Why the study?
Left ventricular ejection fraction is a key indicator in heart failure, and new methods are needed to localize the left ventricle boundary to potentially mitigate the drawbacks of segmentation-based methods.
Does a CNN regression transfer learning model accurately predict left ventricular ejection fraction categories from 2D echocardiography images?
Does a CNN regression transfer learning model accurately predict left ventricular ejection fraction categories from 2D echocardiography images?
A novel deep learning approach using CNN regression models for 2D echocardiography achieved 77% accuracy in categorizing left ventricular ejection fraction, demonstrating potential for automated cardiac assessment.
May support automated EF categorization in echocardiography research; leaves open clinical adoption pending prospective validation.
A key indicator in the diagnosis, prognosis, and management of individuals with heart failure (HF) is the left ventricular ejection fraction (EF). The amount of blood that is forced out of the left ventricle with each contraction provides important details on how well the heart can circulate oxygen-rich blood across the human body. Echocardiography has long been the most commonly used imaging method for determining LVEF due to its availability and cost-effectiveness. This paper makes use of the EchoNet-Dynamic dataset, which has left ventricle coordination data. An organized data preprocessing pipeline is created to extract frames along with coordinates. The suggested model architecture incorporates pre-trained transfer learning models that are optimal for the task of localizing the left ventricle boundaries. By predicting coordinates with CNN regression-type models, we showed how a novel volume tracing method could be used to localize the left ventricle boundary and perhaps mitigate the drawbacks of segmentation-based methods. Based on predetermined thresholds, we divided the ejection fraction (EF) values into "normal," "mild," and "abnormal" categories to detect the patient's heart condition. The analysis revealed a high degree of sensitivity for the "normal" and "abnormal" classes but was lower in the "mild" class. We obtained a confusion matrix accuracy of 77%.
No takes yet. Share an insight, caveat, or question.
Paul et al. (2024) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: