Key result
CNN-Transformer models classify electrohysterography recording periods with ~0.80 ROC-AUC.
Why the study?
Splitting overlapping segments or paired pregnancy recordings between training and test sets can bias deep learning evaluation of electrohysterography.
Does a pregnancy-wise, leakage-controlled CNN-Transformer framework using wavelet scalograms accurately distinguish early- from later-recorded electrohysterography observations in the Induced Cesarean Electrohysterogram Dataset?
Observational (n=91)
Does a pregnancy-wise, leakage-controlled CNN-Transformer framework using wavelet scalograms accurately distinguish early- from later-recorded electrohysterography observations in the Induced Cesarean Electrohysterogram Dataset?
Answers are grounded in this paper and cite their sources.
A pregnancy-wise, leakage-controlled CNN-Transformer ensemble using EHG wavelet scalograms achieved a mean ROC-AUC of 0.804 in distinguishing early from later pregnancy recordings, though external validation is needed to establish clinical utility.
Electrohysterography (EHG) provides a non-invasive measure of uterine electrical activity. Splitting overlapping segments or paired pregnancy recordings between training and test sets can bias deep learning evaluation. We evaluated a pregnancy-wise, leakage-controlled framework for distinguishing early- from later-recorded observations in the Induced Cesarean Electrohysterogram Dataset (ICEHG DS). The dataset comprised 126 records from 91 pregnancies: 35 with paired early and later recordings and 56 with one recording. The S1, S2, and S3 channels were band-pass filtered, segmented into overlapping 120-s windows, and converted to wavelet scalograms. Separate CNN–Transformer models were trained for each channel and a synchronized red–green–blue (RGB) representation. Segment probabilities were averaged within each record, and representation-specific probabilities were combined using a five-outer/three-inner nested GroupKFold design. The optimized ensemble achieved a mean outer-fold raw ROC-AUC of 0.804 ± 0.062 (95% CI: 0.725–0.880), accuracy of 0.635, sensitivity of 0.919, specificity of 0.359, and balanced accuracy of 0.639. Equal-weight four-representation fusion was numerically highest (0.807 ± 0.069); optimized weighting showed no advantage. In 30/35 paired pregnancies (85.7%; exact 95% CI: 69.7–95.2%), early recordings received higher raw early-class scores (sign-test p = 2.24 × 10−5), without establishing physiological changes or clinical utility. External longitudinal validation is required.
No takes yet. Share an insight, caveat, or question.
SAĞLAM et al. (2026) conducted an observational in Pregnancy / Induced Cesarean delivery (n=91). Wavelet scalogram CNN-Transformer ensemble with pregnancy-wise nested GroupKFold validation vs. Equal-weight four-representation fusion was evaluated on Classification of early- vs later-recorded electrohysterography observations (mean outer-fold raw ROC-AUC) (ROC-AUC 0.804, 95% CI 0.725-0.880, p=2.24 × 10−5). Wavelet scalogram CNN-Transformer models classified EHG recording periods with a mean ROC-AUC of 0.804 and correctly ranked 85.7% of paired pregnancies.
Synapse has enriched 10 related papers. Consider them for comparative context: