Prenatal Detection of Ventricular Septal Defects by VGG -16 Model Using Ultrasound Images

Congenital heart defect is one of the important anomaly. The deformed heart can be differentiated from the normal heart based on several parameters such as the size of auricles, ventricles, valve and position of heart, area, circumference, and perimeter. One of the methods to detect the anomalies in fetal heart is by applying advanced image pprocessing techniques and artificial intelligence algorithms. This proposed system is the primary framework for diagnosing the prenatal ventricular septal defects (PVSD). The first step is to denoise the US images using enhanced anisotropic diffusion Enhanced Perona Malik Filter (EPMF), followed by K-means clustering segmentation method as the second step, and finally, VGG-16 architecture was implemented with the pre-trained weights from the database. The original image is compared with the reference image in terms of different parameters using a VGG16 deep learning algorithm to predict PVSD anomalies at the early stage of pregnancy. The experimental results shows that PSNR reach to high value of 33.45, Dice coefficient and jacquard index and conformity coefficient as 0.88-0.86. VGG-16 is the first attempt to diagnose prenatal ventricular septal defects to achieve an accuracy of 90%. This proposed system will give a second opinion for the doctors in diagnosing the abnormalities at the early stages.