The Study Of Intelligent Quality Control In Obstetric Ultrasound Images

Part 1 Research and development of intelligent quality control model for obstetric ultrasound imagesObjectiveLabeling the obstetrics ultrasound images and to develop an intelligent quality control model.Materials and methodsA total of 57,000 images(3000 images/plane)were collected from 19 quality control planes of patients undergoing routine ultrasound examination during the second and third trimester in Shenzhen Maternity&Child Healthcare Hospital affiliated to Southern Medical University from January 1 to December 31,2020.The images were divided into training set,validation set and test set according to the proportion of 8:1:1.Four sonographers labeled the images of plane name,internal structures,amplification ratio and standard level from the training set and validation set,and two experts who qualified to prenatal ultrasound diagnosis reviewed all the labeled images to ensure the accuracy of the labeling.The intelligent quality control model was built,and the accuracy of the model judgment was verified by the test set images.ResultsThe overall accuracy of intelligent classification and standard level evaluation was 99.7%(5573/5700)and 92.1%(5134/5573),respectively,and there was no statistical difference with manual evaluation(P<0.05),the consistency intensity was strong and moderately strong-strong,respectively.ConclusionThe intelligent quality control model can accurately classify the control planes and evaluate the standard level,which can be used to complete the quality control of obstetric ultrasound images.Part 2 Clinical application of artificial intelligent quality control system in obstetric ultrasound imagesObjectiveTo explore the clinical value of artificial intelligent quality control system in obstetric ultrasound images.Materials and methodsAn Internet-based intelligent obstetric ultrasonic quality control system(IUQCS)was established to collect the uploaded ultrasound images of 19 quality control planes of fetuses in the second and third trimester from 64 hospitals in Shenzhen from January to March 2021.20%images(a total of 18,699 images)were randomly selected from each quality control plane,with 100 images as a group,a total of 187 groups.Two ultrasound specialists independently conducted manual quality assessment,and documented the results and time consumption.At the same time,re-uploaded images to IUQCS by group,and recorded the time from clicked the button ’upload’ to quality control progressed to 100%.The results and time consuming of intelligent evaluation and manual evaluation by two experts were pairwise compared between groups.For plane classification,the accuracy of intelligent evaluation was calculated,the difference of results between groups was analyzed by McNemar test,and the consistency of results between groups was analyzed by Cohen’s Kappa.For the standard level evaluation,the accuracy of the intelligent evaluation was calculated,the difference of the results between groups was analyzed by McNemar-bowker test,and the consistency of the results between groups was analyzed by Weighted Kappa.For time consumption assessment,Shapiro-Wilk was applied to conduct normal test for time difference between groups,and paired sample T test was applied for normal distribution.If the distribution is not normal,wilcoxon sign rank test is applied.P<0.05 was considered statistically significant.ResultsCompared to the manual quality control results of the two experts,the accuracy of intelligent classification of each plane was high(99.3%-99.9%),and there were no statistical difference between intelligent evaluation and manual evaluation(McNemar test P>0.05),the consistency were strong(Cohen’s Kappa>0.9);The accuracy of intelligent standard grade evaluation of each plane were moderately high(79.8%-98.6%),and there were no statistical difference between intelligent evaluation and manual evaluation(McNemar-bowker test P>0.05),the consistency were medium to strong(Weighted kappa 0.45-0.93);The average time of intelligent quality control of each group of images was significantly shorter than that of manual quality control(32.77±4.943s vs 699.82±27.694s vs 714.97±26.219s,P<0.05).ConclusionThe intelligent ultrasonic quality control system can complete the quality control tasks accurately and efficiently,and it has important clinical application value.