Automated Digitization of the Cardiotocography Signals from Real Scene Image of Binary Clinic Report

Abstract

ObjectiveCardiotocography (CTG) is a popular clinical tool to assess foetal health status. Automatic CTG analysis can remove the judgement differences of inter- and intra- doctors. However, absence of CTG database has hindered the development of automatic CTG analysis based on deep learning. Therefore, digitization of CTG signal from clinic report is an important way to enrich CTG database, and in turn promote the automatic CTG analysis. MethodThe proposed digitization method extracts digital signal from the commonly used binary CTG paper reports. An adaptive region positioning algorithm based on statistical calculation is used to locate signal regions. Then, by deducing the dynamic minimal weight sum in theory, methods based on statistical calculation and on the weight sum are designed to remove grid lines. Next, according to different breakpoint types and signal line trends, different signal reconstruction methods are formulated. This realize extracting the signal line from binary background grid lines. Finally, a calibration method based on segmented sampling is designed to reduce calibration error due to smartphone lens distortion. ResultsThe experimental results show that the correlation coefficients of FHR and UC of proposed method both reached 0.98. After three expert gynaecologists's evaluations, there are no clinically relevant differences were identified between the extracted signals and the reference ones.