Automated classification of common maternal fetal ultrasound planes using multi-layer perceptron with deep feature integration

Abstract

Ultrasound is a standard diagnostic tool used during prenatal care to monitor the growth and development of the fetus. During routine clinical obstetric examinations, fetal ultrasound standard planes play a significant role in evaluating fetal growth parameters and assessing abnormalities. However, acquiring common fetal ultrasound planes with accurate fetal anatomical structures is tedious and time-consuming, even for skilled sonographers. Therefore, in this article an automated classification technique is presented for common maternal fetal ultrasound planes using deep learning models to improve detection efficiency and diagnostic accuracy. Feature integration and classification modules are the main components of the proposed approach. Initially, the deep features are extracted using AlexNet and VGG-19 with the global average pooling layer as the last pooling layer and are integrated. Fusing the deep features extracted from different convolutional neural networks strengthens the overall feature representation. After that, the integrated deep features are applied to a multi-layer perceptron to classify fetal ultrasound images into six categories. The proposed model is evaluated on a common maternal fetal ultrasound dataset, and its efficiency is computed in terms of accuracy, recall, precision, and F1-score. The proposed method achieved higher classification efficiency compared to other existing state-of-the-art models.