Breathing rate asymmetry for intrapartum detection of severe preeclampsia using machine learning classifiers

Severe preeclampsia significantly contributes to maternal and fetal morbidity and mortality. Present diagnostic techniques, including blood pressure monitoring and proteinuria tests, are often time-consuming and invasive, highlighting the necessity for more effective diagnostic tools. This study explores the use of Breathing Rate Asymmetry (BRA) and entropy-derived indices, combined with machine learning algorithms, to classify severe preeclampsia during labor. Breath-to-breath time series from 21 normotensive and 21 severe preeclampsia pregnant women were assessed. Key features of BRA analysis included asymmetry indices—Guzik’s (G%) and Porta’s (P%)—and Multiscale Fuzzy Entropy (MFE), alongside the mean breathing rate (mBR). Various tree-based classifiers, such as Boosted Trees, Medium Trees, and RUSBoosted Trees, were evaluated using a leave-pair-out cross-validation approach. The Boosted Tree classifier achieved the highest test accuracy of 80.00 % (sensitivity 80.48 %, specificity 79.52 %; F1-score 78.35 %, AUC 0.87). Other models, such as Medium Trees and RUSBoosted Trees, attained moderate accuracies of ∼ 71–78 %. To our knowledge, no previous publication has demonstrated the feasibility of using BRA indices combined with machine learning for intrapartum detection of severe preeclampsia. The timely diagnosis of severe preeclampsia through classifiers based on BRA and entropy-derived indices could support physicians in early interventions. We believe that the future implementation of such AI-driven tools could optimize diagnosis and enhance prenatal care in preeclamptic women.