Foetal Activity Detection Using Deep Convolution Neural Networks

Abstract

Ultrasound is a conventional diagnostic instrument employed in prenatal care to track the progression and advancement of the fetus. In routine clinical obstetric assessments, the standard planes of fetal ultrasound hold considerable importance in evaluating fetal growth metrics and identifying abnormalities. In this work, a method to detect FFSP using deep convolutional neural network (DCNN) architecture to improve detection efficiency is presented. Squeeze net, 16 convolutional layers with small 3x3 large kernel, and all three layers form the proposed DCNN. The final pooling layer uses global average pooling (GAP) to reduce inconsistency in the network. This helps reduce the problem of overfitting and improves the performance from different training data. To improve cognitive performance, data augmentation methods developed specifically for FFSP are used in conjunction with adaptive learning strategies. Extensive testing shows that the proposed method gives accuracy of 96% which outperforms traditional methods, and DCNN is an important tool to identify FFSP in clinical diagnosis.