Artificial intelligence in recognition and segmentation of left ventricle in 2D echocardiography

Abstract

Introduction Echocardiography examination requires a high level of clinical skill and operator training to guarantee good quality image acquisition and analysis. Therefore, the interpretation remains dependent on inter-operator variabilities. Deep learning has achieved remarkable results in a variety of computer vision tasks. It’s application in echocardiography is gaining popularity, in particular in tasks related to automation. Aim To evaluate deep learning for recognition of geometrical features of left ventricle (LV) in automated cardiac measurements. Methods A total of 400 end-systolic and end-diastolic frames from 80 patients (with various indications for the study) were used to train and validate the neural networks. Raw pixel data was extracted from EPIQ 7G (Philips) imaging platform. All of the images were from 2D echocardiography apical four chamber views. LV was annotated in each image, with 367 images used for training and 33 for validation. We used the state of art mask R-CNN (regional convolutional neural network) model. Mask R-CNN is a two stage model: the first stage scans the echocardiogram and detects areas likely to contain LV and the second stage returns region class (LV), bounding boxes (squared region of LV) and masks (list of pixels that belong to LV). To reduce the training time we used a mask R-CNN pretrained on ImageNet data (a common dataset for computer vision tasks) for the initial step. Intersection over Union (IoU) was used as the primary metric for model evaluation. IoU measures the number of pixels common between the tar