OP10.08: Automatic detection of biparietal diameter from ultrasound images using deep learning

Abstract

In recent years, the evolution of artificial intelligence (AI) has made it possible to analyse an image which was previously difficult to evaluate. Deep learning is currently used in studies as a computer aided diagnosis (CAD) system in the CT and MRI fields and has gradually spread in the ultrasound field. The aim of this study is to correctly recognise biparietal diameter (BPD) images via fetal ultrasound using the Convolutional Neural Network (CNN) and to develop a system to detect more accurate BPDs. The data were collected from fetal measurements at 18 to 40 weeks gestation via abdominal ultrasound (Voluson E10) from January 2017 to January 2019. Two gynecologists classified 1000 images into 2 groups as teaching data and non-teaching data. The images were anonymised, edited, and output to JPG format. Among all of the data collected, 75% was divided into training data, and the other 25% was classified as test data; the convolutional neural network (CNN) was then constructed (figure 1). After 100 epochs of learning, the accuracy of the learning model was evaluated by cross-validation. This study was conducted with the approval of the ethics committee of our hospital. As a result, the detection system of the BPD images via fetal ultrasound using CNN had an accuracy of 90.0% and a test loss 0.27. Supporting information can be found in the online version of this abstract Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Ähnliche Arbeiten