An optimized modified faster region convolutional neural network for spina bifida identification from ultrasound images

Abstract

Detecting spina bifida defects in an early parental stage and providing proper remedial measures is the main objective. There it demands effective learning approaches to automatically detect and classify fetal disability. However, the existing approaches face several complicated issues like time complexity, cost, and misclassification.This paper proposes a novel Modified Faster Region Convolutional neural Network based Cheetah Optimizer (Modified FRCNN-CO) model to accurately predict whether the child is defective with spina bifida disease or not.The ultrasound scanning images of the fetus acquired during the second trimester are considered an efficient screening tool for detecting fetus abnormalities. The ultrasound scan image of the fetal spine obtained during the 18thweek of pregnancy is taken as input. The poor quality and noise factors present in ultrasound images are enhanced and removed respectively using preprocessing pipelines namely contrast enhancement, intensity adjustment, and denoising. The enhanced images are segmented through the generative adversarial network (GAN) model. With the capability to capture data distribution, the GAN model segments and emphasizes defective regions of images. This procedure makes the proposed modified FRCNN-CO model accurately identify the images with spina bifida disease. Based on the features extracted, the proposed modified FRCNN-CO model accurately classifies and labels the ultrasound images into two classes normal ‘0′ and defective ‘1′.The experimental outcomes illustrate the supremacy of the proposed modified FRCNN-CO model over other compared methods, especially achieved detection accuracy of about 97.8%.