Stony Brook scientists have received major NIH funding to investigate a pioneering approach to fetal monitoring that could improve outcomes in the delivery room.
A multi-disciplinary team led by the College of Engineering and Applied Sciences (CEAS) has received $3.2 million under the National Institutes of Health (NIH) Research Project Grant Program (RO1) to investigate machine learning methods for classification of intrapartum signals (FHR and uterine activity) that has the potential to significantly outperform the accuracy of contemporary methods. The project is called “Rethinking Electronic Fetal Monitoring to Improve Perinatal Outcomes and Reduce Frequency of Operative Vaginal and Cesarean Deliveries.”
Since the 1970’s, the number of C-sections performed during childbirth has climbed from two percent to 36 percent. While more women are electing to have the procedure, that is not the main driver for this dramatic increase. Most C-sections are the result of doctor decisions during childbirth. According to studies in more than 20 countries that were reviewed by Trinity College in Dublin and published in Science Daily in 2018, a key driver is doctors’ fear of litigation.
During labor, a fetus can be deprived of adequate levels of oxygen. If the oxygen supply drops below a certain threshold, asphyxia occurs, which can lead to permanent brain damage or even death of the fetus/newborn. Current technology employs fetal heart rate (FHR) and uterine activity monitoring to inform decisions taken in the delivery room. The use of these patterns has not reduced the number of unwanted neurologic outcomes despite the increase in surgical interventions, in contrast to what had been widely expected.
Machine learning could help, the Stony Brook team believes.
“Dr. Gerald Quirk and I began looking at this problem five-plus years ago, receiving a small grant to do some exploratory research,” said Professor Petar Djurić, Principal Investigator (PI) and Chair of the Department of Electrical and Computer Engineering. “Through our initial work, we discovered what we believe is a way to use machine learning to ‘see’ what doctors can’t see to provide data-driven guidance for decision-making in the delivery room. This should serve to greatly reduce the number of surgical interventions while leading to better fetal outcomes. It will also lower healthcare costs as C-sections are much more expensive than normal deliveries.”
“We believe that through the collaborative efforts of our team, coming from different disciplines, we will develop systems able to more precisely identify the fetus truly at risk for an adverse neurologic outcome such as cerebral palsy or hypoxic/ischemic encephalopathy while sparing the majority of women who undergo a cesarean for an “abnormal” fetal heart rate pattern but in fact deliver a normal baby,” Quirk said.
Because the intrapartum signals (FHR and uterine activity) are not the only source of information about the fetus, the proposed methods also exploit various physiological data that are acquired on a routine basis such as age and ethnicity of the mother and whether the fetus is a first child or not. The emphasis of this research is on sequential signal processing methods that will capture the dynamics of fetus well-being on a minute-by-minute basis. It is expected that the high accuracy of the proposed methods will place them at the heart of computerized decision support systems, which in turn will pave the way for the wide adoption of these systems in the future.
“Another key outcome of this project will be to build a large, de-identified database on deliveries that will continue to increase the accuracy of information available in real-time for doctors in the delivery room,” said IV Ramakrishnan, Co-PI, Associate Dean of Research and Professor of Computer Science. “It will be an open source learning tool using language familiar to doctors that they can access. Plus, what we are creating will benefit areas of medicine beyond childbirth. For example, the applied methodology may help neurologists monitor brain signals and assist in bringing patients in coma back to normal, if possible.”
Perhaps one of the most crucial pieces for the development of this program is the addition of a psychological component. An understanding of how doctors make decisions is critical to optimizing the exchange between doctor and machine to ensure the best choice is made in any given situation.
“Even the most powerful machine learning systems will be ineffective if human users do not incorporate the provided recommendations into their decision making,” said Christian Luhmann, Professor of Psychology in the College of Arts and Sciences. “Successful adoption of the proposed system will require doctors to thoroughly trust the guidance it generates. Medical doctors are highly trained experts and are unlikely to embrace ‘black box’ recommendations.”
Principal Investigators (PIs):
Petar M. Djurić, Chair, Department of Electrical and Computer Engineering, CEAS
IV Ramakrishnan, Professor of Computer Science and Associate Dean of Research, CEAS
— Dick Wolfe