2D-3D Fully Convolutional Neural Networks for Cardiac MR Segmentation

TL;DR

This paper presents 2D and 3D CNN-based pipelines for fully automated cardiac MR image segmentation, achieving near state-of-the-art results and providing insights into optimal model configurations.

Contribution

It introduces a comprehensive analysis of 2D and 3D CNN architectures for cardiac MRI segmentation, including a novel dice loss function and detailed experimental insights.

Findings

Models achieve high accuracy close to state-of-the-art

Introduction of a novel dice loss function improves segmentation

Analysis of network structures enhances model performance

Abstract

In this paper, we develop a 2D and 3D segmentation pipelines for fully automated cardiac MR image segmentation using Deep Convolutional Neural Networks (CNN). Our models are trained end-to-end from scratch using the ACD Challenge 2017 dataset comprising of 100 studies, each containing Cardiac MR images in End Diastole and End Systole phase. We show that both our segmentation models achieve near state-of-the-art performance scores in terms of distance metrics and have convincing accuracy in terms of clinical parameters. A comparative analysis is provided by introducing a novel dice loss function and its combination with cross entropy loss. By exploring different network structures and comprehensive experiments, we discuss several key insights to obtain optimal model performance, which also is central to the theme of this challenge.