A robust artificial intelligence method detects almost non-reactive Non-stress pattern: What we expect?

Abstract Objective To compare interpretation of prenatal non-stress (NST) pattern between obstetricians and artificial intelligence (AI), and to determine the degree of agreement of AI system. Methods One thousand records of prenatal NST pattern with 20 to 30 minutes were interpreted using AI system, as well as visual interpretation of five obstetricians, to explore the agreement and accuracy of AI system. Weighted kappa was used to assess reliability of AI for interpretation of prenatal NST pattern. Results A total of 967 cases enroll in this study. Moderate agreement (kappa, 0.48) was found among the five obstetricians for FHR pattern during antepartum period. The AI system recognized NST pattern like obstetricians, with a moderate kappa coefficient of agreement of 0.42. When AI was used to assess the strong consistent set of inter-obstetricians, the agreement was high (kappa, 0.75). AI could identify major non-reactive NST pattern, with high sensitivity of 91.67%. A concordant identification was observed 71.76% of preterm cases and 66.05% of term cases. Conclusion Based on the visual interpretation of obstetricians, AI was excellent for antepartum FHR monitoring interpretation, regardless gestational age. Further, AI showed a competitive ability to identify non-reactive NST pattern and the potential avoidance of unnecessary clinical intervention.