Accuracy of computerized detection of fetal heart rate patterns

Abstract

Objective: To measure the agreement of computerized detection of fetal heart rate (FHR) patterns with a panel of clinical experts. Method: Digital signal processing techniques and algorithms were used to measure the FHR features of baseline, variability, accelerations and decelerations. Five obstetricians used proprietary software to review, measure, and label the FHR recordings from 41.8 hours of normal and abnormal examples. A composite opinion was defined based on agreement by a majority (3/5) of the experts. Each individual and the computer were compared with the composite. Results: The following table shows the number of hours spent in baseline or nonbaseline (acceleration or deceleration) according to the composite, the portion of this time correctly labeled by the computer-based method and the range of correct detection by individual doctors. FHR State Hours Percentage by Computer Lowest percentage by MD Highest percentage by MD Baseline 26.7 86.5 90.2 96.4 Nonbaseline 15.1 91.0 89.0 96.5 Total 41.8 88.2 89.3 96.5 Conclusions: These physicians were fairly consistent in identifying fetal heart rate features. The performance of the computer was close to the five experts. Further enhancements to the software should bring the performance to the 90% level. It is now possible to study large numbers of tracings and to quantify the association of FHR patterns with newborn status and cord blood gas results.