Utilization of Artificial Intelligence in Echocardiography

Abstract

Echocardiography has a central role in the diagnosis and management of cardiovascular disease. Precise and reliable echocardiographic assessment is required for clinical decision-making. Even if the development of new technologies (3-dimentional echocardiography, speckle-tracking, semi-automated analysis, etc.), the final decision on analysis is strongly dependent on operator experience. Diagnostic errors are a major unresolved problem. Moreover, not only can cardiologists differ from one another in image interpretation, but also the same observer may come to different findings when a reading is repeated. Daily high workloads in clinical practice may lead to this error, and all cardiologists require precise perception in this field. Artificial intelligence (AI) has the potential to improve analysis and interpretation of medical images to a new stage compared with previous algorithms. From our comprehensive review, we believe AI has the potential to improve accuracy of diagnosis, clinical management, and patient care. Although there are several concerns about the required large dataset and “black box” algorithm, AI can provide satisfactory results in this field. In the future, it will be necessary for cardiologists to adapt their daily practice to incorporate AI in this new stage of echocardiography.

In the modern era, artificial intelligence (AI) is spreading into all parts of daily life. AI is a program that has tasks based on algorithms in an intelligent manner. Machine learning is a subset of AI and focuses on the machine’s ability to receive a set of data and learn for itself. The tasks in machine learning can be classified into supervised and unsupervised learning problems. In the former, the task of assigning data to one of the discrete categories is called classification, whereas the task of fitting the desired output consisting of ≥1 continuous variables is called regression. In the latter, the goal may be to discover some groups categorized with similar variables, or features, called “clustering”. Deep learning is a subset of machine learning that can solve a problem by using multilayered neural networks (Figure 1). Deep learning has led to state-of-the art improvements in word recognition, visual object recognition, object detection, etc.¹ Image recognition by machines trained through deep learning in some situations is superior to that of humans. Just a few years ago, we were surprised by a machine learning-based computer program (“AlphaGo”) that defeated the world champion of Go.² The AI algorithm continues to be enhanced every year. Medical imaging also seems to be changing and undergoing an important revolution because of AI methods such as deep learning based on neural networks.

Figure 1.

Artificial intelligence, including machine learning and deep learning and their tasks.

Data are an essential component of AI, and the quality and size of a dataset used to build a model will strongly influence the outcomes. When datasets are biased, the results are unusable in the clinical setting. Preprocessing is also important because echocardiographic data are nonstructural and there are differences in image properties in the dataset. Cardiologists need to make a labeled dataset to develop the model. After development of models, a clinical trial is key to this flow because developed models must be validated in different cohorts. A typical model development process is shown in Figure 2. AI researchers follow this process when developing a new model.

Figure 2.

Development process for artificial intelligence models.

Recently, there have been some reports on AI in the medical imaging modalities. For example, calcium scoring in low-dose chest computed tomography (CT), identification of functionally significant stenosis in CT angiography, and diagnosis of chronic myocardial infarction on cine magnetic resonance image (MRI) have been developed.^3–5 Compared with CT and MRI, in echocardiography there is an issue of high observer variation in the interpretation of images. Thus, AI might be help to improve observer variation and provide accurate diagnosis in echocardiography. In this review, we focus on the current status and future directions of AI in the field of echocardiography.

Direction of AI in Echocardiography

Echocardiography has a central role in the diagnosis and management of cardiovascular disease.⁶ Precise and reliable echocardiographic assessment is required for clinical decision-making.^7–10 Even in the development of new technologies (3-dimentional echocardiography, speckle-tracking, semi-automated analysis, etc.), the final analytical decision is strongly dependent on operator experience. For example, left ventricular ejection fraction (LVEF) is subjective, and variability could be influenced by observer experience. Several institutes have several readers with a wide range of experience levels.^11,12 Until now, many interventions for reduction of variability in LVEF have been tested to overcome this issue.^13,14 Our multicenter group suggested that a simple teaching intervention can reduce the variability in LVEF assessment, especially for readers with limited experience.¹⁵ However, there are several limitations, including a lack of ground truth, limited number of sample sizes, etc. Thus, diagnostic errors are a major unresolved problem. Moreover, not only can cardiologists differ from one another in image interpretations, but the same observer may come to different conclusion when a reading is repeated. Daily high workloads in clinical practice may lead to this error, and all cardiologists require precise perception in this field.

AI will likely help to overcome these issues. The AI algorithms might provide an aid to diagnostics with fewer errors and provide hidden features for accurate diagnosis. One landmark echocardiographic paper was recently published.¹⁶ The authors used a deep learning model to build a fully automated echocardiogram interpretation program, including view identification, image segmentation, quantification of structure and function, and disease detection. Since then, many cardiologists can see a potential role for AI in the echocardiographic field. Our laboratory also investigated the building of models of automated diagnosis of myocardial infarction using a deep learning algorithm.¹⁷ The model showed several new insights and findings in the development of the algorithm. AI has the potential to improve analysis and interpretation of medical images to an advanced stage compared with previous algorithms. Table summarizes the diagnostic ability of current machine-learning models in the field of echocardiography.^16,18–25 The remainder of this review focuses on previously published deep learning approaches in echocardiography, view classifications, automated analysis of size and function, diagnosis of cardiovascular diseases, and diastolic dysfunction.

Table. Studies of Machine Learning for Echocardiography

Authors	Year	Target	Models	Training/ validation dataset	Test dataset	Accuracy	AUC
Madani et al24	2018	Echocardiography views (Classification)	Neural network	200,000 images	20,000 images	0.92	1.00
Zhang et al16	2018	Echocardiography views (Classification)	Neural network	Total 14,035 studies	–	0.84	–
Raghavendra et al25	2018	Wall motion abnormalities (Classification)	Neural network	279 images	–	0.75	–
Omar et al18	2018	Wall motion abnormalities (Classification)	Neural network	4,392 maps	61 subjects	0.95	–
Kusunose et al17	2019	Wall motion abnormalities (Classification)	Neural network (5 types)	960 images	240 images+120 images from an independent cohort	–	0.97
Zhang et al16	2018	LV size and function (Regression*)	Neural network	Total 14,035 studies	–	Median absolute deviations of 15–17%	–
Sanchez-Martinez et al19	2018	Heart failure with preserved EF (Clustering)	Agglomerative hierarchical clustering	–	–	0.73	–
Tabassian et al20	2018	Heart failure with preserved EF (Clustering and classification)	KNN and PCA	–	–	0.81	–
Narula et al21	2016	Myocardial disease (HCM vs. athlete) (Classification)	Support vector machine	–	–	–	0.80
Sengupta et al22	2016	Myocardial disease (CP vs. RCM) (Classification)	Associative memory classifier	–	–	0.94	0.96
Zhang et al16	2018	Myocardial disease (HCM, amyloidosis, PAH) (Classification)	Neural network	Total 14,035 studies	–	–	0.85–0.93

*Left ventricle was segmented by a classification model and then the size and function were evaluated. AUC, area under the curve; CP, constrictive pericarditis; HCM, hypertrophic cardiomyopathy; KNN, k-nearest neighbor; PAH, pulmonary hypertension; PCA, principle component analysis; RCM, restrictive cardiomyopathy.

AI for View Classifications

Echocardiographic images consist of several video clips, still images (M-mode and B-mode) and Doppler recordings because the heart’s structure and function are complex and require many views to diagnose cardiovascular diseases (Figure 3). Because of the nonstructural data in echocardiography, determination of the view is the essential first step in interpreting an echocardiogram. Recent studies applied deep learning with convolutional neural networks for view classification of echocardiograms.²⁴ They trained a convolutional neural network to simultaneously classify 15 standard views (12 video, 3 still), based on labeled still images and videos from 267 transthoracic echocardiograms with over 800,000 images that captured a range of real-world clinical variations. Their model classified among 12 video views with 97.8% overall test accuracy without overfitting. Another group reported that a model for view classification was successfully trained with more layers and a larger number of echocardiography view classes.¹⁶ Thus, this method may be reasonable for application to image classification. On the other hand, there are some limitations, including lack of explanation of the learning process and less than perfect classification. The utility is questionable in the current version of models. Moreover, they did not adequately resolve the problem of vendor differences and image qualities. A future enhanced model to classify the correct views would be required.

Figure 3.

Variety of echocardiographic images needed to be recognized by artificial intelligence systems.

AI for Size and Function

Quantification of heart size and function is an essential part of echocardiography. Fully automated 3D echocardiographic analysis can obtain quantitative results without any observer interaction (e.g., selection of views, positioning markers and modifying borders). Commercially available software has been tested for accuracy and reproducibility. The algorithms are knowledge-based probabilistic contouring algorithms²⁶ or adaptive analytics algorithms.²⁷ The most frequently used software is the HeartModel algorithm in the Philips EPIQ series (Figure 4). This software shows automated tracings of the left ventricular and left atrial endocardial borders with 3D analysis. There are many studies comparing fully automated methods and either cardiac magnetic resonance or manual echocardiography,^28–31 but there are some limitations from the clinical setting viewpoint. One is the dependency on image quality, which has an important role, because results obtained with poor but analyzable image quality provide inaccurate results.³² On the other hand, although measurement accuracy using this analysis still depends on image quality, its degree becomes obviously smaller than with current semi-automated software. The number of datasets for training also affects measurement accuracy. The current adaptive analytics algorithm does not work very well in patients with distorted LV shape, such as LV aneurysms and apical hypertrophic cardiomyopathy because of the limited number of datasets for machine learning.

Figure 4.

Representative fully automated analysis software, HeartModel.

These limitations also exist for deep learning in echocardiography. Deep learning algorithms require a high-quality database to provide a sound estimation model with a small sample size. Zhang et al proposed a pipeline based on a deep learning approach for a fully automated analysis of echocardiographic data.¹⁶ They proposed to train a U-net deep learning model for automatic segmentation of the LV in the apical 4- and 2-chamber views. For LV segmentation, their model had a Dice score of approximately 85% in the apical 2-chamber view, and approximately 80% in the apical 4-chamber view, and a mean absolute percentage error of approximately 10% for EF from the apical 2-chamber view, and 20% for EF from the apical 4-chamber view. The correlation is good; however, their dataset did not include a wide range of LVEF. In addition, EF values were evaluated after segmentation of the LV, so may include segmentation errors. We believe that a direct evaluation of EF with larger training sets that include extremes of LV range will be needed for training and validation.

AI for Wall Motion Abnormality

One of the most important assessments in echocardiography is evaluating regional wall motion abnormalities (RWMAs) for the management of ischemic coronary artery disease (CAD). Assessment of RWMAs is a Class I recommendation in the guidelines by trained echocardiographic technicians for patients with chest pain in the emergency department.^33–35 Conventional assessment of RWMAs, which is based on visual interpretation of endocardial excursion and myocardial thickening, is subjective and experience-dependent.³⁶ A useful method for reducing the misreading of RWMAs is required.^37–39 Machine-learning models have been evaluated to identify and quantify RWMAs.^18,25 A convolutional neural network provided good models with high sensitivity for diagnosis of CAD. Recently, our laboratory investigated building models of automated diagnosis for myocardial infarction using a deep learning algorithm (Figure 5).¹⁷ For detection of the presence of RWMA, the area under the receiver-operating characteristic curve (AUC) by deep learning algorithm was similar to that for a reading by cardiologist/sonographer, and significantly higher than the AUC for resident readers. Interestingly, deep learning had relatively low ratios of misclassification of the right coronary artery, left circumflex coronary artery, and control groups except for the left anterior descending coronary artery (LAD). It seems to reflect the real-world assessment (e.g., overdiagnoses in ischemic groups by human observers or importance of LAD in the clinical setting). The results of a deep learning model in echocardiography might provide new insights in the medical field.

Figure 5.

An example of a convolutional neural network model for detection of coronary artery disease. LAD, left anterior descending coronary artery; LCX, left circumflex coronary artery; RCA, right coronary artery.

AI for Diagnosis of Cardiovascular Diseases

Several techniques have been applied to identify clinical disease states. In the echocardiographic field, speckle-tracking imaging is widely used in cases of cardiomyopathy. Clinical reports on speckle-tracking imaging show significant differences in regional strain in several cardiomyopathies, even in the absence of ischemia. Knowledge of the characteristic LV strain distribution pattern might facilitate diagnosis of constrictive pericarditis, cardiac amyloidosis, hypertrophic cardiomyopathy, hypertensive heart disease, tachycardia-induced cardiomyopathy, and aortic stenosis (Figure 6).^40–46 On the other hand, recent American and European consensus paper describe reginal longitudinal strain assessment by speckle-tracking analysis as still too immature to adopt in the clinical setting.⁴⁷ The limitation should be overcome in the future. Recently, machine-learning algorithms have revealed clinical disease conditions and new futures. Sengupta et al applied a cognitive machine-learning algorithm to differentiate constrictive pericarditis from restrictive cardiomyopathy with multimodality imaging and pathology.²² The same group showed a machine-learning approach to assessing the potential role diagnosing hypertrophy in athletes and hypertrophic cardiomyopathy.²¹ Sanchez-Martinez et al¹⁹ and Tabassian et al²⁰ showed that machine learning using echocardiographic data, including strain imaging at rest and during exercise, may improve diagnosis and understanding of heart failure with preserved EF. In this field, investigators try to not only assess the accuracy of diagnosis, but also discover new findings in cardiovascular disease. Zhang et al proposed a model based on a deep learning approach for differentiating cardiomyopathy and pulmonary hypertension from the parasternal long-axis views.¹⁶ Unlike other machine-learning approaches, the deep learning approach may automatically encode optimal features from data beyond human recognition. Big data have the potential to lead to precise diagnosis and discovering important features from the echocardiographic images. In the future, AI may aid physicians in accurate diagnosis without requiring pathological samples.

Figure 6.

Examples of strain distribution in amyloidosis, constrictive pericarditis and hypertrophic cardiomyopathy.

Future of AI in Echocardiography

Cardiologists will determine the capability of AI in diagnosis, and they will be responsible for the final decisions. Thus, cardiologists will be required to have the capacity e to manage AI and advanced knowledge. Some recent studies have been concerned about adversarial examples in the medical imaging field.⁴⁸ Adversarial examples are inputs to learning models that an attacker has intentionally designed to cause the model to make a mistake; they are like optical illusions for machines. In echocardiography, data are just pixel images, not structured data. Echocardiographic imaging systems may be vulnerable to adversarial attacks. For example, insurance companies will use a deep learning system that receives images as part of a claim to verify that heart surgery would be necessary in the future. An adversarial example may be used to deceive the insurance company’s system. In these cases, cardiologists should have adequate and solid knowledge in this field. The era of AI is almost here.

Conclusions

From our comprehensive review, we believe AI has the potential to improve accuracy of diagnosis, clinical management, and patient care. Although there are several concerns about the required large dataset and “black box” algorithm, AI seems able to provide satisfactory results in this field. In the future, it will be necessary for cardiologists to incorporate this new horizon of AI in echocardiography into their daily practice.

Sources of Funding

This work was partially supported by JSPS Kakenhi Grants (Number 17K09506 to K. Kusunose, and 19H03654 to M. Sata), the Takeda Science Foundation (to M. Sata), and the Vehicle Racing Commemorative Foundation (to M. Sata).

Conflict of Interest

The authors declare no conflicts of interest.

Acknowledgment

The authors acknowledge Kathryn Brock, BA, for revising the manuscript.

References

• 1.   LeCun Y, Bengio Y, Hinton G. Deep learning. Nature 2015; 521: 436.
• 2.   Mozur P. Google’s AlphaGo defeats Chinese Go Master in win for AI. The New York Times 2017; B3.
• 3.   Lessmann N, van Ginneken B, Zreik M, de Jong PA, de Vos BD, Viergever MA, et al. Automatic calcium scoring in low-dose chest CT using deep neural networks with dilated convolutions. IEEE Trans Med Imaging 2018; 37: 615–625.
• 4.   van Hamersvelt RW, Zreik M, Voskuil M, Viergever MA, Isgum I, Leiner T. Deep learning analysis of left ventricular myocardium in CT angiographic intermediate-degree coronary stenosis improves the diagnostic accuracy for identification of functionally significant stenosis. Eur Radiol 2019; 29: 2350–2359.
• 5.   Zhang N, Yang G, Gao Z, Xu C, Zhang Y, Shi R, et al. Deep learning for diagnosis of chronic myocardial infarction on nonenhanced cardiac cine MRI. Radiology 2019; 291: 606–617.
• 6.   Mitchell C, Rahko PS, Blauwet LA, Canaday B, Finstuen JA, Foster MC, et al. Guidelines for performing a comprehensive transthoracic echocardiographic examination in adults: Recommendations from the American Society of Echocardiography. J Am Soc Echocardiogr 2019; 32: 1–64.
• 7.   Rady M, Ulbrich S, Heidrich F, Jellinghaus S, Ibrahim K, Linke A, et al. Left ventricular torsion: A new echocardiographic prognosticator in patients with non-ischemic dilated cardiomyopathy. Circ J 2019; 83: 595–603.
• 8.   Kawasaki M, Tanaka R, Yoshida A, Nagaya M, Minatoguchi S, Yoshizane T, et al. Non-invasive pulmonary capillary wedge pressure assessment on speckle tracking echocardiography as a predictor of new-onset non-valvular atrial fibrillation: Four-year prospective study (NIPAF Study). Circ J 2018; 82: 3029–3036.
• 9.   Hatazawa K, Tanaka H, Nonaka A, Takada H, Soga F, Hatani Y, et al. Baseline global longitudinal strain as a predictor of left ventricular dysfunction and hospitalization for heart failure of patients with malignant lymphoma after anthracycline therapy. Circ J 2018; 82: 2566–2574.
• 10.   Wang LW, Kesteven SH, Huttner IG, Feneley MP, Fatkin D. High-frequency echocardiography: Transformative clinical and research applications in humans, mice, and zebrafish. Circ J 2018; 82: 620–628.
• 11.   Kaufmann BA, Min SY, Goetschalckx K, Bernheim AM, Buser PT, Pfisterer ME, et al. How reliable are left ventricular ejection fraction cut offs assessed by echocardiography for clinical decision making in patients with heart failure? Int J Cardiovasc Imaging 2013; 29: 581–588.
• 12.   Okuma H, Noto N, Tanikawa S, Kanezawa K, Hirai M, Shimozawa K, et al. Impact of persistent left ventricular regional wall motion abnormalities in childhood cancer survivors after anthracycline therapy: Assessment of global left ventricular myocardial performance by 3D speckle-tracking echocardiography. J Cardiol 2017; 70: 396–401.
• 13.   Gottdiener JS, Bednarz J, Devereux R, Gardin J, Klein A, Manning WJ, et al. American Society of Echocardiography recommendations for use of echocardiography in clinical trials. J Am Soc Echocardiogr 2004; 17: 1086–1119.
• 14.   Bhattacharyya S, Lloyd G. Improving appropriateness and quality in cardiovascular imaging. Circ Cardiovasc Imaging 2015; 8: e003988.
• 15.   Kusunose K, Shibayama K, Iwano H, Izumo M, Kagiyama N, Kurosawa K, et al. Reduced variability of visual left ventricular ejection fraction assessment with reference images: The Japanese Association of Young Echocardiography Fellows multicenter study. J Cardiol 2018; 72: 74–80.
• 16.   Zhang J, Gajjala S, Agrawal P, Tison GH, Hallock LA, Beussink-Nelson L, et al. Fully automated echocardiogram interpretation in clinical practice. Circulation 2018; 138: 1623–1635.
• 17.   Kusunose K, Abe T, Haga A, Fukuda D, Yamada H, Harada M, et al. A deep learning approach for assessment of regional wall motion abnormality from echocardiographic images. JACC Cardiovasc Imaging, doi:10.1016/j.jcmg.2019.02.024.
• 18.   Omar HA, Domingos JS, Patra A, Upton R, Leeson P, Noble JA. Quantification of cardiac bull’s-eye map based on principal strain analysis for myocardial wall motion assessment in stress echocardiography. In: Proceedings of IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018). IEEE, 2018; 1195–1198.
• 19.   Sanchez-Martinez S, Duchateau N, Erdei T, Kunszt G, Aakhus S, Degiovanni A, et al. Machine learning analysis of left ventricular function to characterize heart failure with preserved ejection fraction. Circ Cardiovasc Imaging 2018; 11: e007138.
• 20.   Tabassian M, Sunderji I, Erdei T, Sanchez-Martinez S, Degiovanni A, Marino P, et al. Diagnosis of heart failure with preserved ejection fraction: Machine learning of spatiotemporal variations in left ventricular deformation. J Am Soc Echocardiogr 2018; 31: 1272–1284.e1279.
• 21.   Narula S, Shameer K, Salem Omar AM, Dudley JT, Sengupta PP. Machine-learning algorithms to automate morphological and functional assessments in 2D echocardiography. J Am Coll Cardiol 2016; 68: 2287–2295.
• 22.   Sengupta PP, Huang YM, Bansal M, Ashrafi A, Fisher M, Shameer K, et al. Cognitive machine-learning algorithm for cardiac imaging: A pilot study for differentiating constrictive pericarditis from restrictive cardiomyopathy. Circ Cardiovasc Imaging 2016; 9: pii: e004330.
• 23.   Higaki A, Inoue K, Kinoshita M, Ikeda S, Yamaguchi O. Reconstruction of apical 2-chamber view from apical 4-and long-axis views on echocardiogram using machine learning: Pilot study with deep generative modeling. Circ Rep 2019; 1: 197.
• 24.   Madani A, Arnaout R, Mofrad M, Arnaout R. Fast and accurate view classification of echocardiograms using deep learning. NPJ Digital Med 2018; 1: 6.
• 25.   Raghavendra U, Fujita H, Gudigar A, Shetty R, Nayak K, Pai U, et al. Automated technique for coronary artery disease characterization and classification using DD-DTDWT in ultrasound images. Biomed Signal Process Control 2018; 40: 324–334.
• 26.   Yang L, Georgescu B, Zheng Y, Foran DJ, Comaniciu D. A fast and accurate tracking algorithm of left ventricles in 3D echocardiography. Proc IEEE Int Symp Biomed Imaging 2008; 5: 221–224.
• 27.   Tsang W, Salgo IS, Medvedofsky D, Takeuchi M, Prater D, Weinert L, et al. Transthoracic 3D echocardiographic left heart chamber quantification using an automated adaptive analytics algorithm. JACC Cardiovasc Imaging 2016; 9: 769–782.
• 28.   Otani K, Nakazono A, Salgo IS, Lang RM, Takeuchi M. Three-dimensional echocardiographic assessment of left heart chamber size and function with fully automated quantification software in patients with atrial fibrillation. J Am Soc Echocardiogr 2016; 29: 955–965.
• 29.   Spitzer E, Ren B, Soliman OI, Zijlstra F, Van Mieghem NM, Geleijnse ML. Accuracy of an automated transthoracic echocardiographic tool for 3D assessment of left heart chamber volumes. Echocardiography 2017; 34: 199–209.
• 30.   Carvajal-Rivera JJ, Lopez-Quintero JC, Gonzalez-Menchen C, de Agustin JA, Macaya C, Perez de Isla L. Left ventricular volumes and ejection fraction quantification using an automated three-dimensional adaptive analytic echocardiographic algorithm in pediatric population. Echocardiography 2018; 35: 1827–1834.
• 31.   Medvedofsky D, Mor-Avi V, Amzulescu M, Fernandez-Golfin C, Hinojar R, Monaghan MJ, et al. Three-dimensional echocardiographic quantification of the left-heart chambers using an automated adaptive analytics algorithm: Multicentre validation study. Eur Heart J Cardiovasc Imaging 2018; 19: 47–58.
• 32.   Medvedofsky D, Mor-Avi V, Byku I, Singh A, Weinert L, Yamat M, et al. Three-dimensional echocardiographic automated quantification of left heart chamber volumes using an adaptive analytics algorithm: Feasibility and impact of image quality in nonselected patients. J Am Soc Echocardiogr 2017; 30: 879–885.
• 33.   Amsterdam EA, Wenger NK, Brindis RG, Casey DE Jr, Ganiats TG, Holmes DR Jr, et al. 2014 AHA/ACC Guideline for the Management of Patients with Non-ST-Elevation Acute Coronary Syndromes: A report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. J Am Coll Cardiol 2014; 64: e139–e228.
• 34.   Roffi M, Patrono C, Collet JP, Mueller C, Valgimigli M, Andreotti F, et al. 2015 ESC Guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. Eur Heart J 2016; 37: 267–315.
• 35.   Kimura K, Kimura T, Ishihara M, Nakagawa Y, Nakao K, Miyauchi K, et al. JCS 2018 guideline on diagnosis and treatment of acute coronary syndrome. Circ J 2019; 83: 1085–1196.
• 36.   Parisi AF, Moynihan PF, Folland ED, Feldman CL. Quantitative detection of regional left ventricular contraction abnormalities by two-dimensional echocardiography. II: Accuracy in coronary artery disease. Circulation 1981; 63: 761–767.
• 37.   Amundsen BH, Helle-Valle T, Edvardsen T, Torp H, Crosby J, Lyseggen E, et al. Noninvasive myocardial strain measurement by speckle tracking echocardiography: Validation against sonomicrometry and tagged magnetic resonance imaging. J Am Coll Cardiol 2006; 47: 789–793.
• 38.   Kusunose K, Yamada H, Nishio S, Mizuguchi Y, Choraku M, Maeda Y, et al. Validation of longitudinal peak systolic strain by speckle tracking echocardiography with visual assessment and myocardial perfusion SPECT in patients with regional asynergy. Circ J 2011; 75: 141–147.
• 39.   Qazi M, Fung G, Krishnan S, Rosales R, Steck H, Rao RB, et al. Automated heart wall motion abnormality detection from ultrasound images using Bayesian networks. In: Proceedings of 20th International Joint Conference on Artifical Intelligence, Hyderabad, India, January 6–12, 2007; 519–525.
• 40.   Phelan D, Thavendiranathan P, Popovic Z, Collier P, Griffin B, Thomas JD, et al. Application of a parametric display of two-dimensional speckle-tracking longitudinal strain to improve the etiologic diagnosis of mild to moderate left ventricular hypertrophy. J Am Soc Echocardiogr 2014; 27: 888–895.
• 41.   Kusunose K, Dahiya A, Popovic ZB, Motoki H, Alraies MC, Zurick AO, et al. Biventricular mechanics in constrictive pericarditis comparison with restrictive cardiomyopathy and impact of pericardiectomy. Circ Cardiovasc Imaging 2013; 6: 399–406.
• 42.   Phelan D, Collier P, Thavendiranathan P, Popovic ZB, Hanna M, Plana JC, et al. Relative apical sparing of longitudinal strain using two-dimensional speckle-tracking echocardiography is both sensitive and specific for the diagnosis of cardiac amyloidosis. Heart 2012; 98: 1442–1448.
• 43.   Foell D, Jung B, Germann E, Staehle F, Bode C, Markl M. Hypertensive heart disease: MR tissue phase mapping reveals altered left ventricular rotation and regional myocardial long-axis velocities. Eur Radiol 2013; 23: 339–347.
• 44.   Chang SA, Kim HK, Kim DH, Kim JC, Kim YJ, Kim HC, et al. Left ventricular twist mechanics in patients with apical hypertrophic cardiomyopathy: Assessment with 2D speckle tracking echocardiography. Heart 2010; 96: 49–55.
• 45.   Kusunose K, Torii Y, Yamada H, Nishio S, Hirata Y, Seno H, et al. Clinical utility of longitudinal strain to predict functional recovery in patients with tachyarrhythmia and reduced LVEF. JACC Cardiovasc Imaging 2017; 10: 118–126.
• 46.   Carstensen HG, Larsen LH, Hassager C, Kofoed KF, Jensen JS, Mogelvang R. Basal longitudinal strain predicts future aortic valve replacement in asymptomatic patients with aortic stenosis. Eur Heart J Cardiovasc Imaging 2016; 17: 283–292.
• 47.   Mirea O, Pagourelias ED, Duchenne J, Bogaert J, Thomas JD, Badano LP, et al. Variability and reproducibility of segmental longitudinal strain measurement: A report from the EACVI-ASE Strain Standardization Task Force. JACC Cardiovasc Imaging 2018; 11: 15–24.
• 48.   Szegedy C, Zaremba W, Sutskever I, Bruna J, Erhan D, Goodfellow I, et al. Intriguing properties of neural networks (submitted 21 December 2013, revised 19 February 2014 (v4)). arXiv.org>cs>arXiv,1312.6199. Cornell University. https://arxiv.org/abs/1312.6199 (accessed June 2019).