Cardiac Image Segmentation Based On Deep Learning

Assessing the left ventricle in cardiac magnetic resonance imaging(MRI)through segmentation plays a crucial role in the diagnosis of cardiac diseases for cardiologists.However,conventional manual segmentation is a tedious task that requires excessive human effort,which makes automated segmentation highly desirable in practice to facilitate the process of clinical diagnosis.This paper proposes two methods for automatically outing the left ventricle,namely a left ventricle segmentation algorithm based on boundary weighted loss and residual feature aggregation(RFA)and a left ventricle contouring algorithm based on deep reinforcement learning.The first method is based on the U-Net model,where normal convolutions of the encoder and decoder are replaced with a residual feature aggregation(RFA)module for more efficient feature extraction.At the same time,we add a series of cascaded dilated convolutions in the middle part of the encoder and decoder to expand the receptive field.In addition,we design a boundary weighted loss function,which can effectively address poor segmentation results caused by blurred/incomplete edges of the target object,or high proximity between the target object and others.Through experimental verification,it is proved that the proposed model and the carefully designed loss function both contribute to segmentation performance.In the second method,we propose a novel reinforcement-learning-based framework for left ventricle contouring,which mimics how a cardiologist outlines the left ventricle along a specific trajectory in a cardiac image.Following the algorithm of proximal policy optimization(PPO),we train a policy network,which makes a stochastic decision on the agent’s movement according to its local observation such that the generated trajectory matches the true contour of the left ventricle as much as possible.Moreover,we design a deep learning model with a customized loss function to generate the agent’s landing spot(or coordinate of its initial position on a cardiac image).The experiment results show that the coordinate of the generated landing spot is sufficiently close to the true contour and the proposed reinforcement-learning-based approach outperforms the existing U-net model and its improved version,even with limited training set.The cardiac MRI dataset used in this paper was provided by the Department of Radiology,Peking Union Medical College Hospital,and we train and test the proposed model on this dataset.Compared with other algorithms,it is proved that the method proposed in this paper is able to further improve the segmentation accuracy and achieve better performance.