(817) A Machine Learning Approach to Intrapartum FHRT Analysis To Identify Neonates Requiring Therapeutic Hypothermia

Objective: Examine a "sliding window time frame" as a ML approach to fetal heart rate tracing (FHRT) analysis to identify neonates that developed moderate to severe neonatal encephalopathy (NE) and underwent therapeutic hypothermia.

Study Design:

Retrospective observational case-control involving neonates born at a single academic institution and admitted to the NICU with cord pHs < 7.15. The study group included neonates meeting both physiologic (low cord pH) and neurologic criteria for moderate to severe NE, making them eligible for therapeutic hypothermia. The control group comprised neonates who, despite having low cord pHs, did not show evidence of moderate or severe NE and thus did not undergo therapeutic hypothermia.


 

The entire intrapartum FHRT for eligible neonates was analyzed to develop a ML-based model incorporating an embedded sliding time window. Isolation Forest was used to identify and cluster windows with unusual patterns as anomalies or typical patterns as normal. A Random Forest classifier was then trained to predict abnormal pattern labels. After the optimal sliding time window size was determined, its impact on the ML model's anomaly detection performance was evaluated by calculating the percentage of anomalous time windows during the entire intrapartum course for both subject groups and compared by student t-test.

Results: There were 22 study and 22 control cases with similar clinical characteristics and cord pH values. Using a 90-minute sliding time window, the ML-based model identified a significant difference in the proportion of windows with abnormal patterns between the control and study cases (4.7 % vs 73.3%, p < 0.001). The ML classifier applied to these 90-minute windows achieved 100% accuracy in distinguishing between study and control cases.

Conclusion: The ML-based model incorporating an embedded sliding time window accurately identified neonates with moderate to severe NE. This is a promising approach to intrapartum FHRT assessment with the potential to identify fetuses at risk for neurologic injury and allow for timely intervention.