An attention-based deep learning method for right ventricular quantification using 2D echocardiography: Feasibility and accuracy.

AIM: To test the feasibility and accuracy of a new attention-based deep learning (DL) method for right ventricular (RV) quantification using 2D echocardiography (2DE) with cardiac magnetic resonance imaging (CMR) as reference. METHODS AND RESULTS: We retrospectively analyzed images from 50 adult patients (median age 51, interquartile range 32-62 42% women) who had undergone CMR within 1 month of 2DE. RV planimetry of the myocardial border was performed in end-diastole (ED) and end-systole (ES) for eight standardized 2DE RV views with calculation of areas. The DL model comprised a Feature Tokenizer module and a stack of Transformer layers. Age, gender and calculated areas were used as inputs, and the output was RV volume in ED/ES. The dataset was randomly split into training, validation and testing subsets (35, 5 and 10 patients respectively). Mean RVEDV, RVESV and RV ejection fraction (EF) were 163 ± 70 mL, 82 ± 42 mL and 51% ± 8% respectively without differences among the subsets. The proposed method achieved good prediction of RV volumes (R2  = .953, absolute percentage error [APE] = 9.75% ± 6.23%) and RVEF (APE = 7.24% ± 4.55%). Per CMR, there was one patient with RV dilatation and three with RV dysfunction in the testing dataset. The DL model detected RV dilatation in 1/1 case and RV dysfunction in 4/3 cases. CONCLUSIONS: An attention-based DL method for 2DE RV quantification showed feasibility and promising accuracy. The method requires validation in larger cohorts with wider range of RV size and function. Further research will focus on the reduction of the number of required 2DE to make the method clinically applicable.

Resource efficient aortic distensibility calculation by end to end spatiotemporal learning of aortic lumen from multicentre multivendor multidisease CMR images.

Aortic distensibility (AD) is important for the prognosis of multiple cardiovascular diseases. We propose a novel resource-efficient deep learning (DL) model, inspired by the bi-directional ConvLSTM U-Net with densely connected convolutions, to perform end-to-end hierarchical learning of the aorta from cine cardiovascular MRI towards streamlining AD quantification. Unlike current DL aortic segmentation approaches, our pipeline: (i) performs simultaneous spatio-temporal learning of the video input, (ii) combines the feature maps from the encoder and decoder using non-linear functions, and (iii) takes into account the high class imbalance. By using multi-centre multi-vendor data from a highly heterogeneous patient cohort, we demonstrate that the proposed method outperforms the state-of-the-art method in terms of accuracy and at the same time it consumes [Formula: see text] 3.9 times less fuel and generates [Formula: see text] 2.8 less carbon emissions. Our model could provide a valuable tool for exploring genome-wide associations of the AD with the cognitive performance in large-scale biomedical databases. By making energy usage and carbon emissions explicit, the presented work aligns with efforts to keep DL's energy requirements and carbon cost in check. The improved resource efficiency of our pipeline might open up the more systematic DL-powered evaluation of the MRI-derived aortic stiffness.

Knee Injury Detection Using Deep Learning on MRI Studies: A Systematic Review.

The improved treatment of knee injuries critically relies on having an accurate and cost-effective detection. In recent years, deep-learning-based approaches have monopolized knee injury detection in MRI studies. The aim of this paper is to present the findings of a systematic literature review of knee (anterior cruciate ligament, meniscus, and cartilage) injury detection papers using deep learning. The systematic review was carried out following the PRISMA guidelines on several databases, including PubMed, Cochrane Library, EMBASE, and Google Scholar. Appropriate metrics were chosen to interpret the results. The prediction accuracy of the deep-learning models for the identification of knee injuries ranged from 72.5-100%. Deep learning has the potential to act at par with human-level performance in decision-making tasks related to the MRI-based diagnosis of knee injuries. The limitations of the present deep-learning approaches include data imbalance, model generalizability across different centers, verification bias, lack of related classification studies with more than two classes, and ground-truth subjectivity. There are several possible avenues of further exploration of deep learning for improving MRI-based knee injury diagnosis. Explainability and lightweightness of the deployed deep-learning systems are expected to become crucial enablers for their widespread use in clinical practice.

Three-Dimensional Late Gadolinium Enhancement Cardiovascular Magnetic Resonance Predicts Inducibility of Ventricular Tachycardia in Adults With Repaired Tetralogy of Fallot.

BACKGROUND: Adults with repaired tetralogy of Fallot die prematurely from ventricular tachycardia (VT) and sudden cardiac death. Inducible VT predicts mortality. Ventricular scar, the key substrate for VT, can be noninvasively defined with late gadolinium enhancement (LGE) cardiovascular magnetic resonance but whether this relates to inducible VT is unknown. METHODS: Sixty-nine consecutive repaired tetralogy of Fallot patients (43 male, mean 40±15 years) clinically scheduled for invasive programmed VT-stimulation were prospectively recruited for prior 3-dimensional LGE cardiovascular magnetic resonance. Ventricular LGE was segmented and merged with reconstructed cardiac chambers and LGE volume measured. RESULTS: VT was induced in 22 (31%) patients. Univariable predictors of inducible VT included increased RV LGE (odds ratio [OR], 1.15; P=0.001 per cm3), increased nonapical vent LV LGE (OR, 1.09; P=0.008 per cm3), older age (OR, 1.6; P=0.01 per decile), QRS duration ≥180 ms (OR, 3.5; P=0.02), history of nonsustained VT (OR, 3.5; P=0.02), and previous clinical sustained VT (OR, 12.8; P=0.003); only prior sustained VT (OR, 8.02; P=0.02) remained independent in bivariable analyses after controlling for RV LGE volume (OR, 1.14; P=0.003). An RV LGE volume of 25 cm3 had 72% sensitivity and 81% specificity for predicting inducible VT (area under the curve, 0.81; P<0.001). At the extreme cutoffs for ruling-out and ruling-in inducible VT, RV LGE >10 cm3 was 100% sensitive and >36 cm3 was 100% specific for predicting inducible VT. CONCLUSIONS: Three-dimensional LGE cardiovascular magnetic resonance-defined scar burden is independently associated with inducible VT and may help refine patient selection for programmed VT-stimulation when applied to an at least intermediate clinical risk cohort.

Transmural Remodeling of Cardiac Microstructure in Aged Spontaneously Hypertensive Rats by Diffusion Tensor MRI.

The long-standing high blood pressure (also known as hypertension) overworks the heart. Microstructural remodeling is a key factor of hypertensive heart disease progression. Diffusion tensor magnetic resonance imaging (DT-MRI) is a powerful tool for the rapid noninvasive nondestructive delineation of the cardiomyocyte organization. The spontaneously hypertensive rat (SHR) is a well-established model of genetic hypertension. The goal of this study was to employ high-resolution DT-MRI and the SHR animal model to assess the transmural layer-specific remodeling of myocardial microstructure associated with hypertension. Ex vivo experiments were performed on excised formalin-fixed hearts of aged SHRs (n = 4) and age-matched controls (n = 4). The DT-MRI-derived fractional anisotropy (FA), longitudinal diffusivity (λ L ), transversal diffusivity (λ T ), and mean diffusivity (MD) served as the readout parameters investigated at three transmural zones (i.e., endocardium, mesocardium, and epicardium). The helix angles (HAs) of the aggregated cardiomyocytes and the orientation of laminar sheetlets were also studied. Compared with controls, the SHRs exhibited decreased epicardial FA, while FA changes in the other two transmural regions were insignificant. No substantial differences were observed in the diffusivity parameters and the transmural course of HAs between the two groups. A consistent distribution pattern of laminar sheetlet orientation was not identified for either group. Our findings are in line with the known cellular microstructure from early painstaking histological studies. Biophysical explanations of the study outcomes are provided. In conclusion, our experimental findings indicate that the epicardial microstructure is more vulnerable to high blood pressure leading to more pronounced changes in this region during remodeling. DT-MRI is well-suited for elucidating these alterations. The revealed transmural nonuniformity of myocardial reorganization may shed light on the mechanisms of the microstructure-function relationship in hypertension progression. Our results provide insights into the management of patients with systemic arterial hypertension, thus prevent the progression toward heart failure. The findings of this study should be acknowledged by electromechanical models of the heart that simulate the specific cardiac pathology.

Assessment of Myocardial Microstructural Dynamics by In Vivo Diffusion Tensor Cardiac Magnetic Resonance.

BACKGROUND: Cardiomyocytes are organized in microstructures termed sheetlets that reorientate during left ventricular thickening. Diffusion tensor cardiac magnetic resonance (DT-CMR) may enable noninvasive interrogation of in vivo cardiac microstructural dynamics. Dilated cardiomyopathy (DCM) is a condition of abnormal myocardium with unknown sheetlet function. OBJECTIVES: This study sought to validate in vivo DT-CMR measures of cardiac microstructure against histology, characterize microstructural dynamics during left ventricular wall thickening, and apply the technique in hypertrophic cardiomyopathy (HCM) and DCM. METHODS: In vivo DT-CMR was acquired throughout the cardiac cycle in healthy swine, followed by in situ and ex vivo DT-CMR, then validated against histology. In vivo DT-CMR was performed in 19 control subjects, 19 DCM, and 13 HCM patients. RESULTS: In swine, a DT-CMR index of sheetlet reorientation (E2A) changed substantially (E2A mobility ∼46°). E2A changes correlated with wall thickness changes (in vivo r2 = 0.75; in situ r2 = 0.89), were consistently observed under all experimental conditions, and accorded closely with histological analyses in both relaxed and contracted states. The potential contribution of cyclical strain effects to in vivo E2A was ∼17%. In healthy human control subjects, E2A increased from diastole (18°) to systole (65°; p < 0.001; E2A mobility = 45°). HCM patients showed significantly greater E2A in diastole than control subjects did (48°; p < 0.001) with impaired E2A mobility (23°; p < 0.001). In DCM, E2A was similar to control subjects in diastole, but systolic values were markedly lower (40°; p < 0.001) with impaired E2A mobility (20°; p < 0.001). CONCLUSIONS: Myocardial microstructure dynamics can be characterized by in vivo DT-CMR. Sheetlet function was abnormal in DCM with altered systolic conformation and reduced mobility, contrasting with HCM, which showed reduced mobility with altered diastolic conformation. These novel insights significantly improve understanding of contractile dysfunction at a level of noninvasive interrogation not previously available in humans.

Quantitative analysis of hypertrophic myocardium using diffusion tensor magnetic resonance imaging.

Systemic hypertension is a causative factor in left ventricular hypertrophy (LVH). This study is motivated by the potential to reverse or manage the dysfunction associated with structural remodeling of the myocardium in this pathology. Using diffusion tensor magnetic resonance imaging, we present an analysis of myocardial fiber and laminar sheet orientation in ex vivo hypertrophic (6 SHR) and normal (5 WKY) rat hearts using the covariance of the diffusion tensor. First, an atlas of normal cardiac microstructure was formed using the WKY b0 images. Then, the SHR and WKY b0 hearts were registered to the atlas. The acquired deformation fields were applied to the SHR and WKY heart tensor fields followed by the preservation of principal direction (PPD) reorientation strategy. A mean tensor field was then formed from the registered WKY tensor images. Calculating the covariance of the registered tensor images about this mean for each heart, the hypertrophic myocardium exhibited significantly increased myocardial fiber derangement ([Formula: see text]) with a mean dispersion of 38.7 deg, and an increased dispersion of the laminar sheet normal ([Formula: see text]) of 54.8 deg compared with 34.8 deg and 51.8 deg, respectively, in the normal hearts. Results demonstrate significantly altered myocardial fiber and laminar sheet structure in rats with hypertensive LVH.

On the averaging of cardiac diffusion tensor MRI data: the effect of distance function selection.

Diffusion tensor magnetic resonance imaging (DT-MRI) allows a unique insight into the microstructure of highly-directional tissues. The selection of the most proper distance function for the space of diffusion tensors is crucial in enhancing the clinical application of this imaging modality. Both linear and nonlinear metrics have been proposed in the literature over the years. The debate on the most appropriate DT-MRI distance function is still ongoing. In this paper, we presented a framework to compare the Euclidean, affine-invariant Riemannian and log-Euclidean metrics using actual high-resolution DT-MRI rat heart data. We employed temporal averaging at the diffusion tensor level of three consecutive and identically-acquired DT-MRI datasets from each of five rat hearts as a means to rectify the background noise-induced loss of myocyte directional regularity. This procedure is applied here for the first time in the context of tensor distance function selection. When compared with previous studies that used a different concrete application to juxtapose the various DT-MRI distance functions, this work is unique in that it combined the following: (i) metrics were judged by quantitative-rather than qualitative-criteria, (ii) the comparison tools were non-biased, (iii) a longitudinal comparison operation was used on a same-voxel basis. The statistical analyses of the comparison showed that the three DT-MRI distance functions tend to provide equivalent results. Hence, we came to the conclusion that the tensor manifold for cardiac DT-MRI studies is a curved space of almost zero curvature. The signal to noise ratio dependence of the operations was investigated through simulations. Finally, the 'swelling effect' occurrence following Euclidean averaging was found to be too unimportant to be worth consideration.

Value of Formalin Fixation for the Prolonged Preservation of Rodent Myocardial Microanatomical Organization: Evidence by MR Diffusion Tensor Imaging.

Previous ex vivo diffusion tensor imaging (DTI) studies on formalin-fixed myocardial tissue assumed that, after some initial changes in the first 48 hr since the start of fixation, DTI parameters remain stable over time. Prolonged preservation of cardiac tissue in formalin prior to imaging has been seen many times in the DTI literature as it is considered orderly. Our objective is to define the effects of the prolonged cardiac tissue exposure to formalin on tissue microanatomical organization, as this is assessed by DTI parameters. DTI experiments were conducted on eight excised rodent hearts that were fixed by immersion in formalin. The samples were randomly divided into two equinumerous groups corresponding to shorter (∼2 weeks) and more prolonged (∼6-8 weeks) durations of tissue exposure to formalin prior to imaging. We found that when the duration of cardiac tissue exposure to formalin before imaging increased, water diffusion became less restricted, helix angle (HA) histograms flattened out and exhibited heavier tails (even though the classic HA transmural variation was preserved), and a significant loss of inter-voxel primary diffusion orientation integrity was introduced. The prolonged preservation of cardiac tissue in formalin profoundly affected its microstructural organization, as this was assessed by DTI parameters. The accurate interpretation of diffusivity profiles necessitates awareness of the pitfalls of prolonged cardiac tissue exposure duration to formalin. The acquired knowledge works to the advantage of a proper experimental design of DTI studies of fixed hearts. Anat Rec, 299:878-887, 2016. © 2016 Wiley Periodicals, Inc.

Rapid automatic segmentation of abnormal tissue in late gadolinium enhancement cardiovascular magnetic resonance images for improved management of long-standing persistent atrial fibrillation.

BACKGROUND: Atrial fibrillation (AF) is the most common heart rhythm disorder. In order for late Gd enhancement cardiovascular magnetic resonance (LGE CMR) to ameliorate the AF management, the ready availability of the accurate enhancement segmentation is required. However, the computer-aided segmentation of enhancement in LGE CMR of AF is still an open question. Additionally, the number of centres that have reported successful application of LGE CMR to guide clinical AF strategies remains low, while the debate on LGE CMR's diagnostic ability for AF still holds. The aim of this study is to propose a method that reliably distinguishes enhanced (abnormal) from non-enhanced (healthy) tissue within the left atrial wall of (pre-ablation and 3 months post-ablation) LGE CMR data-sets from long-standing persistent AF patients studied at our centre. METHODS: Enhancement segmentation was achieved by employing thresholds benchmarked against the statistics of the whole left atrial blood-pool (LABP). The test-set cross-validation mechanism was applied to determine the input feature representation and algorithm that best predict enhancement threshold levels. RESULTS: Global normalized intensity threshold levels T PRE  = 1 1/4 and T POST  = 1 5/8 were found to segment enhancement in data-sets acquired pre-ablation and at 3 months post-ablation, respectively. The segmentation results were corroborated by using visual inspection of LGE CMR brightness levels and one endocardial bipolar voltage map. The measured extent of pre-ablation fibrosis fell within the normal range for the specific arrhythmia phenotype. 3D volume renderings of segmented post-ablation enhancement emulated the expected ablation lesion patterns. By comparing our technique with other related approaches that proposed different threshold levels (although they also relied on reference regions from within the LABP) for segmenting enhancement in LGE CMR data-sets of AF patients, we illustrated that the cut-off levels employed by other centres may not be usable for clinical studies performed in our centre. CONCLUSIONS: The proposed technique has great potential for successful employment in the AF management within our centre. It provides a highly desirable validation of the LGE CMR technique for AF studies. Inter-centre differences in the CMR acquisition protocol and image analysis strategy inevitably impede the selection of a universally optimal algorithm for segmentation of enhancement in AF studies.

Heterogeneity of Fractional Anisotropy and Mean Diffusivity Measurements by In Vivo Diffusion Tensor Imaging in Normal Human Hearts.

BACKGROUND: Cardiac diffusion tensor imaging (cDTI) by cardiovascular magnetic resonance has the potential to assess microstructural changes through measures of fractional anisotropy (FA) and mean diffusivity (MD). However, normal variation in regional and transmural FA and MD is not well described. METHODS: Twenty normal subjects were scanned using an optimised cDTI sequence at 3T in systole. FA and MD were quantified in 3 transmural layers and 4 regional myocardial walls. RESULTS: FA was higher in the mesocardium (0.46 ±0.04) than the endocardium (0.40 ±0.04, p≤0.001) and epicardium (0.39 ±0.04, p≤0.001). On regional analysis, the FA in the septum was greater than the lateral wall (0.44 ±0.03 vs 0.40 ±0.05 p = 0.04). There was a transmural gradient in MD increasing towards the endocardium (epicardium 0.87 ±0.07 vs endocardium 0.91 ±0.08×10-3 mm2/s, p = 0.04). With the lateral wall (0.87 ± 0.08×10-3 mm2/s) as the reference, the MD was higher in the anterior wall (0.92 ±0.08×10-3 mm2/s, p = 0.016) and septum (0.92 ±0.07×10-3 mm2/s, p = 0.028). Transmurally the signal to noise ratio (SNR) was greatest in the mesocardium (14.5 ±2.5 vs endocardium 13.1 ±2.2, p<0.001; vs epicardium 12.0 ± 2.4, p<0.001) and regionally in the septum (16.0 ±3.4 vs lateral wall 11.5 ± 1.5, p<0.001). Transmural analysis suggested a relative reduction in the rate of change in helical angle (HA) within the mesocardium. CONCLUSIONS: In vivo FA and MD measurements in normal human heart are heterogeneous, varying significantly transmurally and regionally. Contributors to this heterogeneity are many, complex and interactive, but include SNR, variations in cardiac microstructure, partial volume effects and strain. These data indicate that the potential clinical use of FA and MD would require measurement standardisation by myocardial region and layer, unless pathological changes substantially exceed the normal variation identified.

Improved 2D vector field estimation using probabilistic weights.

We consider the application of tomography to the reconstruction of two-dimensional vector fields. The most practical sensor configuration in such problems is the regular positioning along the boundary of the reconstruction domain. However, such a configuration does not result in uniform distribution in the Radon parameter space, which is a necessary requirement to achieve accurate reconstruction results. On the other hand, sampling the projection space uniformly imposes serious constraints on space or time. In this paper, we propose to place the sensors regularly along the boundary of the reconstruction domain and employ probabilistic weights with the purpose of compensating for the lack of uniformity in the distribution of projection space parameters. Simulation results demonstrate that, when the proposed probabilistic weights are employed, an average 27% decrease in the reconstruction error may be achieved, over the case that projection measurements are not weighed (e.g., in one case the error reduces from 3.7% to 2.6%). When compared with the case where actual uniform sampling of the projection space is employed, the proposed method achieves a 90 times reduction in the number of the required sensors or 180 times reduction in the total scanning time, with only 7% increase in the error with which the vector field is estimated.

Virtual sensors for 2D vector field tomography.

We consider the application of tomography to the reconstruction of 2-D vector fields. The most convenient sensor configuration in such problems is the regular positioning along the domain boundary. However, the most accurate reconstructions are obtained by sampling uniformly the Radon parameter domain rather than the border of the reconstruction domain. This dictates a prohibitively large number of sensors and impractical sensor positioning. In this paper, we propose uniform placement of the sensors along the boundary of the reconstruction domain and interpolation of the measurements for the positions that correspond to uniform sampling in the Radon domain. We demonstrate that when the cubic spline interpolation method is used, a 60 times reduction in the number of sensors may be achieved with only about 10% increase in the error with which the vector field is estimated. The reconstruction error by using the same sensors and ignoring the necessity of uniform sampling in the Radon domain is in fact higher by about 30%. The effects of noise are also examined.

Improved 2-D vector field reconstruction using virtual sensors and the Radon transform.

This paper describes a method that allows one to recover both components of a 2-D vector field based on boundary information only, by solving a system of linear equations. The analysis is carried out in the digital domain and takes advantage of the redundancy in the boundary data, since these may be viewed as weighted sums of the local vector field's Cartesian components. Furthermore, a sampling of lines is used in order to combine the available measurements along continuous tracing lines with the digitised 2-D space where the solution is sought. A significant enhancement in the performance of the proposed algorithm is achieved by using, apart from real data, also boundary data obtained at virtual sensors. The potential of the proposed method is demonstrated by presenting an example of vector field reconstruction.