Edge-Sensitive Left Ventricle Segmentation Using Deep Reinforcement Learning.

Jingjing Xiong,Pengfei Xian,Yuzhi Zhao,Yasar Abbas Ur Rehman,Yujia Zhang,Lai-Man Po,Kwok Wai Cheung

doi:10.3390/s21072375

Abstract

Deep reinforcement learning (DRL) has been utilized in numerous computer vision tasks, such as object detection, autonomous driving, etc. However, relatively few DRL methods have been proposed in the area of image segmentation, particularly in left ventricle segmentation. Reinforcement learning-based methods in earlier works often rely on learning proper thresholds to perform segmentation, and the segmentation results are inaccurate due to the sensitivity of the threshold. To tackle this problem, a novel DRL agent is designed to imitate the human process to perform LV segmentation. For this purpose, we formulate the segmentation problem as a Markov decision process and innovatively optimize it through DRL. The proposed DRL agent consists of two neural networks, i.e., First-P-Net and Next-P-Net. The First-P-Net locates the initial edge point, and the Next-P-Net locates the remaining edge points successively and ultimately obtains a closed segmentation result. The experimental results show that the proposed model has outperformed the previous reinforcement learning methods and achieved comparable performances compared with deep learning baselines on two widely used LV endocardium segmentation datasets, namely Automated Cardiac Diagnosis Challenge (ACDC) 2017 dataset, and Sunnybrook 2009 dataset. Moreover, the proposed model achieves higher F-measure accuracy compared with deep learning methods when training with a very limited number of samples.

Highlights

The movement of the boundary of the left ventricle (LV) could be used to measure mechanical dyssynchrony
Numerous methods have been proposed for the automatic segmentation of LV images, which can be divided into three groups: hand-engineered features-based segmentation, deep learning-based segmentation, and reinforcement learning based segmentation
Unlike the conventional deep learning-based segmentation methods that gradually adjust the segmentation probability of each pixel, we propose a novel edge-based image segmentation model using Deep reinforcement learning (DRL), and train a DRL agent to delineate the outline of the left ventricle

Summary

Introduction

The movement of the boundary of the left ventricle (LV) could be used to measure mechanical dyssynchrony. Due to the motion of LV during image acquisition, the captured images are usually characterized by blurring and intensity inhomogeneity, which makes the segmentation of magnetic resonance LV images challenging. To address this problem, numerous methods have been proposed for the automatic segmentation of LV images, which can be divided into three groups: hand-engineered features-based segmentation, deep learning-based segmentation, and reinforcement learning based segmentation. The deep learning-based LV segmentation methods have achieved state-of-the-art performance [2,3,4,5] It extracts features from the input image in a hierarchical fashion, i.e., from low-level features to more abstract and data specific features. LV dataset is relatively small and hard to acquire, which restricts the performance of deep learning-based approaches

Objectives

Methods

Findings

Conclusion