<![CDATA[Advances in medical imaging and utilization have encouraged the development of more sophisticated image analysis technologies. In this context, image segmentation acts as a fundamental preprocessing step, but fetal echocardiography (FE) image segmentation still faces challenges in terms of accuracy and efficiency. The dataset for developing the FE image segmentation model was obtained from the examination results of patients at Muhammad Husein Hospital (RSMH) in Palembang who had normal conditions, atrial septal defect (ASD), ventricular septal defect (VSD), and atrioventricular septal defect (AVSD), totaling 650 FE images, which have been verified by experts. Compared to previous studies, this study focuses on creating a DL-based segmentation model for FE images using an open-source framework and the Python MIScnn library, which is specifically designed for medical imaging. This differs from previous DL frameworks that are more general, such as TensorFlow or PyTorch, which do not emphasize specialization for medical imaging. Furthermore, in an effort to improve model accuracy and efficiency, various configurations were tested, including variations in batch size and loss functions. the Model performance evaluation was conducted comprehensively using various important metrics in addition to pixel accuracy and IoU, such as F1 score, average accuracy, precision, recall, and False Positive Rate (FPR). This method is expected to provide a more in-depth picture of model performance compared to previous studies that may have only considered a few metrics. The best results were achieved using the U-Net architecture with a batch size of 32 and the binary cross-entropy loss function. This U-Net model demonstrated excellent overall performance, achieving a pixel accuracy of 0.996, an IoU of 0.995, a mean accuracy of 0.965, an FPR of 0.004, a precision of 0.929, a recall of 0.933, and an F1-score of 0.941. These findings highlight the significant potential of deep learning methods in improving the accuracy and efficiency of fetal echocardiography image analysis.]]>
Copyrights © 2025
