Integrating Local Binary Pattern Image Transformations and Customized Deep Learning Models for Enhanced Fetal Cardiac Anomaly Detection

Abstract

This research focuses on developing a deep learning diagnostic model for diagnosing fetal cardiac anomalies from real time ultrasound scan images. The dataset framed in the previous research is transformed using Local Binary Pattern (LBP) technique which is trained with the deep learning models to create the classifiers. These models are used to classify the Real Time images captured by ultrasound scanning machines. The LBP is a texture image feature. LBP captures the local structure and patterns within an image by comparing the intensity values of each pixel with its surrounding neighbours. The LBP operator assigns a binary code to each pixel based on the comparison results, resulting in a texture representation of the image. The FetaEcho_V05 dataset is transformed into an LBP image dataset titled as FetalEcho_V0501. This dataset is used for creating the classifiers by training the custom CNN(CCNN), AlexNet, VGG16 and ResNet50 deep learning models. The classifiers are evaluated for its overall classification performance and class wise evaluation performance using the metrics the precision, recall, accuracy and F1 score metrics. When the overall performance is considered, the CCCNN model performed the best on the FetalEcho_V0501 dataset.