Age Estimation Of Adolescents Based On Knee T1WI MRI Using Deep Learning

Objective: To combine sagittal T1-weighted knee MRI imaging with deep learning to provide a new idea for the application of deep learning to infer the age of adolescents.Methods:A total of 1 212 male knee MRI T1 WI images aged 10 to 18 years from the Affiliated Hospital of Qingdao University(internal data set)and 341 male knee MRI T1 WI images from Qingdao Municipal Hospital(external data set)from January 2015 to December 2021 were retrospectively collected.After image standardization,the distal femoral and proximal tibial epiphyseal plates of the internal dataset images were labeled with itk-SNAP software.Then the labeled images were trained by U-Net neural network for image segmentation,so that the region of interest(distal femur and proximal tibia epiphyseal plate)could be automatically identified by U-Net convolutional neural network.Finally,the internal data set was divided into training group(971)and validation group(241)according to the random number table method according to the age of each group for the establishment of the model,and the external data set(test group)was used for the evaluation of the model.The performance of the model was tested and verified by accuracy,precision,recall,sensitivity,specificity and other indicators.Results: The accuracy,precision,recall,specificity and sensitivity of the validation group were 85.713%,84.732%,85.713%,97.729% and 85.713%,respectively.The accuracy,precision,recall,specificity,and sensitivity of the external test set were 82.578%,83.145%,82.578%,97.442%,and 82.578%,respectively.After homogeneity of variance test and independent sample t test,there was no significant difference between the results of the validation set and the test group(P > 0.05).Conclusion:Deep learning combined with knee MRI sagittal TIWI imaging has high accuracy and can be applied to age estimation of adolescents aged 10-18 years.