Fig. 1
From: Automated interpretation of cardiotocography using deep learning in a nationwide multicenter study
Study flowchart. The figure illustrates the process of data preparation and labeling for the development and external test datasets. The development dataset includes deliveries from 11 hospitals, with specific exclusion criteria and an 8:1:1 split into training, validation, and internal test sets. The external test dataset is derived from 3 hospitals and divided into three subsets after applying similar exclusion criteria. A two-stage labeling process was conducted by obstetricians using 5-minute signal intervals, categorizing labels as normal or abnormal based on specific criteria. Disagreements were resolved by senior obstetricians with over 15 years of experience.
