Related Experiment Video
Updated: Aug 10, 2025
06:56
Human Fetal Blood Flow Quantification with Magnetic Resonance Imaging and Motion Compensation
Published on: January 7, 2021
2.5K
A machine learning pipeline to classify foetal heart rate deceleration with optimal feature set
Sahana Das1, Sk Md Obaidullah2, Mufti Mahmud3
1West Bengal State University, Kolkata, 700126, India.
Scientific Reports
|February 13, 2023
Summary
This study introduces a machine learning pipeline for classifying fetal heart rate (FHR) decelerations using cardiotocography (CTG). A novel fuzzy logic approach for feature extraction achieved 97.94% accuracy with Multilayer Perceptron, outperforming other methods.
Area of Science:
- Obstetrics and Gynecology
- Medical Imaging and Signal Processing
- Computational Biology and Bioinformatics
Background:
- Fetal Heart Rate (FHR) deceleration analysis via Cardiotocography (CTG) is crucial for assessing fetal well-being.
- Accurate classification of FHR decelerations depends on precise estimation of event points (EP) from FHR and Uterine Contraction Pressure (UCP).
- Current visual inspection methods for CTG interpretation can be subjective and prone to variability.
Purpose of the Study:
- To develop and evaluate a machine learning (ML) pipeline for automated classification of FHR decelerations.
- To compare the efficacy of different feature set generation approaches for ML-based deceleration classification.
- To identify the optimal method for feature extraction and ML model selection for accurate FHR deceleration classification.
Main Methods:
- Proposed a deceleration classification pipeline utilizing four ML models: Multilayer Perceptron (MLP), Random Forest (RF), Naïve Bayes (NB), and Simple Logistics Regression.
- Systematically compared three feature set creation approaches from detected EPs: a novel fuzzy logic (FL)-based method, expert clinical annotation, and calculation using National Institute of Child Health and Human Development guidelines.
- Validated classification results using statistical metrics including Receiver Operating Characteristic (ROC) curves, Intra-class Correlation Coefficient (ICC), Deming regression, and Bland-Altman plots.
Main Results:
- The MLP model achieved the highest classification accuracy of 97.94% when using the proposed FL-based feature set for EP annotation.
- In contrast, the RF model achieved only 63.92% accuracy with clinician-annotated EPs.
- The FL-annotated feature set demonstrated superior performance for classifying FHR decelerations compared to expert annotation and guideline-based calculations.
Conclusions:
- The fuzzy logic-based feature set is optimal for classifying FHR decelerations using ML.
- Automated classification of FHR decelerations from CTG data can be significantly improved with advanced feature engineering techniques.
- The proposed ML pipeline offers a promising tool for objective and accurate assessment of fetal status during labor.
