for future comparisons in patients with diseases of the aortic valve or aorta.

Deep Learning Segmentation of the Right Ventricle in Cardiac MRI: The M&Ms Challenge.

In recent years, several deep learning models have been proposed to accurately quantify and diagnose cardiac pathologies. These automated tools heavily rely on the accurate segmentation of cardiac structures in MRI images. However, segmentation of the right ventricle is challenging due to its highly complex shape and ill-defined borders. Hence, there is a need for new methods to handle such structure's geometrical and textural complexities, notably in the presence of pathologies such as Dilated Right Ventricle, Tricuspid Regurgitation, Arrhythmogenesis, Tetralogy of Fallot, and Inter-atrial Communication. The last MICCAI challenge on right ventricle segmentation was held in 2012 and included only 48 cases from a single clinical center. As part of the 12th Workshop on Statistical Atlases and Computational Models of the Heart (STACOM 2021), the M&Ms-2 challenge was organized to promote the interest of the research community around right ventricle segmentation in multi-disease, multi-view, and multi-center cardiac MRI. Three hundred sixty CMR cases, including short-axis and long-axis 4-chamber views, were collected from three Spanish hospitals using nine different scanners from three different vendors, and included a diverse set of right and left ventricle pathologies. The solutions provided by the participants show that nnU-Net achieved the best results overall. However, multi-view approaches were able to capture additional information, highlighting the need to integrate multiple cardiac diseases, views, scanners, and acquisition protocols to produce reliable automatic cardiac segmentation algorithms.

Cascaded neural network-based CT image processing for aortic root analysis.

PURPOSE: Careful assessment of the aortic root is paramount to select an appropriate prosthesis for transcatheter aortic valve implantation (TAVI). Relevant information about the aortic root anatomy, such as the aortic annulus diameter, can be extracted from pre-interventional CT. In this work, we investigate a neural network-based approach for segmenting the aortic root as a basis for obtaining these parameters. METHODS: To support valve prosthesis selection, geometric measures of the aortic root are extracted from the patient's CT scan using a cascade of convolutional neural networks (CNNs). First, the image is reduced to the aortic root, valve, and left ventricular outflow tract (LVOT); within that subimage, the aortic valve and ascending aorta are segmented; and finally, the region around the aortic annulus. From the segmented annulus region, we infer the annulus orientation using principal component analysis (PCA). The area-derived diameter of the annulus is approximated based on the segmentation of the aortic root and LVOT and the plane orientation resulting from the PCA. RESULTS: The cascade of CNNs was trained using 90 expert-annotated contrast-enhanced CT scans routinely acquired for TAVI planning. Segmentation of the aorta and valve within the region of interest achieved an F1 score of 0.94 on the test set of 36 patients. The area-derived diameter within the annulus region was determined with a mean error below 2 mm between the automatic measurement and the diameter derived from annotations. The calculated diameters and resulting errors are comparable to published results of alternative approaches. CONCLUSIONS: The cascaded neural network approach enabled the assessment of the aortic root with a relatively small training set. The processing time amounts to 30 s per patient, facilitating time-efficient, reproducible measurements. An extended training data set, including different levels of calcification or special cases (e.g., pre-implanted valves), could further improve this method's applicability and robustness.

Virtual downsizing for decision support in mitral valve repair.

PURPOSE: Various options are available for the treatment of mitral valve insufficiency, including reconstructive approaches such as annulus correction through ring implants. The correct choice of general therapy and implant is relevant for an optimal outcome. Additional to guidelines, decision support systems (DSS) can provide decision aid by means of virtual intervention planning and predictive simulations. Our approach on virtual downsizing is one of the virtual intervention tools that are part of the DSS workflow. It allows for emulating a ring implantation based on patient-specific lumen geometry and vendor-specific implants. METHODS: Our approach is fully automatic and relies on a lumen mask and an annulus contour as inputs. Both are acquired from previous DSS workflow steps. A virtual surface- and contour-based model of a vendor-specific ring design (26-40 mm) is generated. For each case, the ring geometry is positioned with respect to the original, patient-specific annulus and additional anatomical landmarks. The lumen mesh is parameterized to allow for a vertex-based deformation with respect to the user-defined annulus. Derived from post-interventional observations, specific deformation schemes are applied to atrium and ventricle and the lumen mesh is altered with respect to the ring location. RESULTS: For quantitative evaluation, the surface distance between the deformed lumen mesh and segmented post-operative echo lumen close to the annulus was computed for 11 datasets. The results indicate a good agreement. An arbitrary subset of six datasets was used for a qualitative evaluation of the complete lumen. Two domain experts compared the deformed lumen mesh with post-interventional echo images. All deformations were deemed plausible. CONCLUSION: Our approach on virtual downsizing allows for an automatic creation of plausible lumen deformations. As it takes only a few seconds to generate results, it can be added to a virtual intervention toolset without unnecessarily increasing the pipeline complexity.

Extraction of open-state mitral valve geometry from CT volumes.

PURPOSE: The importance of mitral valve therapies is rising due to an aging population. Visualization and quantification of the valve anatomy from image acquisitions is an essential component of surgical and interventional planning. The segmentation of the mitral valve from computed tomography (CT) acquisitions is challenging due to high variation in appearance and visibility across subjects. We present a novel semi-automatic approach to segment the open-state valve in 3D CT volumes that combines user-defined landmarks to an initial valve model which is automatically adapted to the image information, even if the image data provide only partial visibility of the valve. METHODS: Context information and automatic view initialization are derived from segmentation of the left heart lumina, which incorporates topological, shape and regional information. The valve model is initialized with user-defined landmarks in views generated from the context segmentation and then adapted to the image data in an active surface approach guided by landmarks derived from sheetness analysis. The resulting model is refined by user landmarks. RESULTS: For evaluation, three clinicians segmented the open valve in 10 CT volumes of patients with mitral valve insufficiency. Despite notable differences in landmark definition, the resulting valve meshes were overall similar in appearance, with a mean surface distance of [Formula: see text] mm. Each volume could be segmented in 5-22 min. CONCLUSIONS: Our approach enables an expert user to easily segment the open mitral valve in CT data, even when image noise or low contrast limits the visibility of the valve.

Comparing algorithms for automated vessel segmentation in computed tomography scans of the lung: the VESSEL12 study.

The VESSEL12 (VESsel SEgmentation in the Lung) challenge objectively compares the performance of different algorithms to identify vessels in thoracic computed tomography (CT) scans. Vessel segmentation is fundamental in computer aided processing of data generated by 3D imaging modalities. As manual vessel segmentation is prohibitively time consuming, any real world application requires some form of automation. Several approaches exist for automated vessel segmentation, but judging their relative merits is difficult due to a lack of standardized evaluation. We present an annotated reference dataset containing 20 CT scans and propose nine categories to perform a comprehensive evaluation of vessel segmentation algorithms from both academia and industry. Twenty algorithms participated in the VESSEL12 challenge, held at International Symposium on Biomedical Imaging (ISBI) 2012. All results have been published at the VESSEL12 website http://vessel12.grand-challenge.org. The challenge remains ongoing and open to new participants. Our three contributions are: (1) an annotated reference dataset available online for evaluation of new algorithms; (2) a quantitative scoring system for objective comparison of algorithms; and (3) performance analysis of the strengths and weaknesses of the various vessel segmentation methods in the presence of various lung diseases.