| Objective:To propose a fast,automatic,accurate and comprehensive measurement tool for AIS Cobb angle by using deep learning technology,so as to improve the efficiency of diagnosis and treatment.Through the continuous promotion of the latter model,we hope to improve the diagnostic efficiency of AIS scoliosis angle measurement in primary medical institutions as a whole.Methods:1.1500 X-ray images of scoliosis were collected from the database of Radiology Department of Guilin People’s Hospital,and finally 1051 images met the inclusion criteria.Through the pre-processing of the original data image,such as vertebral body recognition,noise reduction,removal of irrelevant areas,making training labels and so on,based on the original U-Net spine segmentation model,a new model is proposed through improvement.Later,through artificial correction,the measurement accuracy of the new model for coronal Cobb angle of AIS Lenke1 patients is further improved.2.The anteroposterior and lateral images of Lenke 1 spine of AIS patients treated in our hospital from January 2020 to June 2021 were selected,and finally 46 patients were included in the study.The new model was used to measure the Cobb angles of 46 patients included in the study.In addition,two groups of physicians were selected and divided into group A and group B(each group consisted of three physicians).The Cobb angles of patients in group A and group B were measured by Surgimap software and Photoshop software respectively.The measured data were analyzed and sorted out,and the data were grouped and named as new model method,A1(Surgimap method),A2(Photoshop method),B1(Surgimap method)and B2(Photoshop method).The Kolmogorov-Smirnov test was used to verify the normality of the data distribution.Paired t test was used to compare the data in accordance with normal distribution,and Wilcoxon rank sum test was used to compare the data in accordance with abnormal distribution.The reliability of measurement results within and between groups A and B was analyzed by intraclass correlation coefficient(ICC),and Bland-Altman plot was drawn to understand the consistency of measurement results by visual inspection of different inter-observer measurements,and the measurement time of each method was recorded.Results:1.Through AI deep learning technology,a new AIS Cobb angle measurement model with simple operation and quick response is made in this study.The model can automatically select the upper and lower end vertebrae without manual selection.2.The Cobb angle and measurement time measured by the new model method,group A and group B physicians using Surgimap software and Photoshop software were recorded and statistically analyzed.Kolmogorov-Smirnov test was used to test that the Cobb angle data of the five groups did not conform to normal distribution.Wilcoxon rank sum test was used to test the new model method,Surgimap method and Photoshop method,P> 0.05.There was no statistical difference between the new model method and the two methods in measuring the coronal Cobb angle of AIS.ICC > 0.9 within group A and B and between group A and B.Bland-Altman plots showed that different surveyors had good consistency in the measurement of Cobb angle and the selection of upper and lower end vertebrae.The measurement time of the new model method was 1-2 seconds,and the average measurement time of Surgimap software or Photoshop software was 230-240 seconds.The measurement time of the new model method was significantly reduced and could not be accurately recorded,so no statistical analysis was made.Conclusion: The model of adolescent idiopathic scoliosis image recognition based on AI deep learning technology proposed in this study has high reliability,fast speed and accuracy,and has reached the level of senior specialists.Through further improvement,it can be popularized and applied in clinic. |
PDF Full Text Request