Interpretation of fetal monitoring by an artificial neural network

Abstract

Objective: To test the performance of an artificial neural network (ANN) for interpretation of electronic fetal monitor (EFM) traces. Methods: Records from 268 babies without congenital anomalies and older than 36 weeks of gestation were classified into four groups based on the presence or absence of neonatal encephalopathy (NE) and arterial cord blood gas base deficit (BD) levels above or below 12 mmol/L. The last 4 hours of EFM recordings were analyzed by specialized software to measure baseline, variability, accelerations, and decelerations. Summaries of these analyses were used to train the ANN to relate these observations to outcome. Eleven obstetricians classified a sample of the EFM records. Results: The total number of babies correctly classified by the ANN was 224 (84%). The percentages of correct classifications by the ANN and the obstetricians are shown in the table. ∗ Correct classification ANN Obstetricians NE and BD over 12 mmol/L 47/56 83.9% 57/88 64.7% ∗ NE and BD under 12 mmol/L 0/9 0% Not available No NE and BD over 12 mmol/L 60/80 85.0% 39/110 35.4% ∗ No NE and BD under 12 mmol/L 114/123 92.7% 85/110 77.3% ∗ ∗ P < 0.005. Conclusions: The ANN was able to classify babies correctly at rates that exceeded the obstetricians. Detecting more babies at risk for permanent neurological damage with fewer false positives has major ramifications for obstetrics. Studies using another multicenter dataset are underway to assess the reproducibility of these results.