New Study Reveals AI's Potential in Detecting Heart Defects Before Birth

The study was led by a multidisciplinary team from the Nuffield Department of Women’s & Reproductive Health and the Institute of Biomedical Engineering. It forms part of an international, multi-centre collaboration for the development of clinical AI models in Fetal Echocardiography (CAIFE) for the detection of CHD.  

CHDs represent one of the most common birth defects worldwide, affecting approximately 1% of newborns. Early identification is crucial for enabling appropriate care pathways and enhancing postnatal outcomes. However, despite the central role of routine prenatal ultrasound, many heart anomalies are still missed in clinical practice, particularly in low-resource settings or where fetal cardiology expertise is unavailable. 

The study analysed data from over 30,000 fetuses and evaluated various AI models used in fetal ultrasounds, which are performed between 16 and 40 weeks of gestation. The results showed that some AI models could match or even outperform less experienced human doctors in identifying whether the fetal heart was normal or abnormal. However, these results were based on controlled research environments and may not reflect how AI would perform in everyday clinical settings.

While these findings are promising, further research is needed to determine how AI would perform in various populations and healthcare systems, particularly given the diversity of heart defects and the differences in ultrasound equipment and techniques employed. The researchers also pointed out the importance of ensuring that AI tools are fair, transparent, and trustworthy when used in real-life situations.

This research is part of the CAIFE project, which aims to tackle significant global health issues through innovative technology and collaborative efforts.

"Our analysis highlights both the promise and the complexity of using AI in prenatal cardiac screening. AI models can perform remarkably well, sometimes approaching expert-level interpretation. Yet translating these models into everyday clinical practice remains a considerable challenge." -Dr Elena D'Alberti

"Our study found that AI in fetal echocardiography has great promise to facilitate and reduce disparities in the prenatal diagnosis of CHD. This initiative is particularly critical in non-specialist settings and low- and middle-income countries and holds the potential to reduce global neonatal morbidity and mortality. However, such application is yet to be proven, requiring further research and development." - Dr Olga Patey

"Early AI model designs and translational studies, as systematically reviewed in this article, are demonstrating the potential of AI-assistive technology in fetal echocardiography. Challenges of turning that potential into trusted and reliable solutions embedded into clinical workflows and at scale will require inter-disciplinary research teams with expertise in clinical medicine, computer science/engineering, social sciences to work together, and partnerships with industry. However, the technical foundations are now established, and we are entering an incredibly exciting chapter in healthcare AI imaging where we will see AI-assistive technology realise its potential to address real-world clinical needs.”  - Prof. Alison Noble