Deep Learning for Assessment of Left Ventricular Ejection Fraction from Echocardiographic Images

Development of automated neural network prediction for echocardiographic left ventricular ejection fraction

Introduction

The echocardiographic measurement of left ventricular ejection fraction (LVEF) is fundamental to the diagnosis and classification of patients with heart failure (HF).

Methods

This paper aimed to quantify LVEF automatically and accurately with the proposed pipeline method based on deep neural networks and ensemble learning. Within the pipeline, an Atrous Convolutional Neural Network (ACNN) was first trained to segment the left ventricle (LV), before employing the area-length formulation based on the ellipsoid single-plane model to calculate LVEF values. This formulation required inputs of LV area, derived from segmentation using an improved Jeffrey's method, as well as LV length, derived from a novel ensemble learning model. To further improve the pipeline's accuracy, an automated peak detection algorithm was used to identify end-diastolic and end-systolic frames, avoiding issues with human error. Subsequently, single-beat LVEF values were averaged across all cardiac cycles to obtain the final LVEF.

Results

This method was developed and internally validated in an open-source dataset containing 10,030 echocardiograms. The Pearson's correlation coefficient was 0.83 for LVEF prediction compared to expert human analysis (p < 0.001), with a subsequent area under the receiver operator curve (AUROC) of 0.98 (95% confidence interval 0.97 to 0.99) for categorisation of HF with reduced ejection (HFrEF; LVEF<40%). In an external dataset with 200 echocardiograms, this method achieved an AUC of 0.90 (95% confidence interval 0.88 to 0.91) for HFrEF assessment.

Conclusion

The automated neural network-based calculation of LVEF is comparable to expert clinicians performing time-consuming, frame-by-frame manual evaluations of cardiac systolic function.

Deep learning to assess right ventricular ejection fraction from two-dimensional echocardiograms in precapillary pulmonary hypertension

BACKGROUND

Precapillary pulmonary hypertension (PH) is characterized by a sustained increase in right ventricular (RV) afterload, impairing systolic function. Two-dimensional (2D) echocardiography is the most performed cardiac imaging tool to assess RV systolic function; however, an accurate evaluation requires expertise. We aimed to develop a fully automated deep learning (DL)-based tool to estimate the RV ejection fraction (RVEF) from 2D echocardiographic videos of apical four-chamber views in patients with precapillary PH.

METHODS

We identified 85 patients with suspected precapillary PH who underwent cardiac magnetic resonance imaging (MRI) and echocardiography. The data was divided into training (80%) and testing (20%) datasets, and a regression model was constructed using 3D-ResNet50. Accuracy was assessed using five-fold cross validation.

RESULTS

The DL model predicted the cardiac MRI-derived RVEF with a mean absolute error of 7.67%. The DL model identified severe RV systolic dysfunction (defined as cardiac MRI-derived RVEF < 37%) with an area under the curve (AUC) of .84, which was comparable to the AUC of RV fractional area change (FAC) and tricuspid annular plane systolic excursion (TAPSE) measured by experienced sonographers (.87 and .72, respectively). To detect mild RV systolic dysfunction (defined as RVEF ≤ 45%), the AUC from the DL-predicted RVEF also demonstrated a high discriminatory power of .87, comparable to that of FAC (.90), and significantly higher than that of TAPSE (.67).

CONCLUSION

The fully automated DL-based tool using 2D echocardiography could accurately estimate RVEF and exhibited a diagnostic performance for RV systolic dysfunction comparable to that of human readers.

A deep multi-stream model for robust prediction of left ventricular ejection fraction in 2D echocardiography

We propose a deep multi-stream model for left ventricular ejection fraction (LVEF) prediction in 2D echocardiographic (2DE) examinations. We use four standard 2DE views as model input, which are automatically selected from the full 2DE examination. The LVEF prediction model processes eight streams of data (images + optical flow) and consists of convolutional neural networks terminated with transformer layers. The model is made robust to missing, misclassified and duplicate views via pre-training, sampling strategies and parameter sharing. The model is trained and evaluated on an existing clinical dataset (12,648 unique examinations) with varying properties in terms of quality, examining physician, and ultrasound system. We report [Formula: see text] and mean absolute error = 4.0% points for the test set. When evaluated on two public benchmarks, the model performs on par or better than all previous attempts on fully automatic LVEF prediction. Code and trained models are available on a public project repository .

The Role of Artificial Intelligence in Echocardiography: A Clinical Update

PURPOSE OF REVIEW

In echocardiography, there has been robust development of artificial intelligence (AI) tools for image recognition, automated measurements, image segmentation, and patient prognostication that has created a monumental shift in the study of AI and machine learning models. However, integrating these measurements into complex disease recognition and therapeutic interventions remains challenging. While the tools have been developed, there is a lack of evidence regarding implementing heterogeneous systems for guiding clinical decision-making and therapeutic action.

RECENT FINDINGS

Newer AI modalities have shown concrete positive data in terms of user-guided image acquisition and processing, precise determination of both basic and advanced quantitative echocardiographic features, and the potential to construct predictive models, all with the possibility of seamless integration into clinical decision support systems. AI in echocardiography is a powerful and ever-growing tool with the potential for revolutionary effects on the practice of cardiology. In this review article, we explore the growth of AI and its applications in echocardiography, along with clinical implications and the associated regulatory, legal, and ethical considerations.

Artificial Intelligence-Assisted Left Ventricular Diastolic Function Assessment and Grading: Multiview Versus Single View

BACKGROUND

Clinical assessment and grading of left ventricular diastolic function (LVDF) requires quantification of multiple echocardiographic parameters interpreted according to established guidelines, which depends on experienced clinicians and is time consuming. The aim of this study was to develop an artificial intelligence (AI)-assisted system to facilitate the clinical assessment of LVDF.

METHODS

In total, 1,304 studies (33,404 images) were used to develop a view classification model to select six specific views required for LVDF assessment. A total of 2,238 studies (16,794 two-dimensional [2D] images and 2,198 Doppler images) to develop 2D and Doppler segmentation models, respectively, to quantify key metrics of diastolic function. We used 2,150 studies with definite LVDF labels determined by two experts to train single-view classification models by AI interpretation of strain metrics or video. The accuracy and efficiency of these models were tested in an external data set of 388 prospective studies.

RESULTS

The view classification model identified views required for LVDF assessment with good sensitivity (>0.9), and view segmentation models successfully outlined key regions of these views with intersection over union > 0.8 in the internal validation data set. In the external test data set of 388 cases, AI quantification of 2D and Doppler images showed narrow limits of agreement compared with the two experts (e.g., left ventricular ejection fraction, -12.02% to 9.17%; E/e' ratio, -3.04 to 2.67). These metrics were used to detect LV diastolic dysfunction (DD) and grade DD with accuracy of 0.9 and 0.92, respectively. Concerning the single-view method, the overall accuracy of DD detection was 0.83 and 0.75 by strain-based and video-based models, and the accuracy of DD grading was 0.85 and 0.8, respectively. These models could achieve diagnosis and grading of LVDD in a few seconds, greatly saving time and labor.

CONCLUSION

AI models successfully achieved LVDF assessment and grading that compared favorably with human experts reading according to guideline-based algorithms. Moreover, when Doppler variables were missing, AI models could provide assessment by interpreting 2D strain metrics or videos from a single view. These models have the potential to save labor and cost and to facilitate work flow of clinical LVDF assessment.

Artificial intelligence-assisted interpretation of systolic function by echocardiogram

OBJECTIVE

Precise and reliable echocardiographic assessment of left ventricular ejection fraction (LVEF) is needed for clinical decision-making. Recently, artificial intelligence (AI) models have been developed to estimate LVEF accurately. The aim of this study was to evaluate whether an AI model could estimate an expert read of LVEF and reduce the interinstitutional variability of level 1 readers with the AI-LVEF displayed on the echocardiographic screen.

METHODS

This prospective, multicentre echocardiographic study was conducted by five cardiologists of level 1 echocardiographic skill (minimum level of competency to interpret images) from different hospitals. Protocol 1: Visual LVEFs for the 48 cases were measured without input from the AI-LVEF. Protocol 2: the 48 cases were again shown to all readers with inclusion of AI-LVEF data. To assess the concordance and accuracy with or without AI-LVEF, each visual LVEF measurement was compared with an average of the estimates by five expert readers as a reference.

RESULTS

A good correlation was found between AI-LVEF and reference LVEF (r=0.90, p<0.001) from the expert readers. For the classification LVEF, the area under the curve was 0.95 on heart failure with preserved EF and 0.96 on heart failure reduced EF. For the precision, the SD was reduced from 6.1±2.3 to 2.5±0.9 (p<0.001) with AI-LVEF. For the accuracy, the root-mean squared error was improved from 7.5±3.1 to 5.6±3.2 (p=0.004) with AI-LVEF.

CONCLUSIONS

AI can assist with the interpretation of systolic function on an echocardiogram for level 1 readers from different institutions.

Automatic echocardiographic evaluation of the probability of pulmonary hypertension using machine learning

Echocardiography, a simple and noninvasive tool, is the first choice for screening pulmonary hypertension (PH). However, accurate assessment of PH, incorporating both the pulmonary artery pressures and additional signs for PH remained unsatisfied. Thus, this study aimed to develop a machine learning (ML) model that can automatically evaluate the probability of PH. This cohort included data from 346 (275 for training set and internal validation set and 71 for external validation set) patients with suspected PH patients and receiving right heart catheterization. Echocardiographic images on parasternal short axis-papillary muscle level (PSAX-PML) view from all patients were collected, labeled, and preprocessed. Local features from each image were extracted and subsequently integrated to build a ML model. By adjusting the parameters of the model, the model with the best prediction effect is finally constructed. We used receiver-operating characteristic analysis to evaluate model performance and compared the ML model with the traditional methods. The accuracy of the ML model for diagnosis of PH was significantly higher than the traditional method (0.945 vs. 0.892, p = 0.027 [area under the curve [AUC]]). Similar findings were observed in subgroup analysis and validated in the external validation set (AUC = 0.950 [95% CI: 0.897-1.000]). In summary, ML methods could automatically extract features from traditional PSAX-PML view and automatically assess the probability of PH, which were found to outperform traditional echocardiographic assessments.

Single-Site Experience with an Automated Artificial Intelligence Application for Left Ventricular Ejection Fraction Measurement in Echocardiography

Left ventricular ejection fraction (LVEF) is a key parameter in evaluating left ventricular (LV) function using echocardiography (Echo), but its manual measurement by the modified biplane Simpson (MBS) method is time consuming and operator dependent. We investigated the feasibility of a server-based, commercially available and ready-to use-artificial intelligence (AI) application based on convolutional neural network methods that integrate fully automatic view selection and measurement of LVEF from an entire Echo exam into a single workflow. We prospectively enrolled 1083 consecutive patients who had been referred to Echo for diagnostic or therapeutic purposes. LVEF was measured independently using MBS and AI. Test–retest variability was assessed in 40 patients. The reliability, repeatability, and time efficiency of LVEF measurements were compared between the two methods. Overall, 889 Echos were analyzed by cardiologists with the MBS method and by the AI. Over the study period of 10 weeks, the feasibility of both automatic view classification and seamlessly measured LVEF rose to 81% without user involvement. LVEF, LV end-diastolic and end-systolic volumes correlated strongly between MBS and AI (R = 0.87, 0.89 and 0.93, p < 0.001 for all) with a mean bias of +4.5% EF, −12 mL and −11 mL, respectively, due to impaired image quality and the extent of LV function. Repeatability and reliability of LVEF measurement (n = 40, test–retest) by AI was excellent compared to MBS (coefficient of variation: 3.2% vs. 5.9%), although the median analysis time of the AI was longer than that of the operator-dependent MBS method (258 s vs. 171 s). This AI has succeeded in identifying apical LV views and measuring EF in one workflow with comparable results to the MBS method and shows excellent reproducibility. It offers realistic perspectives for fully automated AI-based measurement of LVEF in routine clinical settings.

Revealing Unforeseen Diagnostic Image Features With Deep Learning by Detecting Cardiovascular Diseases From Apical 4-Chamber Ultrasounds

Background

With the increase of highly portable, wireless, and low‐cost ultrasound devices and automatic ultrasound acquisition techniques, an automated interpretation method requiring only a limited set of views as input could make preliminary cardiovascular disease diagnoses more accessible. In this study, we developed a deep learning method for automated detection of impaired left ventricular (LV) function and aortic valve (AV) regurgitation from apical 4‐chamber ultrasound cineloops and investigated which anatomical structures or temporal frames provided the most relevant information for the deep learning model to enable disease classification.

Methods and Results

Apical 4‐chamber ultrasounds were extracted from 3554 echocardiograms of patients with impaired LV function (n=928), AV regurgitation (n=738), or no significant abnormalities (n=1888). Two convolutional neural networks were trained separately to classify the respective disease cases against normal cases. The overall classification accuracy of the impaired LV function detection model was 86%, and that of the AV regurgitation detection model was 83%. Feature importance analyses demonstrated that the LV myocardium and mitral valve were important for detecting impaired LV function, whereas the tip of the mitral valve anterior leaflet, during opening, was considered important for detecting AV regurgitation.

Conclusions

The proposed method demonstrated the feasibility of a 3‐dimensional convolutional neural network approach in detection of impaired LV function and AV regurgitation using apical 4‐chamber ultrasound cineloops. The current study shows that deep learning methods can exploit large training data to detect diseases in a different way than conventionally agreed on methods, and potentially reveal unforeseen diagnostic image features.

Automatic Detection of Secundum Atrial Septal Defect in Children Based on Color Doppler Echocardiographic Images Using Convolutional Neural Networks

Secundum atrial septal defect (ASD) is one of the most common congenital heart diseases (CHDs). This study aims to evaluate the feasibility and accuracy of automatic detection of ASD in children based on color Doppler echocardiographic images using convolutional neural networks. In this study, we propose a fully automatic detection system for ASD, which includes three stages. The first stage is used to identify four target echocardiographic views (that is, the subcostal view focusing on the atrium septum, the apical four-chamber view, the low parasternal four-chamber view, and the parasternal short-axis view). These four echocardiographic views are most useful for the diagnosis of ASD clinically. The second stage aims to segment the target cardiac structure and detect candidates for ASD. The third stage is to infer the final detection by utilizing the segmentation and detection results of the second stage. The proposed ASD detection system was developed and validated using a training set of 4,031 cases containing 370,057 echocardiographic images and an independent test set of 229 cases containing 203,619 images, of which 105 cases with ASD and 124 cases with intact atrial septum. Experimental results showed that the proposed ASD detection system achieved accuracy, recall, precision, specificity, and F1 score of 0.8833, 0.8545, 0.8577, 0.9136, and 0.8546, respectively on the image-level averages of the four most clinically useful echocardiographic views. The proposed system can automatically and accurately identify ASD, laying a good foundation for the subsequent artificial intelligence diagnosis of CHDs.

Echocardiographic Advances in Dilated Cardiomyopathy

Although the overall survival of patients with dilated cardiomyopathy (DCM) has improved significantly in the last decades, a non-negligible proportion of DCM patients still shows an unfavorable prognosis. DCM patients not only need imaging techniques that are effective in diagnosis, but also suitable for long-term follow-up with frequent re-evaluations. The exponential growth of echocardiography’s technology and performance in recent years has resulted in improved diagnostic accuracy, stratification, management and follow-up of patients with DCM. This review summarizes some new developments in echocardiography and their promising applications in DCM. Although nowadays cardiac magnetic resonance (CMR) remains the gold standard technique in DCM, the echocardiographic advances and novelties proposed in the manuscript, if properly integrated into clinical practice, could bring echocardiography closer to CMR in terms of accuracy and may certify ultrasound as the technique of choice in the follow-up of DCM patients. The application in DCM patients of novel echocardiographic techniques represents an interesting emergent research area for scholars in the near future.

Deep Learning Analysis of Echocardiographic Images to Predict Positive Genotype in Patients With Hypertrophic Cardiomyopathy

Genetic testing provides valuable insights into family screening strategies, diagnosis, and prognosis in patients with hypertrophic cardiomyopathy (HCM). On the other hand, genetic testing carries socio-economical and psychological burdens. It is therefore important to identify patients with HCM who are more likely to have positive genotype. However, conventional prediction models based on clinical and echocardiographic parameters offer only modest accuracy and are subject to intra- and inter-observer variability. We therefore hypothesized that deep convolutional neural network (DCNN, a type of deep learning) analysis of echocardiographic images improves the predictive accuracy of positive genotype in patients with HCM. In each case, we obtained parasternal short- and long-axis as well as apical 2-, 3-, 4-, and 5-chamber views. We employed DCNN algorithm to predict positive genotype based on the input echocardiographic images. We performed 5-fold cross-validations. We used 2 reference models—the Mayo HCM Genotype Predictor score (Mayo score) and the Toronto HCM Genotype score (Toronto score). We compared the area under the receiver-operating-characteristic curve (AUC) between a combined model using the reference model plus DCNN-derived probability and the reference model. We calculated the p-value by performing 1,000 bootstrapping. We calculated sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). In addition, we examined the net reclassification improvement. We included 99 adults with HCM who underwent genetic testing. Overall, 45 patients (45%) had positive genotype. The new model combining Mayo score and DCNN-derived probability significantly outperformed Mayo score (AUC 0.86 [95% CI 0.79–0.93] vs. 0.72 [0.61–0.82]; p < 0.001). Similarly, the new model combining Toronto score and DCNN-derived probability exhibited a higher AUC compared to Toronto score alone (AUC 0.84 [0.76–0.92] vs. 0.75 [0.65–0.85]; p = 0.03). An improvement in the sensitivity, specificity, PPV, and NPV was also achieved, along with significant net reclassification improvement. In conclusion, compared to the conventional models, our new model combining the conventional and DCNN-derived models demonstrated superior accuracy to predict positive genotype in patients with HCM.

Artificial intelligence in echocardiography: detection, functional evaluation, and disease diagnosis

Ultrasound is one of the most important examinations for clinical diagnosis of cardiovascular diseases. The speed of image movements driven by the frequency of the beating heart is faster than that of other organs. This particularity of echocardiography poses a challenge for sonographers to diagnose accurately. However, artificial intelligence for detection, functional evaluation, and disease diagnosis has gradually become an alternative for accurate diagnosis and treatment using echocardiography. This work discusses the current application of artificial intelligence in echocardiography technology, its limitations, and future development directions.

How to standardize the measurement of left ventricular ejection fraction

Despite recent advances in imaging for myocardial deformation, left ventricular ejection fraction (LVEF) is still the most important index for systolic function in daily practice. Its role in multiple fields (e.g., valvular heart disease, myocardial infarction, cancer therapy-related cardiac dysfunction) has been a mainstay in guidelines. In addition, assessment of LVEF is vital to clinical decision-making in patients with heart failure. However, notable limitations to LVEF include poor inter-observer reproducibility dependent on observer skill, poor acoustic windows, and variations in measurement techniques. To solve these problems, methods for standardization of LVEF by sharing reference images among observers and artificial intelligence for accurate measurements have been developed. In this review, we focus on the standardization of LVEF using reference images and automated LVEF using artificial intelligence.

Current status and issues regarding reference values for echocardiography: a short review

In general, the reference values of diagnostic test parameters are specified based on the values of 95% confidence intervals of those parameters measured in healthy subjects. As heart size varies according to sex, there are sex-related differences in the reference values of echocardiographic parameters. There have been attempts to minimize the variability in the reference values of echocardiographic parameters worldwide by correcting for age-related, sex-related, and body size-related differences. This short review describes the current status and issues regarding the reference values of echocardiographic parameters and discusses the findings of research aimed at resolving these issues.

Deep learning-based automated left ventricular ejection fraction assessment using 2-D echocardiography

Deep learning (DL) has been applied for automatic left ventricle (LV) ejection fraction (EF) measurement, but the diagnostic performance was rarely evaluated for various phenotypes of heart disease. This study aims to evaluate a new DL algorithm for automated LVEF measurement using two-dimensional echocardiography (2DE) images collected from three centers. The impact of three ultrasound machines and three phenotypes of heart diseases on the automatic LVEF measurement was evaluated. Using 36890 frames of 2DE from 340 patients, we developed a DL algorithm based on U-Net (DPS-Net) and the biplane Simpson's method was applied for LVEF calculation. Results showed a high performance in LV segmentation and LVEF measurement across phenotypes and echo systems by using DPS-Net. Good performance was obtained for LV segmentation when DPS-Net was tested on the CAMUS data set (Dice coefficient of 0.932 and 0.928 for ED and ES). Better performance of LV segmentation in study-wise evaluation was observed by comparing the DPS-Net v2 to the EchoNet-dynamic algorithm (P = 0.008). DPS-Net was associated with high correlations and good agreements for the LVEF measurement. High diagnostic performance was obtained that the area under receiver operator characteristic curve was 0.974, 0.948, 0.968, and 0.972 for normal hearts and disease phenotypes including atrial fibrillation, hypertrophic cardiomyopathy, dilated cardiomyopathy, respectively. High performance was obtained by using DPS-Net in LV detection and LVEF measurement for heart failure with several phenotypes. High performance was observed in a large-scale dataset, suggesting that the DPS-Net was highly adaptive across different echocardiographic systems.NEW & NOTEWORTHY A new strategy of feature extraction and fusion could enhance the accuracy of automatic LVEF assessment based on multiview 2-D echocardiographic sequences. High diagnostic performance for the determination of heart failure was obtained by using DPS-Net in cases with different phenotypes of heart diseases. High performance for left ventricle segmentation was obtained by using DPS-Net, suggesting the potential for a wider range of application in the interpretation of 2DE images.

A deep learning approach for the automatic recognition of prosthetic mitral valve in echocardiographic images

The first step in the automatic evaluation of the cardiac prosthetic valve is the recognition of such valves in echocardiographic images. This research surveyed whether a deep convolutional neural network (DCNN) could improve the recognition of prosthetic mitral valve in conventional 2D echocardiographic images. An efficient intervention to decrease the misreading rate of the prosthetic mitral valve is required for non-expert cardiologists. This intervention could serve as a section of a fully-automated analysis chain, alleviate the cardiologist's workload, and improve precision and time management, especially in an emergent situation. Additionally, it might be suitable for pre-labeling large databases of unclassified images. We, therefore, introduce a large publicly-available annotated dataset for the purpose of prosthetic mitral valve recognition. We utilized 2044 comprehensive non-stress transthoracic echocardiographic studies. Totally, 1597 patients had natural mitral valves and 447 patients had prosthetic valves. Each case contained 1 cycle of echocardiographic images from the apical 4-chamber (A4C) and the parasternal long-axis (PLA) views. Thirteen versions of the state-of-the-art models were independently trained, and the ensemble predictions were performed using those versions. For the recognition of prosthetic mitral valves from natural mitral valves, the area under the receiver-operating characteristic curve (AUC) made by the deep learning algorithm was similar to that made by cardiologists (0.99). In this research, EfficientNetB3 architecture in the A4C view and the EfficientNetB4 architecture in the PLA view were the best models among the other pre-trained DCNN models.

The Role of Artificial Intelligence in Cardiovascular Imaging: State of the Art Review

In this current digital landscape, artificial intelligence (AI) has established itself as a powerful tool in the commercial industry and is an evolving technology in healthcare. Cutting-edge imaging modalities outputting multi-dimensional data are becoming increasingly complex. In this era of data explosion, the field of cardiovascular imaging is undergoing a paradigm shift toward machine learning (ML) driven platforms. These diverse algorithms can seamlessly analyze information and automate a range of tasks. In this review article, we explore the role of ML in the field of cardiovascular imaging.

Steps to use artificial intelligence in echocardiography

Artificial intelligence (AI) has influenced every field of cardiovascular imaging in all phases from acquisition to reporting. Compared with computed tomography and magnetic resonance imaging, there is an issue of high observer variation in the interpretation of echocardiograms. Therefore, AI can help minimize the observer variation and provide accurate diagnosis in the field of echocardiography. In this review, we summarize the necessity for automated diagnosis in the echocardiographic field, and discuss the results of AI application to echocardiography and future perspectives. Currently, there are two roles for AI in cardiovascular imaging. One is the automation of tasks performed by humans, such as image segmentation, measurement of cardiac structural and functional parameters. The other is the discovery of clinically important insights. Most reported applications were focused on the automation of tasks. Moreover, algorithms that can obtain cardiac measurements are also being reported. In the next stage, AI can be expected to expand and enrich existing knowledge. With the continual evolution of technology, cardiologists should become well versed in this new knowledge of AI and be able to harness it as a tool. AI can be incorporated into everyday clinical practice and become a valuable aid for many healthcare professionals dealing with cardiovascular diseases.

Radiomics in Echocardiography: Deep Learning and Echocardiographic Analysis

PURPOSE OF REVIEW

Recent development in artificial intelligence (AI) for cardiovascular imaging analysis, involving deep learning, is the start of a new phase in the research field. We review the current state of AI in cardiovascular field and discuss about its potential to improve clinical workflows and accuracy of diagnosis.

RECENT FINDINGS

In the AI cardiovascular imaging field, there are many applications involving efficient image reconstruction, patient triage, and support for clinical decisions. These tools have a role to support repetitive clinical tasks. Although they will be powerful in some situations, these applications may have new potential in the hands of echo cardiologists, assisting but not replacing the human observer. We believe AI has the potential to improve the quality of echocardiography. Someday AI may be incorporated into the daily clinical setting, being an instrumental tool for cardiologists dealing with cardiovascular diseases.

Clinically Feasible and Accurate View Classification of Echocardiographic Images Using Deep Learning

A proper echocardiographic study requires several video clips recorded from different acquisition angles for observation of the complex cardiac anatomy. However, these video clips are not necessarily labeled in a database. Identification of the acquired view becomes the first step of analyzing an echocardiogram. Currently, there is no consensus whether the mislabeled samples can be used to create a feasible clinical prediction model of ejection fraction (EF). The aim of this study was to test two types of input methods for the classification of images, and to test the accuracy of the prediction model for EF in a learning database containing mislabeled images that were not checked by observers. We enrolled 340 patients with five standard views (long axis, short axis, 3-chamber view, 4-chamber view and 2-chamber view) and 10 images in a cycle, used for training a convolutional neural network to classify views (total 17,000 labeled images). All DICOM images were rigidly registered and rescaled into a reference image to fit the size of echocardiographic images. We employed 5-fold cross validation to examine model performance. We tested models trained by two types of data, averaged images and 10 selected images. Our best model (from 10 selected images) classified video views with 98.1% overall test accuracy in the independent cohort. In our view classification model, 1.9% of the images were mislabeled. To determine if this 98.1% accuracy was acceptable for creating the clinical prediction model using echocardiographic data, we tested the prediction model for EF using learning data with a 1.9% error rate. The accuracy of the prediction model for EF was warranted, even with training data containing 1.9% mislabeled images. The CNN algorithm can classify images into five standard views in a clinical setting. Our results suggest that this approach may provide a clinically feasible accuracy level of view classification for the analysis of echocardiographic data.