Dataset for AI-Driven Centimetric Regression of Cervical Dilation from Electrohysterography and Maternal Data

This dataset contains electrohysterography (EHG) signal features and maternal clinical data used to develop and validate a machine learning model for centimetric cervical dilation regression. The records were obtained from 67 independent patients. Feature Extraction Methodology: To capture the dynamic temporal evolution of uterine electrical activity, the extraction process was designed as follows: 1. Signal Segmentation & Frequency Bands: Each 10-minute EHG recording was divided into 9 overlapping 120-second segments. The signal was filtered into three distinct frequency bands to capture different physiological components: Band 1 (B1): 0.2–1 Hz (total uterine activity) Band 2 (B2): 0.34–1 Hz (fast-wave-high / uterine-dominant component) Band 3 (B3): 0.2–0.34 Hz (lower-frequency component) 2. Base Feature Extraction: For each of the 9 segments and across all 3 frequency bands, 7 base EHG metrics were computed: Root-Mean-Square amplitude (RMS), Area under the rectified envelope (AUC), Zero-crossing rate (ZCR), Mean frequency (MNF), Median frequency (MDF), Sample entropy (SampEn), and Bubble entropy (BubbEn). This generated a longitudinal trajectory of 9 sequential values for each base feature. 3. Temporal Descriptors & Final Concatenation: To summarize these trajectories into a single static profile per patient, 8 temporal descriptors were extracted from the 9-segment trajectory of each base feature: mean: Average value across the 9 segments. slope: Linear rate of change (trend) over the segments. max: Peak value reached. std: Standard deviation, representing temporal variability. maxdiff: Steepest positive rate of change between adjacent segments. mindiff: Steepest negative rate of change between adjacent segments. VT: Total variation (sum of absolute differences). area: Discrete integral of the feature's longitudinal curve. This temporal transformation projected the data into 168 EHG features per recording (7 base metrics × 3 bands × 8 temporal descriptors). This array was then concatenated with 4 recording-level contraction and clinical variables: Low-Contraction count (LC), High-Contraction count (HC), Maternal Age (MA), and Gestational Age (GA), yielding a 172-dimensional feature vector. Finally, the ground-truth target variable—the manual cervical dilation measurement expressed in centimeters (cm)—was appended as the last column of the dataset. Usage: These data are intended for the reproduction of the regression results presented in the study: "AI-Driven Centimetric Regression of Cervical Dilation from Electrohysterography and Maternal Data: Proof of Concept for Non-Invasive Labor-Dystocia Screening" (DOI: [to be defined]). Format & Nomenclature: CSV file containing numeric features. Columns are named following the strict convention:[BaseFeature][Band]_[TemporalDescriptor]. For example, RMS1_slope represents the "slope" descriptor of the "RMS" base feature extracted from "Band 1". The appended clinical variables retain their standard abbreviations (MA, GA, LC, HC). The final column, representing the regression target, contains the integer values of the cervical dilation (cm).