Fetal cardiac ultrasound standard section detection model based on multitask learning and mixed attention mechanism

Abstract

Fetal cardiac ultrasound is a valuable tool for screening fetal cardiac health during pregnancy. Ultrasound standard section testing is an essential part of fetal heart ultrasound diagnosis. Due to the many scattered and partially similar anatomical structures in standard ultrasound views of the fetal heart, the results strongly depend on the clinical experience and knowledge of the sonographer. To improve detection efficiency and reduce misdiagnosis and omission, we propose a single-stage fetal cardiac ultrasound standard plane detection model (FCUM) based on multitask learning and a hybrid attention mechanism to assist sonographers in diagnosis. The feature fusion pyramids of the backbone and detection networks of this model are each embedded with a hybrid attention mechanism module of our design, which enables the multitasking network to extract shared features more accurately and efficiently and improves the accuracy of the detection and classification networks. We designed a classification module for multilayer residual network feature fusion that leads to better classification and faster convergence time. We conducted comprehensive experiments on a dataset of fetal cardiac ultrasound images acquired with several types of devices and in different geographic regions. The experimental results show that our model outperforms baseline models such as YOLOv8 and ResNet-50 in terms of detection precision and classification accuracy.