Explainable Temporal Risk Trajectory Learning for Multi- Horizon Prediction of Fetal Compromise Using Intrapartum Cardiotocography and Clinical Data

Abstract Intrapartum fetal monitoring using cardiotocography (CTG) provides continuous information about fetal heart rate and uterine contraction patterns, but conventional computational approaches often formulate fetal assessment as classification at an individual temporal interval. Such formulations provide limited information about the temporal evolution of estimated fetal risk and its behaviour at future prediction horizons. This study presents an Explainable Temporal Risk Trajectory Learning (E-TRTL) framework for investigating the temporal evolution of model-estimated fetal risk and future fetal-state prediction from intrapartum CTG recordings. The framework was evaluated using the CTU-UHB Intrapartum Cardiotocography Database containing 552 CTG recordings. Quality-controlled fetal heart rate and uterine contraction signals were represented using 5-, 10-, and 20-minute temporal contexts. The framework integrates temporal feature extraction and fusion, temporal risk-trajectory modelling, fetal-state transition analysis, multi-horizon prediction, and model-level explainability. Future-state prediction was evaluated at 5-, 10-, 15-, and 30-minute horizons. Performance was assessed using multiclass classification measures, one-vs-rest ROC-AUC, and continuous risk-trajectory measures. The E-TRTL framework achieved an overall multiclass accuracy of 91.48%, with 89.00% macro precision, 87.00% macro recall, and 88.00% macro F1-score. The one-vs-rest AUC values were 0.884, 0.823, and 0.978 for the Normal, Suspicious, and Pathological categories, respectively. Future-state prediction performance was 95.72%, 94.62%, 93.72%, and 91.48% at the 5-, 10-, 15-, and 30-minute horizons, respectively. Continuous risk-trajectory evaluation produced an RMSE of 0.0442, MAE of 0.0308, R^2of 0.9536, EVS of 0.9536, and correlation coefficient of 0.9766. The results indicate that E-TRTL provides a temporal modelling formulation for examining changes in model-estimated fetal risk and evaluating future-state prediction across multiple horizons. The framework extends conventional classification-oriented analysis by jointly considering temporal risk trajectories, fetal-state transitions, forecasting, and model-level interpretation. The estimated risk trajectory represents a model-derived quantity rather than a directly measured clinical severity score. Further evaluation using independent datasets, clinically defined outcomes, calibration and uncertainty analysis, and prospective validation is required to assess generalizability and potential clinical utility.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-12
DOI
https://doi.org/10.21203/rs.3.rs-11216423/v1
Primary Topic
Artificial Intelligence in Healthcare
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Explainable Temporal Risk Trajectory Learning for Multi- Horizon Prediction of Fetal Compromise Using Intrapartum Cardiotocography and Clinical Data

K. Mekala Devi, Rajeswari P, Joseph Emerson Raja
Research Square
Artificial Intelligence in Healthcare
preprint

Explainable Temporal Risk Trajectory Learning for Multi- Horizon Prediction of Fetal Compromise Using Intrapartum Cardiotocography and Clinical Data

K. Mekala Devi, Rajeswari P, Joseph Emerson Raja
preprint en

Abstract

Abstract Intrapartum fetal monitoring using cardiotocography (CTG) provides continuous information about fetal heart rate and uterine contraction patterns, but conventional computational approaches often formulate fetal assessment as classification at an individual temporal interval. Such formulations provide limited information about the temporal evolution of estimated fetal risk and its behaviour at future prediction horizons. This study presents an Explainable Temporal Risk Trajectory Learning (E-TRTL) framework for investigating the temporal evolution of model-estimated fetal risk and future fetal-state prediction from intrapartum CTG recordings. The framework was evaluated using the CTU-UHB Intrapartum Cardiotocography Database containing 552 CTG recordings. Quality-controlled fetal heart rate and uterine contraction signals were represented using 5-, 10-, and 20-minute temporal contexts. The framework integrates temporal feature extraction and fusion, temporal risk-trajectory modelling, fetal-state transition analysis, multi-horizon prediction, and model-level explainability. Future-state prediction was evaluated at 5-, 10-, 15-, and 30-minute horizons. Performance was assessed using multiclass classification measures, one-vs-rest ROC-AUC, and continuous risk-trajectory measures. The E-TRTL framework achieved an overall multiclass accuracy of 91.48%, with 89.00% macro precision, 87.00% macro recall, and 88.00% macro F1-score. The one-vs-rest AUC values were 0.884, 0.823, and 0.978 for the Normal, Suspicious, and Pathological categories, respectively. Future-state prediction performance was 95.72%, 94.62%, 93.72%, and 91.48% at the 5-, 10-, 15-, and 30-minute horizons, respectively. Continuous risk-trajectory evaluation produced an RMSE of 0.0442, MAE of 0.0308, R^2of 0.9536, EVS of 0.9536, and correlation coefficient of 0.9766. The results indicate that E-TRTL provides a temporal modelling formulation for examining changes in model-estimated fetal risk and evaluating future-state prediction across multiple horizons. The framework extends conventional classification-oriented analysis by jointly considering temporal risk trajectories, fetal-state transitions, forecasting, and model-level interpretation. The estimated risk trajectory represents a model-derived quantity rather than a directly measured clinical severity score. Further evaluation using independent datasets, clinically defined outcomes, calibration and uncertainty analysis, and prospective validation is required to assess generalizability and potential clinical utility.

Research Square
Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology (IN), Multimedia University (MY), Jaipur National University (IN)
Multimedia University
Good health and well-being
Artificial Intelligence in Healthcare
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon