Automatic segmentation based on optimization U-Net neural network (OU-NetNN) for fetal cardiac ultrasound images

The primary cause of significant birth defects is congenital heart disease (CHD), which is a significant worldwide health issue. The method remains a hard challenge because of the broad variances ascribed to several circumstances, including maternal obesity, amniotic fluid volume, and great vessel links. This work suggests deep learning-based computer-aided fetal heart echocardiogram exams with an instance segmentation strategy, which naturally segments the four standard heart images. In this study, a real-time fetal cardiac identification is performed utilizing ultrasound (US) pictures with the Modified You Only Look Once (MYOLO) method for accurate and timely detection. Firstly, the speckle noise of the US pictures is reduced using the Denoising Autoencoder Network (DAN). Secondly, localization is achieved in a fuzzy attention U-Net (FAU-Net). Next, MYOLO is optimized to function at its best, utilizing few resources. Finally, Beetle Swarm Optimization (BSO) is embedded in the U-Net Neural Network (OU-NetNN) to determine which neural network layout is most suitable for segmentation. The average fitness values in various dimensions were compared to the associated optimal values to validate the effectiveness of the recommended optimization technique. There are three assessment techniques, such as intersection over union (IoU), pixel accuracy (PA), and dice coefficient (Dice), to measure segmentation accuracy. The suggested approach works much better than current techniques in the area of segmenting fetal cardiac ultrasound pictures by ventricular septal defect (VSD), atrioventricular septal defect (AVSD), and normal, according to quantitative and visual assessments.