Deep Learning Supplants Visual Analysis by Experienced Operators for the Diagnosis of Cardiac Amyloidosis by Cine-CMR

Background: Diagnosing cardiac amyloidosis (CA) from cine-CMR (cardiac magnetic resonance) alone is not reliable. In this study, we tested if a convolutional neural network (CNN) could outperform the visual diagnosis of experienced operators. Method: 119 patients with cardiac amyloidosis and 122 patients with left ventricular hypertrophy (LVH) of other origins were retrospectively selected. Diastolic and systolic cine-CMR images were preprocessed and labeled. A dual-input visual geometry group (VGG ) model was used for binary image classification. All images belonging to the same patient were distributed in the same set. Accuracy and area under the curve (AUC) were calculated per frame and per patient from a 40% held-out test set. Results were compared to a visual analysis assessed by three experienced operators. Results: frame-based comparisons between humans and a CNN provided an accuracy of 0.605 vs. 0.746 (p < 0.0008) and an AUC of 0.630 vs. 0.824 (p < 0.0001). Patient-based comparisons provided an accuracy of 0.660 vs. 0.825 (p < 0.008) and an AUC of 0.727 vs. 0.895 (p < 0.002). Conclusion: based on cine-CMR images alone, a CNN is able to discriminate cardiac amyloidosis from LVH of other origins better than experienced human operators (15 to 20 points more in absolute value for accuracy and AUC), demonstrating a unique capability to identify what the eyes cannot see through classical radiological analysis.

CinE caRdiac magneTic resonAnce to predIct veNTricular arrhYthmia (CERTAINTY)

Better models to identify individuals at low risk of ventricular arrhythmia (VA) are needed for implantable cardioverter-defibrillator (ICD) candidates to mitigate the risk of ICD-related complications. We designed the CERTAINTY study (CinE caRdiac magneTic resonAnce to predIct veNTricular arrhYthmia) with deep learning for VA risk prediction from cine cardiac magnetic resonance (CMR). Using a training cohort of primary prevention ICD recipients (n = 350, 97 women, median age 59 years, 178 ischemic cardiomyopathy) who underwent CMR immediately prior to ICD implantation, we developed two neural networks: Cine Fingerprint Extractor and Risk Predictor. The former extracts cardiac structure and function features from cine CMR in a form of cine fingerprint in a fully unsupervised fashion, and the latter takes in the cine fingerprint and outputs disease outcomes as a cine risk score. Patients with VA (n = 96) had a significantly higher cine risk score than those without VA. Multivariate analysis showed that the cine risk score was significantly associated with VA after adjusting for clinical characteristics, cardiac structure and function including CMR-derived scar extent. These findings indicate that non-contrast, cine CMR inherently contains features to improve VA risk prediction in primary prevention ICD candidates. We solicit participation from multiple centers for external validation.

Deep Learning for Classification and Selection of Cine CMR Images to Achieve Fully Automated Quality-Controlled CMR Analysis From Scanner to Report

Introduction: Deep learning demonstrates great promise for automated analysis of CMR. However, existing limitations, such as insufficient quality control and selection of target acquisitions from the full CMR exam, are holding back the introduction of deep learning tools in the clinical environment. This study aimed to develop a framework for automated detection and quality-controlled selection of standard cine sequences images from clinical CMR exams, prior to analysis of cardiac function.Materials and Methods: Retrospective study of 3,827 subjects that underwent CMR imaging. We used a total of 119,285 CMR acquisitions, acquired with scanners of different magnetic field strengths and from different vendors (1.5T Siemens and 1.5T and 3.0T Phillips). We developed a framework to select one good acquisition for each conventional cine class. The framework consisted of a first pre-processing step to exclude still acquisitions; two sequential convolutional neural networks (CNN), the first (CNNclass) to classify acquisitions in standard cine views (2/3/4-chamber and short axis), the second (CNNQC) to classify acquisitions according to image quality and orientation; a final algorithm to select one good acquisition of each class. For each CNN component, 7 state-of-the-art architectures were trained for 200 epochs, with cross entropy loss and data augmentation. Data were divided into 80% for training, 10% for validation, and 10% for testing.Results: CNNclass selected cine CMR acquisitions with accuracy ranging from 0.989 to 0.998. Accuracy of CNNQC reached 0.861 for 2-chamber, 0.806 for 3-chamber, and 0.859 for 4-chamber. The complete framework was presented with 379 new full CMR studies, not used for CNN training/validation/testing, and selected one good 2-, 3-, and 4-chamber acquisition from each study with sensitivity to detect erroneous cases of 89.7, 93.2, and 93.9%, respectively.Conclusions: We developed an accurate quality-controlled framework for automated selection of cine acquisitions prior to image analysis. This framework is robust and generalizable as it was developed on multivendor data and could be used at the beginning of a pipeline for automated cine CMR analysis to obtain full automatization from scanner to report.

Aortic stenosis reexpanded – a novel approach to determine aortic valve area with phase contrast cardiovascular magnetic resonance imaging

Background Transthoracic echocardiography (TTE) has become the diagnostic standard for evaluating aortic stenosis (AS) severity, mainly because of its advantages in comparison to the gold standard of cardiac catheterization. However, its inaccuracies in determining stroke volume (SV) and consequentially computing aortic valve area (AVA) call for a more precise and dependable method. Phase contrast cardiovascular magnetic resonance imaging (PC-CMR) is an aspiring tool to push these boundaries. Purpose The aim of this study was to validate a novel and simple approach based on PC-CMR against the invasive and echocardiographic determination of SV and AVA in patients with moderate and severe AS. Methods A total of 50 patients with moderate or severe AS underwent TTE, cardiac catheterization and CMR; AVA by PC-CMR was determined via plotting momentary flow across the valve against momentary flow velocity. SV via CMR was measured directly via PC-CMR and volumetrically using cine images. Invasive SV and AVA were determined via Fick principle and Gorlin formula, respectively. TTE yielded SV and AVA using the continuity equation. Finally, gradients were calculated via the modified Bernoulli equation. Results SV by PC-CMR showed a strong correlation with cine-CMR with no significant bias (r: 0.730, p&lt;0.001; SV by PC-CMR: 85±31ml; SV by cine-CMR: 85±19ml, p=0.829). Peak gradients determined by PC-CMR were 65±29mmHg and correlated inversely with AVA by PC-CMR (r: −0.371; p=0.008). Mean AVA during the whole systolic phase showed a moderate correlation (r: 0.544, p&lt;0.001) to invasive AVA with a small bias (AVA by CMR: 0.78±0.25cm2 versus invasive AVA: 0.70±0.23cm2, bias: 0.08cm2, p=0.017). Inter-methodical correlation and bias of AVA as measured by TTE and invasive AVA (AVA by TTE: 0.81±0.23cm2, r: 0.580, p&lt;0.001, bias 0.11cm2, p&lt;0.001) were similar to AVA by PC-CMR and invasive AVA. Conclusion PC-CMR provides a great option to yield reliable and solid SV values in patients with moderate and severe AS. Furthermore, continuous determination of flow volumes and flow velocities is able to determine AVA in these patients in an easy and reproducible manner. Our novel approach shines a light on the diagnostic potential of PC-CMR for non-invasive AS grading, especially in cases where echocardiography reaches its limits and where clinical findings appear inconclusive. FUNDunding Acknowledgement Type of funding sources: None. Central Illustration Cine (l,r) and PC-CMR (m) images in AS

Longitudinal strain measurement by 3D modelling from cine CMR: feasibility and comparison to 2D speckle tracking echocardiography

Background Global longitudinal strain (GLS) has emerged as a sensitive index of left ventricular (LV) systolic function with greater prognostic value than LV ejection fraction (LVEF) in a variety of cardiac disorders. While GLS is routinely derived from 2D speckle tracking echocardiography (STE) and feature tracking in cardiac magnetic resonance (CMR) imaging, calculation of strain via 3D geometric modelling enables analyses of deformation that are independent of 2D image plane constraints. Purpose We sought to compare longitudinal strain measurements extracted from geometric 3D analysis of CMR against values obtained from conventional 2D-STE. Methods Consecutive 2D-echocardiography (2D-echo) and steady-state free precession multiplanar cine CMR scans were performed in 80 prospectively recruited participants (48 healthy controls with LVEF range 53–74%, 30 patients with non-ischaemic cardiac disease with LVEF range 25–77%, and 2 heart transplant recipients with LVEF 53% and 58%), &lt;1 hour apart. Average endocardial peak GLS from 2D-STE was calculated offline using vendor-independent clinical software from apical triplane (2, 3 and 4-chamber) images for each of the standardised LV walls (anterior, anteroseptal, inferoseptal, inferior, inferolateral, anterolateral). Dynamic 3D geometric models of the LV were reconstructed from 3 long- and 6 short-axis CMR slices over one cardiac cycle. Corresponding longitudinal strain measurements were then evaluated by extracting analogous endocardial arc lengths (apex to base of each LV wall) from the 3D LV model. Finally, an average peak GLS was calculated as the mean of the peak longitudinal strains in each LV wall. Results GLS measured by 2D-STE ranged between −6.5% and −27.9% for the study population. A two-way mixed-effects intraclass correlation coefficient (ICC) for absolute agreement of 0.820 (95% CI: [0.720, 0.885]) demonstrated good correlation between average GLS obtained from 2D-STE and CMR. A Bland-Altman analysis revealed a minimal bias (&lt;1%) and 95% limits of agreement (LOA) between −6.3% and 5.5% (Fig. 1), with no apparent proportional bias. Comparatively lower correlation and wider LOA between longitudinal strains from 2D-STE and CMR were observed for each LV wall (Table I). Conclusions Fully automated calculation of LV GLS can be obtained from geometric 3D CMR analysis. Average peak GLS from cine CMR exhibits good agreement with 2D-STE, despite showing only moderate agreement at each LV wall. The increased discrepancy in regional longitudinal strain may be attributed to subjective plane positioning in 2D-echo, which can be expected to improve with advances in 3D-STE. The calculation of GLS by 3D geometric modelling may enhance the diagnostic value of routine cine CMR examinations for LV systolic function assessment. FUNDunding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Health Research Council (HRC) of New Zealand and National Heart Foundation (NHF) of New Zealand Figure 1. Bland-Altman analysis Table I. Regional correlations

Automated non-invasive pressure-volume loop analysis of cardiac aging in a large cohort of healthy community dwellers

Background/Introduction Pressure-volume loops (PVloops) provide a wealth of information on cardiac function that is not readily available from cardiac imaging alone. Methods To estimate left ventricular (LV) PVloops non-invasively have been available, but have so far not been used to interrogate ventricular function in large patient cohorts, due to the complexity of estimating PVloops. A new method was recently validated that construct PVloops non-invasively from cine cardiac magnetic resonance (CMR), based on the time-varying elastance model [1]. At the same time, we have validated a framework for automated, quality controlled analysis of cine CMR in large cohorts of patients/subjects [2]. Combining these two methods could automated PVloop estimation, enabling analysis of ventricular pressure-volume relationships in large study populations. Purpose Evaluate if CMR-based non-invasive PVloops can be used to interrogate the impact of cardiac ageing on LV function occurring in a large population of healthy community dwellers. Methods Non-invasive PVloops were calculated from a full cardiac cycle LV volume curve and brachial blood pressure data using a recently validated method based on the time-varying elastance model [1], in 7,650 healthy community dwellers from the UKBiobank population study. The LV volume curve was automatically obtained using our state-of-the-art, quality controlled deep learning (DL) based cine CMR analysis framework [2]. External Work, pressure-volume-area (PVA), end-systolic pressure (Pes), ventricular elastance (Ees, an estimate of contractility) and arterial elastance (Ea) and energy per ejected volume (EEV: PVA/ stroke volume) were calculated from the PVloops. We performed univariate regression between PVloop parameters and age. We also calculated the additional impact of cardiovascular risk-factors in a multivariate analysis. Results See results in table 1. With age, LV volumes fall (p&lt;0.001) in healthy subjects, while systolic blood pressure and Pes increases (both p&lt;0.001). As a result of the higher afterload, PVA (p=0.894) and EW (p=0.499) do not significantly change with age despite a lower SV. Arterial elastance (Ea) increased, and so did contractility, as measured by Ees (p&lt;0.001). Due to all these changes, EEV increased with age (p&lt;0.001). In multivariate analysis, cardiovascular risk factors hypercholesterolemia and hypertension negatively impacted Pes, PVA, Ees and EEV. Diabetes and smoking habits did not. Conclusion Non-invasive CMR-based PVloop analyses capture the impact of known changes occurring during cardiac ageing on cardiac work, contractility and energetic expenditure. Obtaining PVloops automatically using our AI analysis system in this large cohort of healthy subjects allows to formulate reference for assessment of cardiac disease. FUNDunding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): The authors acknowledge financial support (support) the National Institute for Health Research (NIHR) Cardiovascular MedTech Co-operative (previously existing as the Cardiovascular Healthcare Technology Co-operative 2012 - 2017) award to the Guy's and St Thomas' NHS Foundation Trust, in partnership with King's College London and the NIHR comprehensive Biomedical Research Centre of the Guy's & St Thomas' NHS Foundation Trust. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health Univariate regression analysis Example of estimated PV loop

Fairness in Cardiac Magnetic Resonance Imaging: Assessing sex and racial bias in deep learning-based segmentation

Background: Artificial intelligence (AI) techniques have been proposed for automation of cine CMR segmentation for functional quantification. However, in other applications AI models have been shown to have potential for sex and/or racial bias. Objectives: To perform the first analysis of sex/racial bias in AI-based cine CMR segmentation using a large-scale database. Methods: A state-of-the-art deep learning (DL) model was used for automatic segmentation of both ventricles and the myocardium from cine short-axis CMR. The dataset consisted of end-diastole and end-systole short-axis cine CMR images of 5,903 subjects from the UK Biobank database (61.5±7.1 years, 52% male, 81% white). To assess sex and racial bias, we compared Dice scores and errors in measurements of biventricular volumes and function between patients grouped by race and sex. To investigate whether segmentation bias could be explained by potential confounders, a multivariate linear regression and ANCOVA were performed. Results: We found statistically significant differences in Dice scores (white ~94% vs minority ethnic groups 86-89%) as well as in absolute/relative errors in volumetric and functional measures, showing that the AI model was biased against minority racial groups, even after correction for possible confounders. Conclusions: We have shown that racial bias can exist in DL-based cine CMR segmentation models. We believe that this bias is due to the unbalanced nature of the training data (combined with physiological differences). This is supported by the results which show racial bias but not sex bias when trained using the UK Biobank database, which is sex-balanced but not race-balanced.

Feature-Tracking Strain Derived from Compressed Sensing Cine Cardiovascular Magnetic Resonance Imaging for Myocardial Infarct Detection: A Feasibility Study

Purpose: This study aimed to use gadolinium-enhanced cardiovascular magnetic resonance (LGE-CMR) scanning to examine the clinical feasibility of feature-tracking strain (FT-strain) analysis on compressed sensing (CS) cine cardiovascular magnetic resonance (CMR) imaging for detecting myocardial infarction (MI).Methods: We enrolled 37 patients who underwent conventional cine CMR, CS cine CMR, and LGE-CMR scanning to assess cardiovascular disease. FT-strain analysis was used to assess peak circumferential strain (p-CS) based on an 18-segment model in both cine CMR imaging modalities. Based on LGE-CMR imaging findings, myocardial segments were classified as remote, adjacent, subendocardial infarcted, and transmural infarcted. The diagnostic performance of p-CS for detecting MI was compared between CS cine CMR imaging and conventional cine CMR imaging using the receiver operating characteristic (ROC) curve analysis.Results: A total of 440 remote, 85 adjacent, 76 subendocardial infarcted, and 65 transmural infarcted segments were diagnosed on LGE-CMR imaging. There were significant between-group differences in p-CS on both conventional and CS cine CMR (p <0.05 in each) imaging. The sensitivity and specificity of p-CS for identifying MI were 85% and 79% for conventional cine CMR imaging, and 82% and 77% for CS cine CMR imaging, respectively. There was no significant difference between conventional and CS cine CMR imaging in the area under the curve of p-CS (0.89 vs. 0.87, p = 0.15).Conclusion: FT-strain analysis of CS cine CMR imaging may help identify MI; it may be used alongside or instead of conventional CMR imaging.

Cine MR-Based Radiomics to Predict Myocardial Segments With Infarction

Background: Although myocardial infarction (MI) can be assessed quantitatively and qualitatively by using late gadolinium-enhanced (LGE) cardiovascular magnetic resonance (CMR) imaging, intravenous administration of gadolinium can expose patients to high risk of nephrogenic systemic fibrosis, especially in those with cardiovascular diseases. The purpose of this study is to harness cine CMR-based radiomics for predicting MI without introducing gadolinium.Methods: In this retrospective study, we included 48 patients with acute myocardial infarction (AMI) confirmed by later gadolinium enhancement (LGE) at CMR. CMR examinations were performed within 2 to 6 days after PCI. According to the LGE, each myocardial segment was dichotomized into with and without MI. Radiomic features of myocardial segments were extracted from cine CMR images and the myocardial segments were divided into training and validation sets randomly at a ratio of 0.7:0.3. Pearson correlation and Mann-Whitney U rank test were used to eliminate redundant and irrelevant features. A least absolute shrinkage and selection operator (LASSO) algorithm was used for features selection in the training set. Radiomic signatures were constructed in both the training and validation sets and its predictive performance was assessed using area under the cure of receiver operating characteristic (AUC-ROC).Results：Of 768 myocardial segments in the 48 patients, there were 291 (38%) segments with MI and 477 (62%) segments without MI. After univariate analysis, there were 22 RFs related to MI with statistical significance. LASSO regression selected 18 RFs for radiomics signature builting. AUC-ROC of radiomic signatures in prediction of segments with MI was 0.74(95% CI:0.69-0.78)and 0.68 (95%CI: 0.60-0.75) in the training and validation sets, respectively. The difference was not statistically significant (p=0.14).Conclusion: Cine MR-based radiomics signature can achieve a good prediction performance for MI, which showed the potential to be a promising imaging biomarker for MI without the administration of contrast agent.