Segmentation of Heart Wall Muscles and Detection of Hypertrophic Cardiomyopathy from 2D Echo Images Using U-Nets

EasyChair Preprint 10465

Abstract

In this paper, we focus on the development of an automatic technique to obtain the segmentation of the heart wall in ultrasound images using U-Nets. The detection of hypertrophic cardiomyopathy (HCM) and similar conditions is achieved by measuring the thickness of the posterior heart wall in the left ventricle. We measure the thickness of the segmented heart wall using repeated morphological operations in the form of erosion to give us an idea of its real thickness. Medical literature suggests that if the thickness of the heart wall is greater than 15mm, then we classify the image as a potential case of HCM. In our experiment, we have taken 139 images in our training, and 34 images in our test set in order to examine the accuracy of our technique. We find that the U-Net obtains an accuracy of 0.85 in terms of the Dice similarity coefficient.

Keyphrases: HCM, LVH, ROI, Segmentation, U-Net

@booklet{EasyChair:10465,
  author    = {A Aiswarya and Hisham Ahamed and Nagesh Subbanna},
  title     = {Segmentation of Heart Wall Muscles and Detection of Hypertrophic Cardiomyopathy from 2D Echo Images Using U-Nets},
  howpublished = {EasyChair Preprint 10465},
  year      = {EasyChair, 2023}}