Transformer imputation in CTG: a length-dependent evaluation of reconstruction methods

Clinical Snapshot

Key Findings

• P Value: Statistical significance reported selectively; spline significantly worse than others for RMSE (p-values not specified in abstract); Transformer vs linear interpolation significant only under specific (unspecified) conditions for RMSE

• Effect Size: Not reported with formal effect sizes in the abstract; relative superiority described qualitatively as consistent for correlation and conditional for RMSE

• Primary Outcome: Transformer-based imputation achieved superior waveform preservation (correlation) compared to linear interpolation across all gap lengths, and lower RMSE than spline interpolation consistently. Transformer showed greatest advantage in high-complexity imputation tasks. All methods demonstrated systematic performance degradation as gap length increased.

• Nnt Or Sensitivity: Not applicable to this signal processing methodological study; key threshold identified: reliability declines beyond 30-second gap length for all methods

• Confidence Interval: Not reported in the abstract

Clinical Application

Linear interpolation remains computationally trivial and is widely implemented in existing CTG software. Transformer-based imputation requires substantially greater computational infrastructure, model training, and validation before clinical deployment. The 30-second threshold is immediately actionable as a preprocessing rule for existing CTG AI research pipelines without requiring adoption of the Transformer model itself. In Australia, intrapartum CTG monitoring is standard of care in all tertiary and most secondary obstetric units, governed by RANZCOG guidelines. The TGA has not yet approved any standalone CTG AI decision support system for clinical use, though several are under evaluation. The RACGP and RANZCOG have emphasised the need for rigorous validation of AI tools in obstetric contexts. The 30-second gap threshold identified in this study could inform preprocessing standards for Australian CTG AI research consortia (e.g., those using the AIHW perinatal data collections). PBS and MBS implications are indirect at this stage, as the study addresses a preprocessing methodology rather than a diagnostic or therapeutic intervention. Australian researchers should note that the CTU-UHB dataset does not include Indigenous Australian pregnancies, which carry distinct risk profiles and may require separate validation. Clinicians and clinical informaticists involved in intrapartum CTG monitoring, particularly those developing or implementing CTG-based AI decision support systems requiring preprocessing of FHR signals with missing data segments

Abstract

Objective.Computer and artificial intelligence (AI) analyses are being increasingly used in intrapartum cardiotocography (CTG). However, fetal heart rate (FHR) signal loss, which frequently occurs in clinical practice, hinders visual interpretation and reduces accuracy. Although the impact is well recognized, there is no consensus on the maximum continuous gap length that can be reliably reconstructed under clinical conditions. Therefore, we aim to identify suitable imputation methods and clarify the clinical limits of valid missing-segment lengths.Methods.Using an open FHR dataset (CTU-UHB), we extracted continuous segments and artificially introduced Removed data of varying lengths. Using performance metrics such as difference and similarity, we compared the performance among a Transformer-based model and linear and spline interpolation. Additionally, we quantified the similarity between the Pre-impute and removed data to assess task difficulty.Results.We analyzed 2727 segments from 552 cases across multiple gap lengths. In terms of numerical accuracy root mean square error (RMSE), spline consistently performed significantly worse than others. The Transformer generally maintained a better mean accuracy than linear interpolation, although significant differences were observed only under specific conditions. Conversely, for waveform preservation (correlation), the Transformer consistently outperformed linear interpolation. Notably, in highly complex imputation tasks, the Transformer proved most robust, yielding the lowest RMSE and highest correlation. However, performance systematically degraded for all methods as gap lengths increased.Conclusion.The Transformer provides an effective baseline for FHR imputation under clinical conditions, achieving a favorable balance between waveform and numerical accuracy. By clarifying the clinical limits of valid missing-segment lengths-specifically the decline in reliability beyond 30 s-this study provides guidance for standardizing preprocessing in future CTG AI research and clinical implementation.Significance.For imputing intrapartum FHR data, the Transformer generally improves waveform reproducibility over linear interpolation for short-to-moderate gaps. Defining the reliability limit provides a crucial baseline for future CTG AI.

References

1. 1.Hirono, Y., Kai, C., Ishizuka, S., & Kasai, S. (2026). Transformer imputation in CTG: a length-dependent evaluation of reconstruction methods. Biomedical Physics & Engineering Express. https://doi.org/10.1088/2057-1976/ae7c08