Performance Analysis of a Deep Convolutional Network and a Deep Belief Network for Biometric Anomalies Detection Using fetal Ultrasound Images
Year: 2024
Type: article
Abstract: The uterine pregnancy, the fetal heartbeat, and the general health and anatomy of the fetus are all assessed with ultrasound technology. Additionally, ultrasounds are used to determine the fetus’s position, heart rate, gender, gestational age, and weight. The extremely difficult tasks of finding and evaluating fetal standard scan planes during Second trimester 2D ultrasound require years of training to master. Along with the original ultrasound images, several augmentation techniques are used to enhance the input datasets. In this procedure, fetal anomalies are identified by utilizing ultrasound (US) images. The proposed approach preprocesses the image data & then applies Convolutional neural networks with deep learning (CNNs) and Deep Belief Network (DBN) to autonomously calculate fetal biometrics which includes head circumference and femur length. The proposed approach preprocesses the image data using image processing techniques and then applies deep convolutional neural networks (CNNs) and Deep Belief Network (DBN) to autonomously estimate fetal biometrics, such as head circumference and femur length. First the images are categorized into typical cases and atypical cases. Then the atypical cases are categorized as Macrocephaly, Microcephaly and achondroplasia. The input sources are trained using several CNN layer configurations and also with the help of DBN, and classification accuracy is checked during validation. The foundation of the network is built to retrieve data at various scales. Because of this architecture, the suggested approach may be expanded to whole-slide ultrasound pictures. Our solution beats the state-of-the-art significantly; the suggested study provides more accuracy for the healthcare categorization system, which will give greater precision such as 72.05, 83.95, and 95.00 using Convolutional Neural Network and DBN with prediction accuracy for three classes of 93%. We have proposed an automated diagnosis method based on biometric features and a supervised and unsupervised classification methodology.
Author: R . Chandralekha
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7217389878
