# A Fully Convolutional Neural Network for Cardiac Segmentation in Short-Axis MRI

**Authors:** Phi Vu Tran  
**arXiv:** [1604.00494](https://arxiv.org/abs/1604.00494)  
**Date:** 2016-04-02  
**Summary page:** https://www.summarizepaper.com/en/paper/1604.00494v3/

## Key points

- Automated cardiac segmentation from MRI datasets is crucial for timely diagnosis and management of cardiac pathologies.
- The authors propose a novel approach using a deep fully convolutional neural network architecture.
- Their model is trained end-to-end in a single learning stage, making pixel-wise inferences.
- This is the first application of such an architecture for pixel-wise labeling in cardiac MRI.
- The model outperforms previous methods across multiple evaluation measures on various datasets.
- It demonstrates robustness and accuracy in segmenting both ventricles.
- The model offers computational efficiency by leveraging GPUs for massive scale segmentation.
- Models and code are available on GitHub for further exploration and implementation.

## Summary

Automated cardiac segmentation from magnetic resonance imaging (MRI) datasets plays a crucial role in the timely diagnosis and management of cardiac pathologies. In this study, the authors propose a novel approach to automate the segmentation of the left and right ventricles using a deep fully convolutional neural network architecture. Unlike previous methods, their model is trained end-to-end in a single learning stage, taking whole-image inputs and ground truths to make pixel-wise inferences. This is the first application of such an architecture for pixel-wise labeling in cardiac MRI. The authors conducted numerical experiments to evaluate the performance of their model and found that it outperforms previous fully automated methods across multiple evaluation measures on various cardiac datasets. Notably, their model demonstrates robustness and accuracy in segmenting both ventricles. Additionally, it offers computational efficiency by leveraging commodity compute resources like graphics processing units (GPUs), enabling state-of-the-art cardiac segmentation at massive scales. The availability of the models and code on GitHub provides researchers and practitioners with valuable resources for further exploration and implementation. Overall, this study presents a significant advancement in automated cardiac segmentation using deep learning techniques, offering potential benefits for efficient diagnosis and management of cardiac pathologies.