Visual Ejection Fraction Research Articles

Pragmatic Evaluation of a Deep-Learning Algorithm to Automate Ejection Fraction on Hand-Held, Point-of-Care Echocardiography in a Cardiac Surgical Operating Room

ObjectiveTest the correlation of ejection fraction (EF) estimated by a deep learning-based, automated algorithm (Auto EF) vs an EF estimated by Simpson's method. DesignProspective observational study. SettingSingle-center study at the Hospital of the University of Pennsylvania. ParticipantsStudy participants were ≥18 years of age, scheduled to undergo valve, aortic, coronary artery bypass graft (CABG), heart or lung transplant surgery. InterventionsThis non-interventional study involved acquiring apical 4 Chamber (A4C), transthoracic echocardiographic clips using Philips hand-held ultrasound device; Lumify. Measurements and Main ResultsIn the primary analysis of 54 clips, compared to the Simpson's method for EF estimation, bias was similar for Auto EF (-10.17%) and the experienced reader-estimated EF (-9.82%), but correlation was lower for Auto EF (r = 0.56) than the experienced reader-estimated EF (r = 0.80). In the secondary analyses, correlation between EF estimated by Simpson's method and Auto EF increased when applied to 27 acquisitions classified as adequate (r = 0.86) but decreased when applied to 27 acquisitions classified as inadequate (r = 0.46). ConclusionsApplied to acquisitions of adequate image quality, Auto EF produced a numerical EF estimate equivalent to Simpson's method. However, when applied to acquisitions of inadequate image quality, discrepancies arose between EF estimated by Auto EF and Simpson's method. Visual EF estimates by experienced readers correlated highly with Simpson's method in both variable and inadequate imaging conditions – emphasizing its enduring clinical utility. Trial RegistrationClinicaltrials.gov trial identifier: NCT04943965

Abstract 14515: Initial Experience With Artificial Intelligence-Guided Echocardiography Acquisition in Rural America: Implementation and Quality Assessment

Background: Use of artificial intelligence-guided echocardiography (AI echo) may increase access to imaging, but there is little experience of its application in rural and low resource settings. The Risk Underlying Rural Areas Longitudinal (RURAL) cohort is the first NHLBI population-based cohort study to employ AI echo. Hypothesis: AI echo is feasible and produces adequate quality images to assess cardiac structure and function among rural populations. Methods: The RURAL study, in partnership with Caption Health (Brisbane, CA), is using AI echo in a multiethnic cohort of 4600 participants, performed in a mobile exam unit (MEU) in 10 rural U.S. communities. Non-sonographer MEU technicians underwent 10 hrs in-person competency-based training before scanning. Cardiac structure and function were analyzed in an independent core laboratory. Ejection fraction (EF) was visually estimated and calculated using Caption Health’s Auto EF technology. Results: Overall, 138 participants had AI echoes analyzed from Sept 2021 to May 2022 of whom 62% were women, with 70% obese (body mass index (BMI) ≥ 30kg/m 2 ), median BMI 33.8kg/m 2 (Table 1). Median time per scan was 20.0mins (15.5, 29.7). Image quality was adequate for visual EF in 97%, with left ventricular dimensions measurable in 88% and left atrial diameter in 91%. Adequate images of the right heart and ascending aorta were less common: base of the right ventricle measurable in 62%, right atrium 61% and ascending aorta 60%. Most participants (96%) had LVEF ≥ 50% by visual estimation, and there was 96% agreement between visual and Auto EF for this group. Image quality, but not measurability, varied with BMI. Conclusion: AI echo imaging technology can be used by non-sonographers to acquire adequate quality images characterizing cardiac structure and function in a rural and predominantly obese population, suggesting utility across the spectrum of BMI and applicability in low resource environments with limited access to healthcare.

AI do it better. Reproducibility of fully automated versus manual analysis of left ventricle function using a new software

Abstract Funding Acknowledgements Type of funding sources: Public hospital(s). Main funding source(s): University Hospital of Padua Background/Introduction: Ejection fraction (EF) is a parameter widely used in Echolab to evaluate left ventricular function. Recently, in parallel with the growing interest in artificial intelligence (AI), attemps have been made to create automated systems for EF assessment, in order to reduce time and improve the accuracy of the analysis. Purpose to compare results of different methods of EF assessment: visual estimation (visual EF), manual and fully automated analysis. Methods 28 consecutive pediatric patients were enrolled. This cohort of previously healthy patients was screened at our Center for cardiac evaluation within 6 months after an asymptomatic or paucisymptomatic COVID19 infection. All they were in sinus rhythm. Optimized apical 4- and 2- chamber views were collected for each patient using Canon Aplio i900. Off-line EF assessment was first evaluated visually by pediatric cardiologists with experience in echocardiography, then performed by both fully automated analysis (AI) using two different methods (Automatic Simpson -AI Simpson- and Wall Motion Tracking -AI WMT-) and pediatric cardiologists through manual tracing of endocardial border (Manual Simpson and Manual WMT respectively). Operators were blinded to the AI analysis. To measure intraobserver variability, evaluations of 16 patients’ datasets were performed twice by both operators and AI. Results Patients’ demographic data were: age 9,8+/-4,7 years; males 22 (78%); height 134,3+/- 34,9 cm; weight 41,8+/-28,7 kg; BSA 1,2+/-0,4 mq, HR 85+/-15/min. The time taken for off-line analysis was 0.3-0.7 minutes, 1-1.5 minutes, 1-3 minutes and 3-4 minutes, respectively for AI WMT, AI Simpson, Manual WMT and Manual Simpson. As expected, visual EF showed high intraobserver variability and a poor reproducibility (ICC 43%). AI analysis revealed a good to excellent reproducibility (ICC from 80% to 99%, depending on the method used). WMT methods had the best reproducibility both for manual tracing of endocardial border and fully automated analysis (Table 1). The comparison between different methods (Table 2) showed a good agreement between AI Simpson and AI WMT (mean bias 2,9, from -3,2 to 9,0, ICC 86%). A moderate correlation was found between different methods of AI analysis while only poor correlation was found between manual Simpson and manual WMT (Table 2). Conclusion(s): Automatic Simpson and Wall Motion Tracking are two different fully automated methods which can be used for left ventricular function assessment. AI reproducibility is high for both methods, higher for WMT. WMT method is also less time consuming and improves reproducibility of manual tracing of endocardial borderd analysis. Abstract Table 1. Echocardiographic measurements Abstract Table 2. Comparison between different me

139 Creating an Automated Artificial Intelligence Left Ventricular Ejection Fraction Estimator for Parasternal Long Axis Scans from Unrelated Apical Window Echo Videos

A major hindrance to point-of-care ultrasound (POCUS) deep learning (DL) algorithm development is lack of large publicly available image databases, like ones found in traditional imaging. Our objective was to test the potential of using only apical 4-chamber (A4C) images from a large publically available dataset to train a DL algorithm to visually estimate left ventricle (LV) Ejection Fraction (EF) from a parasternal long axis (PSL) window. Researchers embedded a VGG16 convolutional neural network inside a long short term memory (LSTM) algorithm for ultrasound video analysis. We obtained access to the Stanford EchoNet-Dynamic database of approximately 10, 000 echo A4C videos. All echo examinations were read during comprehensive echocardiography examination and included calculated EF results. Ninety percent of the data were used for algorithm training and 10% for validation during training. The LV in A4C images take on a similar appearance to that in PSL images when flipped horizontally, across the vertical axis, and rotated 90o counter clockwise (CCW) and could potentially be used to simulate PSL data. As part of DL algorithm training, researchers tested 3 training options by applying image manipulation to Stanford A4C echo videos. Tested were: training the DL algorithm only on unaltered Stanford A4C videos, training on unaltered videos and those rotated 90o CCW and finally training on unaltered then rotated 90o CCW and then horizontally flipped A4C videos. As a real-world test, we obtained 569 echo examinations from a different medical center (UCLA) showing PSL window videos from comprehensive echo examinations, with respective EF, and tested the DL algorithm variants’ performance on these actual PSL videos. We calculated mean absolute error (MAE) on algorithm EF results per field standards for evaluating algorithm accuracy compared to Echo Lab calculated EF and Bland-Altman analyses. MAE for skilled echo techs ranges from 4 to 5%. Prior algorithm training with Stanford A4C videos and testing only on Stanford A4C videos achieved an MAE of 8.08% (95% CI 7.60 to 8.55). In this study, the DL algorithm trained only on unaltered A4C videos and tested on unrelated PSL achieved MAE of 27.03% (95% CI 25.59 to 28.46) for visual EF estimation. Training on unaltered A4C videos and then on 90o CCW rotated videos achieved an MAE of 17.41 (95% CI 16.26 to 18.19). Training on unaltered, then 90o CCW rotated and then horizontally flipped videos achieved an MAE of 16.18 (95% CI 15.21 to 17.14). The Bland-Atlman plots showed the vast majority of points falling within the 95% CI for the third and best training iteration. Our study results indicate the potential for POCUS AI researchers to use non-POCUS image data in algorithm development by adapting it for training purposes via video rotation and manipulation designed to simulate the desired imaging window. This may be important for future POCUS algorithm development which may benefit from adaptation of traditional imager data, and help overcome the paucity of POCUS databases.

Evaluating the Association Between the Three Different Ejection Fraction Measurement Techniques and Left Ventricle Global Strain.

The prognosis of cardiovascular diseases (CVDs) is directly associated with systolic function based on the measurement of ejection fraction (EF), and many studies have indicated that the left ventricular global strain (LVGS) provides better predictivity than the EF measurement in the diagnosis, prognosis, survival, and CVD staging. However, these studies did not investigate the correlation between the EF measurement and the LVGS parameters, or which parameters are better correlated with LVGS, but we analyzed the association between three EF measurement methods and LVGS. This study included 62 patients that applied to the clinic between October 2015 and March 2016. An echocardiography examination of these patients was performed. The exclusion criteria were atrial fibrillation and suboptimal image quality. Sixty-two patients (the average age 61.0±12.6 years; 56% male and 44% female) were enrolled in the study. A statistically significant association was found between the visual EF and Simpson EF measurements and the LVGS parameters (p<0.001). While the visual EF was moderately correlated with the LVGS parameters (r=0.44), there was a good correlation between the Simpson EF and the LVGS parameters (r=0.710). In this study, we demonstrate that the Simpson's rule LVEF correlates better with LVGS than the Teicholtz method or visual EF and that it has a better area under the curve value for determining an abnormal LVGS. Therefore, we recommend the use of the Simpson EF for the EF measurement that has a better correlation with the LVGS values in the patients whose ventricle functions should be evaluated.

Visual versus fully automated assessment of left ventricular ejection fraction

The aim of this study is to compare three different methods commonly used in the assessment of left ventricle (LV) ejection fraction (EF) by echocardiography. all patients underwent full echocardiography imaging that includes assessment of LVEF using M-mode, Automated EF (Auto-EF) through tracing the myocardial borders during systole and diastole, and visual EF estimation by two readers. We enrolled 268 patients. Auto-EF measurement was feasible in 240 (89.5%) patients. The averaged LVEF was (52%12) with the visual assessment, (51%11) with Auto-EF and (57%13) with M-mode. Using Bland-Altman analysis we found that the difference between the mean visual and the Auto-EF was not significant [−0.3% (−0.5803–0.0053), p = 0.054]. However, we found a significant difference in the mean EF between the visual versus M-mode and Auto-EF versus M-mode with the mean differences: [−2.4365(−2.9946–1.8783), p < 0.0001] and [−2.1490 (−2.7348–1.5631), p < 0.0001] respectively. Inter-observer variability analysis of the visual EF assessment between the two readers showed that intraclass correlation coefficient was 0.953, (95% confidence interval: 0.939–0.965, p < 0.0001), with excellent correlation between the two readers: R = 0.911, p < 0.0001). The two-dimensional echocardiographic methods using Biplane Auto-EF or visual assessment were significantly comparable, whereas M-mode results in an overestimation of the LV ejection fraction.

Visual versus fully automated assessment of left ventricular ejection fraction.

Introduction:The aim of this study is to compare three different echocardiographic methods commonly used in the assessment of left ventricle (LV) ejection fraction (EF).Methods:All patients underwent full echocardiography including LVEF assessed using M-mode, automated EF (Auto-EF), and visual estimation by two readers.Results:We enrolled 268 patients. Auto-EF measurement was feasible in 240 (89.5%) patients. The averaged LVEF was (52% ± 12) with the visual assessment, (51% ± 11) with Auto-EF and (57% ± 13) with M-mode. Using Bland-Altman analysis, we found that the difference between the mean visual and the Auto-EF was not significant (−0.3% [−0.5803–0.0053], P = 0.054). However, the mean EF was significantly different when comparing visual versus M-mode and Auto-EF versus M-mode with the mean differences: (−2.4365 [−2.9946–−1.8783], P < 0.0001) and (−2.1490 [−2.7348–−1.5631], P < 0.0001) respectively. Inter-observer variability analysis of the visual EF assessment between the two readers showed that intraclass correlation coefficient was 0.953, (95% confidence interval: 0.939–0.965, P < 0.0001), with excellent correlation between the two readers: R = 0.911, P < 0.0001).Conclusion:The two-dimensional echocardiographic methods using Biplane Auto-EF or visual assessment were significantly comparable, whereas M-mode results in an overestimation of the LVEF.

Fully Automated Versus Standard Tracking of Left Ventricular Ejection Fraction and Longitudinal Strain: The FAST-EFs Multicenter Study

BackgroundEchocardiographic determination of ejection fraction (EF) by manual tracing of endocardial borders is time consuming and operator dependent, whereas visual assessment is inherently subjective. ObjectivesThis study tested the hypothesis that a novel, fully automated software using machine learning-enabled image analysis will provide rapid, reproducible measurements of left ventricular volumes and EF, as well as average biplane longitudinal strain (LS). MethodsFor a total of 255 patients in sinus rhythm, apical 4- and 2-chamber views were collected from 4 centers that assessed EF using both visual estimation and manual tracing (biplane Simpson’s method). In addition, datasets were saved in a centralized database, and machine learning-enabled software (AutoLV, TomTec-Arena 1.2, TomTec Imaging Systems, Unterschleissheim, Germany) was applied for fully automated EF and LS measurements. A reference center reanalyzed all datasets (by visual estimation and manual tracking), along with manual LS determinations. ResultsAutoLV measurements were feasible in 98% of studies, and the average analysis time was 8 ± 1 s/patient. Interclass correlation coefficients and Bland-Altman analysis revealed good agreements among automated EF, local center manual tracking, and reference center manual tracking, but not for visual EF assessments. Similarly, automated and manual LS measurements obtained at the reference center showed good agreement. Intraobserver variability was higher for visual EF than for manual EF or manual LS, whereas interobserver variability was higher for both visual and manual EF, but not different for LS. Automated EF and LS had no variability. ConclusionsFully automated analysis of echocardiography images provides rapid and reproducible assessment of left ventricular EF and LS.

High-risk percutaneous coronary intervention is associated with reverse left ventricular remodeling and improved outcomes in patients with coronary artery disease and reduced ejection fraction

Therapies that reverse pathologic left ventricular (LV) remodeling are often associated with improved outcomes. The incidence and impact of reverse LV remodeling after high-risk percutaneous coronary intervention (PCI) are unknown. The PROTECT II study was a multicenter trial in patients with complex, multivessel coronary artery disease and reduced ejection fraction (EF) that revealed an increase in visual EF after high-risk PCI. Among patients with quantitative echocardiography (LV volumes and biplane EF), we assessed the extent and predictors of reverse LV remodeling, defined as improved systolic function with an absolute increase in EF ≥5% and correlated these findings with clinical events. Quantitative echocardiography was performed in 184 patients at baseline and longest follow-up. Mean EF at baseline was 27.1%. Ninety-three patients (51%) demonstrated reverse LV remodeling with an absolute increase in EF of 13.2% (P < .001). End-systolic volume decreased from 137.7 to 106.6 mL (P = .002). No significant change in EF or end-systolic volume was seen among non-remodelers. Reverse LV remodeling occurred more frequently in patients with more extensive revascularization (odds ratio, 7.52; 95% CI [1.31-43.25]) and was associated with significantly fewer major adverse events (composite of death/myocardial infarction/stroke/transient ischemic attack): 9.7% versus 24.2% (P = .009). There was also a greater reduction in New York Heart Association class III/IV heart failure among reverse LV remodelers (66.7% to 24.0%) than non-remodelers (56.3% to 34.4%), P = .045. Reverse LV remodeling can occur after high-risk PCI in patients with complex coronary artery disease and reduced EF and is associated with improved clinical outcomes.

Performance of Conventional Echocardiographic Parameters and Myocardial Measurements in the Sequential Evaluation of Left Ventricular Function

Echocardiography is frequently used for sequential evaluation of left ventricular (LV) function, although the reproducibility of such conventional measurements as LV ejection fraction (EF) have been questioned. The utility of such newer measurements as tissue Doppler imaging and left atrial (LA) size in serial clinical testing are undefined. The magnitude and clinical relevance of changes in conventional and new measurements of LV function were investigated and compared in 346 consecutive patients undergoing sequential echocardiography. Change in LA area, LVEF, tissue E velocity (Em), and transmitral E to Em ratio (E/Em) were compared over 304 +/- 239 days. Changes within and between parameters (after mean correction to make measurements comparable) were assessed in groups designated as stable (n = 144) or unstable (n = 202) according to clinical progress. A single observer remeasured these parameters in stable patients individually and with paired studies side by side. Significant variability was seen in all measurements, with change in LVEF the only parameter differing between stable and unstable groups (6.4 +/- 8.9% vs 9.4 +/- 5.4%; p <0.001). Tissue Em and E/Em ratio were more variable than LA area or LVEF. In stable patients, LVEF changed the least and E/Em changed the most over time (p <0.05). With a single blinded observer, Em had improved reproducibility (1.5 +/- 1.9 vs 2.3 +/- 2.6 cm/s; p <0.001), as did visual EF. In conclusion, variability in all measurements was high irrespective of clinical status. Newer measurements were no better than LVEF in detecting changes in clinical status. Sequential LV assessment should be interpreted with caution, and more robust measurements are needed.

A Novel Two-Dimensional Echocardiographic Image Analysis System Using Artificial Intelligence-Learned Pattern Recognition for Rapid Automated Ejection Fraction

We sought to test the hypothesis that a novel 2-dimensional echocardiographic image analysis system using artificial intelligence-learned pattern recognition can rapidly and reproducibly calculate ejection fraction (EF). Echocardiographic EF by manual tracing is time consuming, and visual assessment is inherently subjective. We studied 218 patients (72 female), including 165 with abnormal left ventricular (LV) function. Auto EF incorporated a database trained on >10,000 human EF tracings to automatically locate and track the LV endocardium from routine grayscale digital cineloops and calculate EF in 15 s. Auto EF results were independently compared with manually traced biplane Simpson's rule, visual EF, and magnetic resonance imaging (MRI) in a subset. Auto EF was possible in 200 (92%) of consecutive patients, of which 77% were completely automated and 23% required manual editing. Auto EF correlated well with manual EF (r = 0.98; 6% limits of agreement) and required less time per patient (48 +/- 26 s vs. 102 +/- 21 s; p < 0.01). Auto EF correlated well with visual EF by expert readers (r = 0.96; p < 0.001), but interobserver variability was greater (3.4 +/- 2.9% vs. 9.8 +/- 5.7%, respectively; p < 0.001). Visual EF was less accurate by novice readers (r = 0.82; 19% limits of agreement) and improved with trainee-operated Auto EF (r = 0.96; 7% limits of agreement). Auto EF also correlated with MRI EF (n = 21) (r = 0.95; 12% limits of agreement), but underestimated absolute volumes (r = 0.95; bias of -36 +/- 27 ml overall). Auto EF can automatically calculate EF similarly to results by manual biplane Simpson's rule and MRI, with less variability than visual EF, and has clinical potential.

Development of an ultra-portable echo device connected to USB port

In practical cardiology, a stethoscope based auscultation has been used to reveal the patient's clinical status. Recently, several hand-held echo devices are going on market and they are expected to play a role as “visible” auscultation instead of stethoscope. We have developed a portable and inexpensive echo device which can be used for screening of cardiac function. Two single element transducers were attached 180° apart to a rotor with 14-mm diameter. The mechanical scanner, integrated circuits for transmitting and receiving ultrasonic signals and an A/D converter were encapsulated in a 150 × 40 mm probe weighing 200 g. The scan was started and the image was displayed on a Windows based personal computer (PC) as soon as the probe was connected to USB 2.0 port of the PC. The central frequency was available between 2.5 and 7.5 MHz, the image depth was 15 cm and the frame rate was 30/s. The estimated price of this ultra-portable ultrasound is about 3000 US$ with software. For 69 cardiac patients with informed consent, image quality was compared with those obtained with basic range diagnostic echo machines. Left ventricular ejection fraction (EF) derived from normal M-mode image of standard machines (EFm) were compared with visual EF of the ultra-portable ultrasound device (EFv). The image quality was comparable to the basic range diagnostic echo machines although short axis view of aortic root was not clearly visualized because the probe was too large for intercostal approach. EFv agreed well with EFm. The ultra-portable ultrasound may provide useful information on screening and health care.