How physiological signals and trustworthy AI can improve maternal care
Yanqi Wu defended his PhD thesis at the Department of Electrical Engineering on 12 February 2026.
Pregnancy brings profound changes to the human body. While these adaptations are essential for fetal development, they can sometimes lead to severe complications such as gestational diabetes, high blood pressure, preterm birth, and abnormal fetal growth. For his PhD research, Yanqi Wu developed new methodologies combining routine clinical data with early physiological signals to predict these pregnancy risks much earlier, paving the way for proactive maternal care.
Currently, risk assessment for pregnancy complications often relies on standard clinical factors and late-stage screening. This delayed diagnosis limits the window for early interventions, such as lifestyle or dietary changes, that could protect the health of both the mother and the baby. We wanted to shift the focus from late detection to early prediction, explains . If we can identify at-risk mothers during the first trimester, healthcare providers can intervene much sooner, which can fundamentally improve long-term health outcomes.
Uncovering early physiological markers
To achieve earlier and more accurate predictions, Wu looked beyond standard medical files. His research investigated the predictive power of physiological signals鈥攕pecifically, heart rate patterns and the synchronization between the heart and breathing.
By analyzing these signals, which were measured during overnight sleep in early pregnancy, Wu found that subtle changes in the body rhythms act as early warning signs months before clinical symptoms appear. When these physiological markers were combined with routine clinical information, the predictive models significantly outperformed standard approaches. While this research utilized clinical-grade sensors, these fundamental body rhythms could eventually be measured by simple, unobtrusive home-monitoring devices, making early risk assessment much more accessible in the future.
An AI assistant that knows its limits
A major challenge in adopting Artificial Intelligence in healthcare is ensuring it is safe and reliable. Traditional AI models often force a high risk or low risk guess even when the patient's data is unclear, which can lead to clinical mistakes. Wu tackled this challenge by designing an AI system that calculates its own uncertainty. If the system is not confident about a specific case, rather than forcing a prediction, it categorizes the patient as uncertain and routes the file to a clinician for human review. This I'm not sure mechanism can reduce incorrect predictions. Ultimately, this research demonstrates how AI can be designed as a trustworthy decision-support tool that works in tandem with doctors to enhance clinical expertise, rather than attempting to replace it.
Title of PhD thesis: . Supervisors: Dr. Xi Long and Dr. Elisabetta Peri and Dr. Myrthe van der Ven.