A machine-learning approach using pubic CT based on radiomics to estimate adult ages. Issue 156 (November 2022)

• Record Type: Journal Article
• Authors: Zhang, Yiying; Wang, Zhenping; Liao, Yuting; Li, Tiansheng; Xu, Xiaoling; Wu, Wenyuan; Zhou, Jie; Huang, Weiyuan; Luo, Shishi; Chen, Feng
• Is Part Of: European journal of radiology. Issue 156(2022)
• Publication Date: 2022-11
• Subjects: Age estimation -- Radiomics -- Pubis -- Computed tomography -- Machine learning; CT Computed tomography -- ROI Region of interest -- VIF Variance Inflation Factor -- AIC Akaike information criteria -- GLCM Gray level co-occurrence matrix -- GLDM Gray level dependence matrix -- GLSZM Gray level size zone matrix -- NGTDM Neighborhood Graytone Difference Matrix -- FDR False discovery rate; Medical radiology -- Periodicals; Radiology -- Periodicals; Radiologie médicale -- Périodiques; Medical radiology; Periodicals; 616.075705
• DOI: 10.1016/j.ejrad.2022.110516
• Languages: English
• ISSNs: 0720-048X

Abstract

Highlights: CT imagings present an ideal pattern of research, which can be obtained in a non-invasive way and investigated repetitively. Multivariate stepwise regression can assess correlations of radiomics texture parameters with age. Combining with computer analysis technologies, the CT radiomics-based machine learning model of age estimation is a promising research approach. Abstract: Purpose: Adult skeletal age estimation is an active research field. To evaluate the performance of a pubic CT radiomics-based machine learning model for estimating age, we established a multiple linear regression model based on radiomics and machine learning methods. Methods: A total of 355 subjects were enrolled in this retrospective study from August 2016 to August 2021, and divided into a training cohort (N = 325) and a testing cohort (N = 30). Computerized texture analysis of the semi-automatically segmentation was performed and 107 texture features were extracted from the regions. Then we used univariate linear regression and multivariate stepwise regression to assess correlations of texture parameters with age. The most vital features were used to make the best predictive model. Eventually, the established radiomics model was tested with an additional 30 patients. Results: Clinical characteristics include age, sex, height, weight and BMI were not statistically significant different between training and testing cohort ( p = 0.098–0.888). Through a multivariate regression analysis using stepwise regression, six texture parameters were found to have significant correlations with age. The regression formula estimating the age was constructed. Conclusions: The radiomics model using machine learning is considered as a new approach for age estimation from pubic symphysis CT features. Digital osteology is obtained in a non-invasive way so that it can be an ideal collection for anthropological studies.