# Automatic Segmentation of Left Ventricle in Cardiac Magnetic Resonance Images

**Authors:** Garvit Chhabra, J. H. Gagan, J. R. Harish Kumar  
**arXiv:** [2201.12805](https://arxiv.org/abs/2201.12805)  
**Date:** 2022-01-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2201.12805v1/

## Key points

- Importance of segmenting the left ventricle in cardiac MRI scans for calculating volume and determining ejection fraction
- Ejection fraction as a crucial metric indicating the percentage of blood leaving the heart with each contraction
- Proposal of a novel approach combining multiscale template matching and elliptical active disc for automated segmentation
- Utilization of gradient descent optimization and Green's theorem to optimize a local energy function based on five free parameters defining the disc shape
- Validation on a dataset comprising 320 scans with 5,273 annotated slices from the M&Ms Challenge
- Successful localization of the left ventricle in 89.63% of cases with Dice coefficients of 0.873 on diastole slices and 0.770 on systole slices
- Comparable performance to deep learning methods despite being rooted in traditional image processing techniques

## Summary

In their paper titled "Automatic Segmentation of Left Ventricle in Cardiac Magnetic Resonance Images," authors Garvit Chhabra, J. H. Gagan, and J. R. Harish Kumar discuss the importance of segmenting the left ventricle in cardiac MRI scans for calculating its volume and determining ejection fraction. The ejection fraction is a crucial metric that indicates the percentage of blood leaving the heart with each contraction and serves as a key indicator of cardiac function. To address this need, the authors propose a novel approach that combines multiscale template matching for detection and an elliptical active disc for automated segmentation of the left ventricle in MR images. The proposed method utilizes gradient descent optimization and Green's theorem to streamline computation expenses while optimizing a local energy function based on five free parameters defining the disc shape. This technique demonstrates promising results and was validated on a dataset comprising 320 scans with 5,273 annotated slices from the Multi-Centre, Multi-Vendor, and Multi-Disease Cardiac Segmentation (M&Ms) Challenge. The results showed successful localization of the left ventricle in 89.63% of cases with Dice coefficients of 0.873 on diastole slices and 0.770 on systole slices. Notably, despite being rooted in traditional image processing techniques, their approach shows comparable performance to deep learning methods. Overall, this study presents an innovative solution for automating left ventricle segmentation in cardiac MRI images , offering potential benefits for cardiologists in accurately assessing cardiac health and function efficiently.