On August 12, npj Digital Medicine, one of the top journals in medical informatics/health care science and services, published the research article A multicenter study on two-stage transfer learning model for duct-dependent CHDs screening in fetal echocardiography.

Jiajie Tang, a 2020-grade doctoral student of our School, is the first author, Yuxuan Jiang, a 2023-grade doctoral student of our School, is the third author, Kanghui Zhang, a 2021-grade doctoral student of our School, and Fanfan Zhu, a 2022-grade doctoral student of our School, participated in the study. Long Lu, Professor at our School, is co-correspondent of this article.

This study focuses on the intelligent screening of duct-dependent congenital heart diseases(CHDs) using transfer learning method. Duct-dependent CHDs are a serious form of CHD with a low detection rate, especially in underdeveloped countries and areas. Although existing studies have developed models for fetal heart structure identification, there is a lack of comprehensive evaluation of the long axis of the aorta. In this study, a total of 6698 images and 48 videos are collected to develop and test a two-stage deep transfer learning model named DDCHD-DenseNet for screening critical duct-dependent CHDs. The model achieves a sensitivity of 0.973, 0.843, 0.769, and 0.759, and a specificity of 0.985, 0.967, 0.956, and 0.759, respectively, on the four multicenter test sets. The results of the heatmaps analysis show that the model can accurately focus on the malformation regions of the heart on fetal ultrasound images, and the results of human and AI competition show that DDCHD-DenseNet demonstrated superior performance approaching that of senior sonographers in distinguishing normal from duct-dependent CHD via the screening aortic arch view. It is expected that DDCHD-DenseNet to be employed as a potential automatic screening tool for hierarchical care and computer-aided diagnosis. This study promotes the application of artificial intelligence method in prenatal diagnosis and broadens the research scope of data-driven artificial intelligence method in medical information.

Furthermore, the study also preliminarily discussed the utilization of medical resources in clinical practice. In view of the fact that some medical knowledge is wasted by excluding medical data that do not meet the experimental conditions in the existing research, a transfer learning method is proposed, which is inspired by human learning transfer mode, and knowledge transfer is carried out by setting up deep learning modes from easy to difficult and from difficult to easy. Experiments show that this method can make use of medical knowledge in non-specific data sets that are usually abandoned in the process of data-driven development, providing a new research direction for the management and utilization of medical data in clinical practice.

npj Digital Medicine is a top journal in the first-rank of medical science, Chinese Academy of Sciences, with an impact factor of 15.2 at present. In Web of Science, it is classified as Q1, ranking first among 105 journals in health care science and service in the world and second among 31 journals in medical informatics in the world.

Links to the full article: https://doi.org/10.1038/s41746-023-00883-y