Digital signal extraction approach for cardiotocography image
Junhong Xiang,Wanrong Yang,Hua Zhang,Fangyu Zhu,Shanshan Pu,Rui Li,Che Wang,Zhonghong Yan,Wang Li
DOI: https://doi.org/10.1016/j.cmpb.2022.107089
IF: 6.1
2022-10-01
Computer Methods and Programs in Biomedicine
Abstract:Background and objectiveCardiotocography, commonly called CTG, has become an indispensable auxiliary examination in obstetrics. Generally, CTG is provided in the form of a report, so the fetal heart rate and uterine contraction signals have to be extracted from the CTG images. However, most studies focused on reading data for a single curve, and the influence of complex backgrounds was usually not considered.MethodsAn efficient signal extraction method was proposed for the binary CTG images with complex backgrounds. Firstly, the images' background grids and symbol noise were removed by templates. Then a morphological method was used to fill breakpoints of curves. Moreover, the projection map was utilized to localize the area and the starting and ending positions of curves. Subsequently, data of the curves were extracted by column scanning. Finally, the amplitude of the extracted signal was calibrated.ResultsThis study had tested 552 CTG images simulated using the CTU-UHB database. The correlation coefficient between the extracted and original signals was 0.9991 ± 0.0030 for fetal heart rate and 0.9904 ± 0.0208 for uterine contraction, and the mean absolute error of fetal heart rate and uterine contraction were 2.4658 ± 1.8446 and 1.8025 ± 0.6155, and the root mean square error of fetal heart rate and uterine contraction were 4.2930 ± 2.9771 and 2.5214 ± 0.9640, respectively. After being validated using 293 clinical authentic CTG images, the extracted signals were remarkably similar to the original counterparts, and no significant differences were observed.ConclusionsThe proposed method could effectively extract the fetal heart rate and uterine contraction signals from the binary CTG images with complex backgrounds.
engineering, biomedical,computer science, interdisciplinary applications,medical informatics, theory & methods