Automated left ventricular dimension assessment using artificial intelligence

Background and purpose

Artificial intelligence (AI) has the potential to greatly improve efficiency and reproducibility of quantification in echocardiography, but to gain widespread use it must both meet expert standards of excellence and have a transparent methodology. We developed an online platform to enable multiple collaborators to annotate medical images for training and validating neural networks.

Methods

Using our online collaborative platform 9 expert echocardiographers labelled 2056 images that comprised the training dataset. They labelled the four points from where the standard parasternal long axis (PLAX) measurements (interventricular septum, posterior wall, left ventricular dimension) would be made. Using these labelled images we trained a 2d convolutional neural network to replicate these labels. Separately, we curated an external validation dataset of the systolic and diastolic frames of 100 PLAX acquisitions. Each of these images were labelled twice by 13 different experts, and the average of the 26 measurements was taken as the consensus standard.

We then compared the individual experts and the AI measurements on the external validation dataset to the consensus standard, and calculated the precision standard deviation (SD) of the signed differences from the consensus standard.

Results

For diastolic septum thickness, the AI had a precision SD of 1.8 mm (ICC 0.81; 95% CI 0.73 to 0.97), compared with 2.0 mm for the individual experts (ICC 0.64; 95% CI 0.57 to 0.72). For diastolic posterior wall thickness, the AI had a precision SD 1.4 mm (ICC 0.54; 95% CI 0.38 to 0.66), and the individual experts 2.2 mm (ICC 0.37; 95% CI 0.29 to 0.46).

The AI's precision SD for left ventricular internal dimension was 3.5 mm (ICC 0.93, 95% CI 0.90 to 0.94), and for individual experts was 4.4mm (ICC 0.82, 95% CI 0.78 to 0.95). Both the experts and AI performed better in diastole than systole (precision SD AI 2.5mm vs 4.3mm, p<0.0001; experts 3.3mm vs 5.3mm, p<0.0001).

Conclusions

AI trained by a group of echocardiography experts was able to perform PLAX measurements which matched the reference standard more closely than any individual expert's own measurements.

This open, collaborative approach may be a model for the development of AI that is explainable to, and trusted by clinicians.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NIHR Imperil BRC ITMATDr Howard was additionally funded by Wellcome. Online collaborative platformResults of AI and experts

Three-month cardiac outcomes of survivors of COVID-19 presenting to a district general hospital in West Hertfordshire, UK

Background

COVID-19 is novel entity associated with significant morbidity and mortality. Most patients recover completely, however, a proportion describe persistent symptoms consistent with cardiopulmonary disturbance. The long-term cardiac and respiratory outcomes of COVID-19 are not known.

Purpose

The aim of this study was to undertake a comprehensive cardiopulmonary assessment of survivors of COVID-19 with post recovery symptoms. The cardiac findings are presented here.

Methods

Survivors of COVID-19 presenting to a general hospital in West Hertfordshire between 18 March 2020 and 15 May 2020, were reviewed at 3 months using a structured prespecified protocol. Patients with persistent symptoms and those admitted to the intensive care unit (ICU) were invited to attend a clinical assessment comprising an electrocardiogram (ECG), echocardiography, chest x-ray (CXR) and pulmonary function tests. At follow up, patients were categorised according to disease course: 1) monitored via a community based virtual hospital, 2) admitted for supplemental oxygen, 3) requiring non-invasive ventilation and 4) ICU admission.

Results

448 eligible patients were evaluated by telephone. 11 patients admitted to ICU and 147 patients with persistent symptoms were invited for further assessment. At presentation, hospitalised patients were older, had higher levels of obesity and increased rates of hypertension than those managed virtually (p<0.05). Among hospitalised patients, the degree of pulmonary infiltration on CXR was higher, ROX index for intubation lower, eGFR lower, C-reactive protein levels higher and lymphocyte counts lower, compared to those managed virtually (p<0.05). The prevalence of known respiratory conditions was higher among patients admitted to hospital, with a trend towards statistical significance (p=0.051). There were no differences in the prevalence of known cardiac disorders and other co-morbidities amongst both patient groups (table 1).

At follow up, CXR appearances were improved and similar among patients monitored virtually, those admitted for supplemental oxygen, those requiring non-invasive ventilation and those treated on ICU. There were no differences in heart rhythm and ECG parameters in the four patient groups. Left ventricular systolic and diastolic dimensions, Simpson's biplane ejection fraction, left atrial volume and left ventricular filling pressures were similar in all four patient groups. There were no differences in right ventricular dimensions, right ventricular fractional area change, tricuspid annular plane systolic excursion and pulmonary artery systolic pressures among patients irrespective of disease severity (table 2).

Conclusions

At 3 months, we identified no differences in ECG indices and echocardiographic parameters of left and right ventricular function among survivors of COVID-19, independent of disease course. The findings of this study argue against significant cardiac sequalae following COVID-19 infection.

Funding Acknowledgement

Type of funding sources: None.

Fully automated global longitudinal strain assessment using artificial intelligence developed and validated by a UK-wide echocardiography expert collaborative

Background

Left ventricular longitudinal strain has been reported to deliver reproducibility, sensitivity and prognostic value over and above ejection fraction. However, it currently relies on uninspectable proprietary algorithms and suffers from a lack of widespread clinical use. Uptake may be improved by increasing user trust through greater transparency.

Purpose

We therefore developed a machine-learning based method, trained, and validated with accredited experts from our AI Echocardiography Collaborative. We make the dataset, code, and trained network freely available under an open-source license.

Methods

AI enables strain to be calculated without relying on speckle tracking by directly locating key points and borders across frames. Strain can then be calculated as the fractional shortening of the left ventricular perimeter. We first curated a dataset of 7523 images, including 2587 apical four chamber, each labelled by a single expert from our collaboration of 17 hospitals, using our online platform (Figure 1). Using both this dataset and a semi-supervised approach, we trained a 3d convolutional neural network to identify the annulus, apex, and the endocardial border throughout the cardiac cycle.

Separately, we constructed an external validation dataset of 100 apical 4 chamber video-loops. The systolic and diastolic frame were identified, and each image was separately labelled by 11 experts. From these labels we then derived the expert consensus strain for each of the 100 video loops. These experts also ordered all 100 echocardiograms by their visual grading of left ventricular longitudinal function. Finally, a single expert calculated strain using two different proprietary commercial packages (A and B).

Results

Consensus strain measurements (obtained by averaging individual assessments by the 11 experts) across the 100 cases ranged from −4% to −27%, with strong correlations with the individual experts and machine methods (Figure 2). Using each cases' consensus across experts as the gold standard, median error from consensus was 3.1% for individual experts, 3.4% for Propriety A, 2.6% for Proprietary B, 2.6% for our AI.

Using the visual grading of longitudinal strain as the reference, the 11 individual experts and 4 machine methods each showed significant correlation: coefficients ranged from 0.55 to 0.69 for experts, and for Proprietary A was 0.68, Proprietary B 0.69, and our AI 0.69.

Conclusions

Our open-source, vendor-independent AI-based strain measure automatically produces values that agree with expert consensus, as strongly as the individual experts do. It also agrees with the subjective visual ranking by longitudinal function. Our open-source AI strain performs at least as well as closed-source speckle-based approaches, and may enable increased clinical and research use of longitudinal strain.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NIHR Imperial BRC ITMAT.Dr Howard was additionally funded by Wellcome. Figure 1. Collaborative online platformFigure 2. Correlations between strain methods

Automated Left Ventricular Dimension Assessment Using Artificial Intelligence Developed and Validated by a UK-Wide Collaborative

BACKGROUND

requires training and validation to standards expected of humans. We developed an online platform and established the Unity Collaborative to build a dataset of expertise from 17 hospitals for training, validation, and standardization of such techniques.

METHODS

The training dataset consisted of 2056 individual frames drawn at random from 1265 parasternal long-axis video-loops of patients undergoing clinical echocardiography in 2015 to 2016. Nine experts labeled these images using our online platform. From this, we trained a convolutional neural network to identify keypoints. Subsequently, 13 experts labeled a validation dataset of the end-systolic and end-diastolic frame from 100 new video-loops, twice each. The 26-opinion consensus was used as the reference standard. The primary outcome was precision SD, the SD of the differences between AI measurement and expert consensus.

RESULTS

In the validation dataset, the AI's precision SD for left ventricular internal dimension was 3.5 mm. For context, precision SD of individual expert measurements against the expert consensus was 4.4 mm. Intraclass correlation coefficient between AI and expert consensus was 0.926 (95% CI, 0.904-0.944), compared with 0.817 (0.778-0.954) between individual experts and expert consensus. For interventricular septum thickness, precision SD was 1.8 mm for AI (intraclass correlation coefficient, 0.809; 0.729-0.967), versus 2.0 mm for individuals (intraclass correlation coefficient, 0.641; 0.568-0.716). For posterior wall thickness, precision SD was 1.4 mm for AI (intraclass correlation coefficient, 0.535 [95% CI, 0.379-0.661]), versus 2.2 mm for individuals (0.366 [0.288-0.462]). We present all images and annotations. This highlights challenging cases, including poor image quality and tapered ventricles.

CONCLUSIONS

Experts at multiple institutions successfully cooperated to build a collaborative AI. This performed as well as individual experts. Future echocardiographic AI research should use a consensus of experts as a reference. Our collaborative welcomes new partners who share our commitment to publish all methods, code, annotations, and results openly.

Role of T1 mapping in identification of convalescent Takotsubo cardiomyopathy

Funding Acknowledgements

Type of funding sources: None.

Background

Takotsubo cardiomyopathy remains an elusive entity to diagnose, especially with increasing time following the initial clinical event. Patients are often referred for cardiac MR, but when performed on a convalescent patient, the conventional CMR exam can be normal. However CMR elevated native T1 values on mapping may persist, aiding diagnosis.

Purpose

We sought to review a series of confirmed Takotsubo cases to evaluate which CMR features were most helpful in making a diagnosis.

Methods

We reviewed 2 years of CMR exams in our institution and identified 14 patients with a confirmed diagnosis of Takotsubo cardiomyopathy after a read of all clinical data by two consultant cardiologists. All of the 14 patients had a troponin positive cardiac event, unobstructed coronary arteries on invasive angiography and echocardiographic evidence of apical hypo/akinesia (or basal, in the non-apical variants) at the time of presentation. All CMR exams were performed at 1.5T. All included functional assessment, T1 mapping and late gadolinium enhancement(LGE). Some exams also included T2 mapping and STIR.

Results

All patients were female. The time interval between presentation and CMR exam varied from 24 hours to 2 years. 2 of the scans showed evidence of non-apical Takotsubo cardiomyopathy. At the time of CMR, only 7 patients(50%) had a residual regional wall motion abnormality, but 13 patients(93%) had a region of elevated native T1 (>1 myocardial segment). High signal was identified on STIR in 6(43%) and elevated native T2 in 10(71%). Of interest, we identified small areas of patchy LGE in 2 patients(14%).

Conclusion

Increased native T1 signal may be the only residual cardiac MRI marker of previous Takotsubo cardiomyopathy, persisting for months after the initial cardiac event, even after resolution of regional wall motion abnormalities and recovery of the left ventricular function. T1 mapping should be performed in CMR exams for Takotsubo cardiomyopathy to improve diagnostic yield.

Abstract Figure. CMR in convalescent Takotsubo

Diffusion tensor cardiovascular magnetic resonance predicts adverse remodelling after myocardial infarction

Background

Changes in myocardial microstructure that underlie post-myocardial infarction (MI) left ventricular (LV) remodelling may contribute to progressive deterioration in cardiac function and increased risk of adverse clinical events. Diffusion Tensor Cardiovascular Magnetic Resonance (DT-CMR) derived parameters provide in vivo measures of helix angle (HA) and sheetlet angle (SA), which allow non-invasive characterization of microstructural dynamics associated with cardiac contraction.

Purpose

To evaluate the relationship between DT-CMR metrics of myocardial microstructure with the development of adverse LV remodelling and to determine the relationship between DT-CMR and strain post- MI.

Methods

We performed a longitudinal pre-clinical CMR study whereby DT-CMR, cine imaging and strain were acquired pre-MI and 3 days and 16 weeks post- MI in a pig model. HA (E1A), SA (E2A), and sheetlet angle mobility (ΔE2A = E2Asystole– E2Adiastole), as well as circumferential (Ecc) and radial (Err) strain were calculated at each timepoint and related to change in left ventricular end-diastolic volume (ΔLVEDV) and change in left ventricular end-systolic volume (ΔLVESV) between 3 days post-MI and 16 weeks post-MI. The translational value of this preclinical study was further assessed in six patients with chronic MI.

Results

ΔE2A over the whole LV (global ΔE2A) at 3 days post-MI correlated significantly with ΔLVEDV (R2=0.89, p=0.0013, Fig.1A;) and ΔLVESV (R2=0.81, p=0.0055, Fig.1B). Global Ecc at 3 days post-MI also correlated with both ΔLVEDV (R2=0.75, p=0.012, Fig.1C) and ΔLVESV (R2=0.71, p=0.018, Fig.1D). Global Err at 3 days post-MI did not show significant correlation with either ΔLVEDV (R2=0.32, p=0.19, Fig.1E) or ΔLVESV (R2=0.35, p=0.17, Fig.1F). Global ΔE2A correlated strongly with global Ecc 3 days post-MI (R2=0.9, p=0.00099, Fig.1G) but less strongly with global Err 3 days post-MI (R2=0.57, p=0.049, Fig.1H). Global ΔE2A at the chronic stage correlated significantly with ejection Fraction (EF), in both clinical (R2=0.87, p=0.007) and preclinical data (R2=0.87, p=0.0024). Global ΔE2A correlated well with LVEDV (clinical: R2=0.72, p=0.033; preclinical: R2=0.8, p=0.0066) and LVESV (clinical: R2=0.78, p=0.020; preclinical: R2=0.89, p=0.0013). In vivo E1A maps at 16 weeks post-MI and ex vivo DT-CMR demonstrated reduced right-handed helix angles in the endocardium of the infarct region.

Conclusion

ΔE2A measured at 3 day post-reperfused MI is strongly correlated with the development of increased end-systolic and end-diastolic volumes, and may therefore serve as a novel CMR early predictor of adverse LV remodelling after reperfused MI. Strong correlations between ΔE2A, LV volumes and EF in a small cohort of stable patients with remodelled hearts after chronic MI confirm the feasibility of performing these measurements in patients and the plausibility of further evaluation of ΔE2A as a predictor of adverse remodelling after reperfused MI.

Sheetlet mobility predicts volume change

Funding Acknowledgement

Type of funding source: Foundation. Main funding source(s): National Institutes of Health by the Division of Intramural Research (NHLBI, NIH, DHHS); British Heart Foundation

1641 Bicuspid pulmonary valve - an unusual cause of dilated pulmonary artery

Introduction

Dilated pulmonary artery is a common finding with a variety of possible aetiologies. We present a case of unexplained dilated pulmonary artery that was associated with bicuspid pulmonary valve identified by CMR.

Case Presentation

A 67-year-old female underwent mitral valve repair for severe mitral regurgitation. She had presented with breathlessness, and had severe mitral regurgitation due to a myxomatous valve with bileaflet prolapse, confirmed with TOE. Echo showed: normal LV dimensions, LVEF 66%. RV size and systolic function normal. Coronary angiography was normal and pulmonary hypertension was not present on right heart catheterization.

At postoperative follow up she was well and asymptomatic, with echo showing good LV systolic function, intact mitral valve repair, and normal systolic pulmonary artery pressures (36mmHg). New T wave inversion was identified across the precordial leads so a CMR was performed.

CMR showed normal LV volumes and function with no wall motion abnormality or LVH. There was no fibrosis or infarction in the late phase after Gadolinium contrast. The RV was dilated (73mm diameter at base) with normal systolic function. The main pulmonary artery was dilated at 42mm. Flow mapping showed no evidence of shunt (Qp:Qs = 1:1) and the atrial septum was intact. Pulmonary valve function was normal but the valve was bicuspid in morphology.

Discussion

Bicuspid pulmonary valve is a rare finding. In our case this was identified by CMR and was associated with a dilated pulmonary artery.

Bicuspid pulmonary valve is most commonly identified in the context of pulmonary valve stenosis. Isolated pulmonary valve stenosis is a rare finding; it is most often associated with other congenital defects like Fallot. Pulmonary stenosis has been associated with pulmonary artery aneurysm, attributed to hemodynamic changes caused by the stenotic valve. However as with bicuspid aortic valve it is recognized that the "post stenotic dilation" is part of the pathology of bicuspid valve and does not need the changed haemodynamics of a stenotic valve.

During embryogenesis neural crest cells are necessary for the normal development of the semilunar valves, the septation of the outflow tract into the aorta and pulmonary artery, and the remodeling of the aortic arch. Any defect in this migration can cause abnormalities or aneurysm of the aorta or pulmonary artery together with semilunar valve defects. A study of 3,861 donor hearts dissected at the European Homograft Bank revealed only four cases (0.1%) of bicuspid pulmonary valves.

Bicuspid pulmonary valve with normal function can be difficult to identify at echocardiography. It can be well visualized by CMR and this can be combined with an assessment of the pulmonary arteries. It should be remembered as a rarer cause of dilated pulmonary artery when other causes have been excluded.

Figure

A. Dilated MPA. B. Dilated RV. C. and D. Bicuspid pulmonary valve.

Abstract 1641 Figure. Dilated MPA. Bicuspid pulmonary valve

P2243CTCA alone demonstrates superior diagnostic accuracy, prognostic utility and is less expensive than CTCA combined with subsequent multi-modality functional imaging in patients with ischaemic symptoms

Background

The evaluation of suspected ischaemic symptoms incorporates multi-modality anatomical and functional imaging tests. The 2016 update to the UK's NICE guidelines recommends CT coronary angiogram (CTCA) first line in patients without known coronary artery disease. Additive multi-modality functional imaging may provide synergistic diagnostic and prognostic information.

Purpose

To investigate the diagnostic accuracy, prognostic utility and cost of CTCA combined with subsequent multi-modality functional testing versus (vs) CTCA alone.

Methods

772 consecutive patients were referred to a single UK tertiary centre with symptoms suggestive of ischaemia. 657 individuals (“CTCA group”) underwent CTCA alone, and 115 individuals (“Combined group”) underwent CTCA and then either perfusion cardiac MRI (n=25), stress echocardiogram (n=16), or myocardial perfusion scintigraphy (n=74). Patients underwent invasive angiography (n=79) +/− revascularisation at the discretion of the referring clinician. All readers and operators were aware of previous imaging findings. Revascularised patients (n=52) were excluded from long term follow-up. The remaining patients were followed-up for a mean of 38.1±17.4 months and the incidence of major adverse cardiovascular events (MACE) recorded. Costs were derived from the NICE guidelines.

Results

Baseline characteristics were similar between groups. The Combined group underwent significantly more invasive angiograms than the CTCA group (29.6% vs 6.8%, p=0.0001) with no significant difference in the rate of revascularisation (73% vs 67%, p=0.72). Mean time from CTCA to angiogram was significantly longer in the Combined group (81.2 vs 38.1 days, p=0.0001). Both sensitivity and specificity were lower in the Combined group than in the CTCA group (sensitivity: 70% vs 93%, specificity: 75% vs 100%). The rate of long term MACE was significantly higher in the Combined group (8.7% vs 2.6%, p=0.0026). Multivariate analysis of CTCA and functional imaging findings found that CTCA-derived four vessel aggregate stenosis score (0–12) was the strongest predictor of MACE for the whole cohort (OR 4.4, p<0.0001), and also for those with negative functional tests (OR 3.9, p<0.0001). Per patient, a combined strategy was more expensive than CTCA alone (£1551 vs £368, p=0.0001).

CTCA and functional data vs outcomes

Conclusions

Combining multi-modality functional testing with CTCA increased costs but did not improve diagnostic accuracy or long term outcomes. Further reductions in both MACE and unnecessary invasive angiography are desirable; CT-derived functional data such as FFRCT may be implicated.

A Low-Frequency Inactivating AKT2 Variant Enriched in the Finnish Population Is Associated With Fasting Insulin Levels and Type 2 Diabetes Risk

To identify novel coding association signals and facilitate characterization of mechanisms influencing glycemic traits and type 2 diabetes risk, we analyzed 109,215 variants derived from exome array genotyping together with an additional 390,225 variants from exome sequence in up to 39,339 normoglycemic individuals from five ancestry groups. We identified a novel association between the coding variant (p.Pro50Thr) in AKT2 and fasting plasma insulin (FI), a gene in which rare fully penetrant mutations are causal for monogenic glycemic disorders. The low-frequency allele is associated with a 12% increase in FI levels. This variant is present at 1.1% frequency in Finns but virtually absent in individuals from other ancestries. Carriers of the FI-increasing allele had increased 2-h insulin values, decreased insulin sensitivity, and increased risk of type 2 diabetes (odds ratio 1.05). In cellular studies, the AKT2-Thr50 protein exhibited a partial loss of function. We extend the allelic spectrum for coding variants in AKT2 associated with disorders of glucose homeostasis and demonstrate bidirectional effects of variants within the pleckstrin homology domain of AKT2.