Automated assessment of global heart shape in stress echocardiography using convolutional neural networks

• Grant number: 23/07883-7
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date until: November 01, 2023
• End date until: October 31, 2024
• Field of knowledge: Engineering - Biomedical Engineering - Medical Engineering
• Principal Investigator: Rangel Arthur
• Grantee: Gabriel Alves Baltazar
• Host Institution: Faculdade de Tecnologia (FT). Universidade Estadual de Campinas (UNICAMP). Limeira , SP, Brazil

Abstract

Stress echocardiography plays a fundamental role in assessing cardiac function and detecting heart diseases, providing valuable information about heart performance during physiological stress situations. However, the analysis of these exams still faces significant challenges, especially in evaluating the global shape of the heart before analyzing the details.Traditional approaches in echocardiographic analysis often focus on analyzing specific details of cardiac structures, such as wall thickness or segmental movement. While these detailed analyses are important, it is equally relevant to evaluate the global shape of the heart for a comprehensive understanding of its function and possible pathological changes.However, evaluating the global shape of the heart has been challenging due to its subjective nature and dependence on the expertise of the echocardiography specialist. Additionally, manual analysis is time-consuming and can lead to inter-observer variations.Given these limitations, there is a need for an objective and automated approach to assess the global shape of the heart before analyzing the details in stress echocardiographic images. In this context, convolutional neural networks (CNNs) have proven to be a powerful tool in medical image analysis, capable of learning complex patterns and extracting relevant information.The aim of this study is to propose the use of CNNs, specifically the MASK-RCNN and U-NET models, to evaluate the overall shape before analyzing the details in stress echocardiography. These models have shown innovative potential in various image segmentation tasks, but so far, no studies exploring their specific application for this task in stress echocardiography have been found.By using these models, we seek to obtain an accurate segmentation of the global shape of the heart, allowing for automated and reliable analysis. This innovative approach has the potential to overcome the limitations of traditional techniques and provide a more objective evaluation of cardiac shape during stress echocardiography.