Recent Advances in Cardiotocography (CTG)
Computerized CTG analysis using machine learning and artificial intelligence represents the most significant recent advance in cardiotocography, though current evidence shows it has not yet demonstrated clear superiority over conventional visual interpretation in reducing adverse perinatal outcomes.
Computerized CTG Analysis and Clinical Outcomes
The primary advancement in CTG technology involves computerized analysis systems designed to reduce inter- and intra-observer variability, which remains the fundamental limitation of conventional visual CTG interpretation 1, 2.
Evidence on Clinical Effectiveness
Four out of five recent randomized controlled trials demonstrated that computerized CTG analysis showed no significant reduction in metabolic acidosis rates or obstetric interventions compared to visual interpretation 1.
One study found conventional CTG with fetal blood sampling actually had lower incidence of adverse perinatal outcomes compared to computerized analysis alone 1.
The positive predictive value of abnormal CTG remains relatively poor regardless of interpretation method, leading to both under-treatment (risking fetal injury) and over-treatment (unnecessary operative interventions) 3.
Machine Learning and Deep Learning Applications
Machine learning algorithms applied to CTG signals show promise for predicting fetal hypoxia more accurately than visual interpretation alone, but significant barriers prevent clinical implementation 3, 2.
Technical Advances in Signal Processing
Time-frequency features extracted using Continuous Wavelet Transform, Wavelet Coherence, and Cross-wavelet Transform combined with Ensemble Cost-sensitive Support Vector Machine classifiers achieve sensitivity of 85.2% and specificity of 66.1% for abnormality detection 4.
These time-frequency features show statistically significant differences (p<0.05) in distinguishing abnormal CTG signals, whereas traditional nonlinear features do not 4.
The area under the receiver operating characteristic curve improved from 0.64 with standard SVM to 0.77 with ensemble cost-sensitive methods 4.
Critical Limitations Preventing Clinical Adoption
Most machine learning studies use the same open-access database with insufficient subject numbers for robust ML validation 3.
Varying pH thresholds are used as benchmarks for fetal hypoxia across different studies, preventing standardization and comparison 3.
Current systems lack interpretable indicators that practitioners can trust and appropriate alongside risk predictions 2.
Large, open, anonymized multicentric databases of perinatal and CTG data are not yet available for developing more accurate systems 2.
Signal Acquisition Technology Improvements
Despite advances in analysis methods, the quality of the original ultrasonic signal remains practically unchanged after introduction of pulsed operating modes and multi-chip sensors 5.
Traditional ultrasonic monitoring technology has been used for over 70 years with multiple upgrades but continues to face fundamental accuracy limitations 5.
Novel signal processing methods focusing on identifying correct radiation direction to the fetal heart and reliable beat-to-beat heart rate evaluation are proposed but not yet validated 5.
Current Clinical Reality and Recommendations
Until computerized systems demonstrate clear clinical benefit in prospective trials, conventional visual CTG interpretation following established guidelines remains standard practice 1, 2.
Key Clinical Caveats
Inter-observer and intra-observer variability in CTG interpretation remains substantial regardless of technology used 1, 3.
Clinical risk factors and human factors significantly impact CTG interpretation quality 3.
No reliable replacement exists that is equally safe, non-invasive, and accessible for widespread screening 5.
Future Directions Requiring Investigation
Development of common standards for evaluating and comparing computerized systems on retrospective cohorts 2.
Large-scale multicenter randomized controlled trials comparing computerized versus conventional CTG with standardized hypoxia benchmarks 1, 3.
Further development of decision-support software that produces interpretable, clinically actionable outputs 1, 2.