基于深度学习的分娩宫缩乏力子宫形态学量化研究OA

Objective To segment uterine MRI during pregnancy using a Bidirectional Convolutional Long Short-Term Memory Dense U-Net(BCDU-Net)deep learning model and evaluate the feasibility of computer-automated quantification of uterine longitudinal diameter and volume as an alternative to manual measurements by physicians.Methods Imaging data were retrospectively collected from 61 parturients in our hospital from July 2019 to November 2023 with uterine atony during delivery who had underwent pelvic MRI during pregnancy.Regions of interest(ROIs)for the uterine wall and uterine volume were manually delineated using 3D Slicer,a BCDU-Net model was developed for automatic segmentation of the uterine wall and volume.The dataset was randomly divided into training,validation,and test sets according to the task.The trained segmentation model was applied to automatically segment the uterine wall and volume,followed by computer-based measurement of uterine longitudinal diameter and volume.Results In the test set,the BCDU-Net model achieved a dice similarity coefficient(DSC)of 0.866 and an intersection over union(IoU)of 0.773 for uterine wall segmentation,and a DSC of 0.978 and an IoU of 0.955 for uterine volume segmentation.Computer-based measurements from the automated segmentation showed high agreement with manual measurements(longitudinal diameter)obtained via the Picture Archiving and Communication System,formula-based calculations(volume),and manual delineations using 3D Slicer(volume),with no statistically significant differences(P>0.05).Conclusion The BCDU-Net deep learning model was successfully developed for the automatic segmentation of the uterine wall and volume.Computer-based measurements of uterine longitudinal diameter and volume,derived from the automated segmentation results,are feasible and can serve as an alternative to manual measurements performed by physician.