| Objective:Multi-slice spiral computed tomography(MSCT)of the pelvis was applied to explore the differences between the virtual reconstructed skeletal model and the real skeleton,to establish a virtual reference sample library for age inference of the pubic symphysis,and to construct an automatic age assessment model of the pubic symphysis using artificial intelligence technology machine learning algorithms.Methods:(1)Retrospective collection of pelvic CT imaging data from patients who attended Hanzhong Hospital and underwent clinical abdominal CT scans from 2015 to 2021,and a total of 895 cases meeting the inclusion and exclusion criteria were screened as the reference sample.Image data from subjects undergoing virtual autopsy(postmortem CT,PMCT)prior to forensic pathology procedures at the Institute of Forensic Science between 2011 and 2015 and 100 pairs of well-preserved skeletal remains of the pubic symphysis collected from 2000 to 2008 at the Institute of Forensic Science of Shanghai Public Security Bureau were collected as a test set.All CT images were reconstructed into virtual skeletal models using 3D reconstruction software for modeling.First,the 100 pairs of pubic symphysis skeletons were modeled after CT scans and compared with the real skeletons to verify whether the differences in morphological performance between the two were statistically significant when age was inferred.Secondly,all virtual skeletons of the reference sample were classified into1-6 stages according to the description of Suchey-Brooks method,and descriptive statistics were performed to construct a virtual reference sample library.Finally three test samples were inferred pubic bone age using the reference sample library and compared with the original Suchey-Brooks reference sample library to assess accuracy.(2)649 abdominal CT samples of Han Chinese in Shaanxi aged 15-90 years were collected as the training set,and seven morphological features(Ridges and furrows,pubic tubercle,Lower extremity of the symphysial surface,Ventral rampart,Dorsal margin,Ventral beveling,Upper extremity of the symphysial surface)were selected with reference to the existing scoring system after obtaining the 3D reconstruction model of the pubic symphysis to establish a new scoring system for the morphological features of the pubic symphysis,using six machine learning regression algorithms respectively The male and female age inference models were developed.Each model was tested using 85 clinical CT samples(35 males and 50 females),96 postmortem CT samples(PMCT)(51 males and 45 females),and 82 real pubic symphysis samples(40 males and 42 females),respectively,to select the optimal age inference model.Results:(1)The results showed using our Chinese sample has a lower inaccuracy for the dry bones(male over 35 years and female over 45 years),and for virtual bone models of dry bone’s CT scans(male over 55 years and female over 45 years),and for virtual bone models from post-mortem computed tomography(male over 35 years of age and female over 55 years),compared to using the SB sample.Additionally,we compared virtual bone models and physical bone specimens from the 100 pubic symphyses in the test sample,Cohen’s Kappa results showed that there was substantial agreement between the Suchey-Brooks method for estimating the ages of virtual and real bones models.(2)The newly established pubic symphysis feature classes all had a strong correlation with age(r>0.700).The Ada Boost model performed best in the male sample,with lower MAEs of 5.23,7.04,and 7.55 years after examination on the clinical CT 3D reconstruction model,PMCT 3D reconstruction model,and real pubic symphysis 3D reconstruction model,respectively.The MAE was less than 10 years in the age interval of 40 to 70 years.The GBR model performed best in the female sample and also obtained lower MAE after examination on the clinical CT 3D reconstruction model,PMCT 3D reconstruction model,and real pubic symphysis 3D reconstruction model,with 5.16,7.02,and 7.71 years,respectively.The MAE was less than 10 years in the age interval of 10 to 70 years.Conclusions:1.The established virtual pubic symphysis reference sample library can be used as a regional standard for age inference of skeletal remains and virtual skeletal models by Suchey-Brock method.2.The established virtual pubic bone joint morphological feature scoring system combined with machine learning regression algorithm to establish age inference model has certain reliability and accuracy,and can be used for age inference of adults. |
PDF Full Text Request