Ultrasound Segmentation of Rat Hearts Using Convolution Neural Networks

Area of Science:

• Biomedical Engineering
• Medical Imaging

Background:

• Cardiovascular disease diagnosis relies heavily on ultrasound imaging.
• Manual segmentation of cardiac ultrasound images for volume estimation is time-consuming and prone to errors.
• Image artifacts like shadowing and mirror images complicate automated segmentation.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) model for automated segmentation of rat heart ultrasound images.
• To assess the impact of including diseased heart data in the training set on segmentation accuracy.

Main Methods:

• Utilized convolutional neural networks (CNNs) for image segmentation.
• Segmented rat heart ultrasound images into four classes: background, myocardium, left ventricle cavity, and right ventricle cavity.
• Compared segmentation accuracy between models trained on healthy hearts only versus models trained on both healthy and diseased hearts.

Main Results:

• Achieved an average overall segmentation accuracy of 70.0% ± 7.3% when combining healthy and diseased data.
• Models trained on healthy hearts alone yielded a slightly higher accuracy of 72.4% ± 6.6% on healthy hearts.
• Testing on diseased hearts showed variable accuracy across different segmented classes.

Conclusions:

• Including diseased hearts in training data shows potential for improving segmentation results.
• CNNs show promise for automated cardiac ultrasound segmentation, but performance varies with disease states.
• Further research with larger datasets is necessary to enhance the accuracy and robustness of CNN-based segmentation for cardiovascular applications.