Diagnostic Performance of Machine Learning Models Analyzing Cardiotocography for Intrapallial Fetal Acidosis Detection

International Journal of Obstetrics and Gynecology | Vol. 14, No. 8, August 2026 | pp. 109–120

DOI: 10.46882/2026/IJOG/000486

Original Research Article

Title: Diagnostic Performance of Machine Learning Models Analyzing Cardiotocography for Intrapallial Fetal Acidosis Detection

Names of Authors: Fatima B. Aliyu¹, Tunde M. Bakare¹

Authors’ Affiliations:
¹ Department of Obstetrics and Gynecology, Aminu Kano Teaching Hospital, Kano, Nigeria

Abstract:
Intrapartum cardiotocography (CTG) interpretation is prone to high inter-observer variability, often contributing to unnecessary operative deliveries or delayed interventions for fetal distress. This retrospective diagnostic accuracy study assessed a machine learning (ML) framework developed to interpret intrapartum CTG traces for the prediction of neonatal umbilical artery academia (pH < 7.15). A database of 1,500 laboring traces recorded within 60 minutes prior to delivery was utilized to train and validate three ML models: Random Forest, Support Vector Machine, and a Deep Convolutional Neural Network (CNN). Performance metrics were checked against standard evaluations performed by expert obstetricians blinded to the clinical outcomes. The reference standard was newborn umbilical cord blood gas analysis. Neonatal acidosis (pH < 7.15) was present in 12.4% of cases. The deep CNN model outperformed both the alternative ML algorithms and clinical observers, demonstrating a sensitivity of 86.4%, a specificity of 91.2%, a positive predictive value of 62.4%, and a negative predictive value of 97.4% (Area Under the Curve [AUC] = 0.93). In comparison, visual expert interpretation according to traditional guidelines achieved a sensitivity of 61.2% and a specificity of 76.5% (p < 0.001), showing a higher rate of false-positive indications for emergency interventions. Automated ML analysis of intrapartum cardiotocography features substantially increases the detection of intrapartum fetal acidosis while providing a robust diagnostic baseline to lower false-positive interpretations and avoidable surgical deliveries.

Keywords: Cardiotocography; Machine Learning; Deep Learning; Fetal Acidosis; Intrapartum Surveillance; Umbilical Cord pH.

Manuscript Timeline: Received: April 11, 2026; Revised: June 03, 2026; Accepted: July 01, 2026; Published: August 05, 2026.

Citation: Aliyu FB, Bakare TM. Diagnostic Performance of Machine Learning Models Analyzing Cardiotocography for Intrapallial Fetal Acidosis Detection. International Journal of Obstetrics and Gynecology. 2026; 14(8): 109–120.

Subscribe to read the full article: https://mail.internationalscholarsjournals.org/subscribe-to-read