Neural architecture search of echocardiography view classifiers

Abstract

Purpose: Echocardiography is the most commonly used modality for assessing the heart in clinical practice. In an echocardiographic exam, an ultrasound probe samples the heart from different orientations and positions, thereby creating different viewpoints for assessing the cardiac function. The determination of the probe viewpoint forms an essential step in automatic echocardiographic image analysis.Approach: In this study, convolutional neural networks are used for the automated identification of 14 different anatomical echocardiographic views (larger than any previous study) in a dataset of 8,732 videos acquired from 374 patients. Differentiable architecture search approach was utilised to design small neural network architectures for rapid inference while maintaining high accuracy. The impact of the image quality and resolution, size of the training dataset, and number of echocardiographic view classes on the efficacy of the models were also investigated.Results: In contrast to the deeper classification architectures, the proposed models had significantly lower number of trainable parameters (up to 99.9% reduction), achieved comparable classification performance (accuracy 88.4-96.0%, precision 87.8-95.2%, recall 87.1-95.1%) and real-time performance with inference time per image of 3.6-12.6ms.Conclusion: Compared with the standard classification neural network architectures, the proposed models are faster and achieve comparable classification performance. They also require less training data. Such models can be used for real-time detection of the standard views.

Accuracy of Deep Learning Echocardiographic View Classification in Patients with Congenital or Structural Heart Disease: Importance of Specific Datasets

“…Automated view classification is an important part of completely autonomous echocardiography interpretation by deep neural networks [13]. Zhang et al [12] as well as other working groups [13,14] and our present study document the high accuracy in the identification of echocardiographic views by a DNN trained and validated in a cohort of general cardiology patients when applied to these patients or patients without a cardiac abnormality. However, the present study indicates that this DNN's precision is considerably decreased in a patient population with underlying congenital or structural heart disease.…”

Automated assessment of transthoracic echocardiogram image quality using deep neural networks

“…A neural network model, previously developed in our research group [19], was then used to detect different echocardiographic views and separate the A4C and PLAX views. This resulted in a total of 33,784 frames from different patients: 15,476 and 18,308 frames for A4C and PLAX, respectively.…”

The emerging roles of machine learning in cardiovascular diseases: a narrative review

“…The recognition accuracy of a single frame and 2 images was 98.3% and the time spent on recognition and classification was satisfied (15). Besides, Azarmehr et al developed a new neural architecture search algorithm to improve image classification speed and performance (16). Overall, ML can assist doctors to collect, recognize, classify and distinguish echocardiogram data.…”