Assisting low dose CT measurement of bone mineral density with 3D-Densenet neural network technology:a study on consistency with quantitative CT
10.3969/j.issn.1002-1671.2024.09.028
- VernacularTitle:基于3D-Densenet神经网络技术助力低剂量CT测量骨密度:与定量CT一致性研究
- Author:
Duoshan MA
1
;
Danyang SU
;
Yan WANG
;
Jianbo GAO
;
Yan WU
Author Information
1. 郑州大学第一附属医院 放射科,河南 郑州 450052
- Keywords:
osteoporosis;
bone mineral density;
quantitative computed tomography;
artificial intelligence
- From:
Journal of Practical Radiology
2024;40(9):1518-1522
- CountryChina
- Language:Chinese
-
Abstract:
Objective To evaluate the correlation and consistency between an artificial intelligence(AI)bone mineral density(BMD)measurement system based on 3D-Densenet neural network technology and quantitative computed tomography(QCT)in measuring BMD,as well as to assess its effectiveness in diagnosing osteoporosis(OP).Methods A total of 1 201 participants who underwent low dose computed tomography(LDCT)were retrospectively included.The AI BMD measurement system and QCT were utilized to measure the BMD of T12,L1,L2 vertebrae,and the average BMD.Consistency and correlation of BMD measurements between the two methods were assessed using Bland-Altman,Pearson,and Kappa analyses.With QCT results as the reference standard,the receiver operating characteristic(ROC)curve was drawn to evaluate the accuracy of AI BMD measurement system in diagnosing OP.Results The r and r2 for the average BMD measured by the two methods were 0.997 and 0.993,respectively.The Kappa value for the diagnosis of normal BMD,low bone mass,and OP using the AI BMD measurement system was 0.905.The area under the curve(AUC)for diagnosing OP using the AI BMD measurement system was 0.998,with a sensitivity of 0.888 and specificity of 0.997.Conclusion The AI BMD measurement system based on 3D-Densenet neural network technology has a high correlation and consistency with the QCT measurement result,which can accurately diagnose normal BMD,low bone mass,and OP.
