Enhancing CNN for Forensics Age Estimation Using CGAN and Pseudo-Labelling

Age estimation using forensics odontology is an important process in identifying victims in criminal or mass disaster cases.Traditionally,this process is done manually by human expert.However,the speed and accuracy may vary depending on the expertise level of the human expert and other human factors such as level of fatigue and attentiveness.To improve the recognition speed and consistency,researchers have proposed automated age estimation using deep learning techniques such as Convolutional Neural Network(CNN).CNN requires many training images to obtain high percentage of recognition accuracy.Unfortunately,it is very difficult to get large number of samples of dental images for training the CNN due to the need to comply to privacy acts.A promising solution to this problem is a technique called Generative Adversarial Network(GAN).GAN is a technique that can generate synthetic images that has similar statistics as the training set.A variation of GAN called Conditional GAN(CGAN)enables the generation of the synthetic images to be controlled more precisely such that only the specified type of images will be generated.This paper proposes a CGAN for generating new dental images to increase the number of images available for training a CNN model to perform age estimation.We also propose a pseudolabelling technique to label the generated images with proper age and gender.We used the combination of real and generated images to trainDentalAge and Sex Net(DASNET),which is a CNN model for dental age estimation.Based on the experiment conducted,the accuracy,coefficient of determination(R2)and Absolute Error(AE)of DASNET have improved to 87%,0.85 and 1.18 years respectively as opposed to 74%,0.72 and 3.45 years when DASNET is trained using real,but smaller number of images.

Estimation of age in forensic anthropology:historical perspective and recent methodological advances

Estimation of age represents a central focus of forensic anthropological analysis of human skeletal remains and of the living. Advances registered in recent research include the topics of taphonomic impact, new anatomical areas of interest, histology, population variation, the dental pulp chamber, technology, mathematical approaches, biochemical analysis and techniques specifically targeting the living. This article reviews the historical development of age estimation methods and considers likely future directions.

Estimating age of Brazilians using the methods of Demirjian and Nicodemo,Moraes and Medici:A comparative analysis

This study proposed to evaluate the precision of Nicodemo, Moraes and Médici method and the method proposed by Demirjian and compare both methodologies in relation to the efficiency and applicability in the Brazilian population. The sample consisted of 124 panoramic radiographs of subjects in the age group from 7 to 16 years old, 66 patients female and 58 male. For the comparison of the genders in relation to the average percentage difference, the t-Student test was considered. In average the method of Nicodemo, Moraes and Médici presented for the female gender an estimated result 14.7% lower than the real age. For the male gender the estimative obtained was that for each year of variation of the real age the method of Nicodemo estimates a variation of 0.48 years, with a confidence interval of 95% given by (0.43;0.54). For the method of Demirjian the result was that for each year of variation of the real age the method of Demirjian estimates a variation of 1.00 year, with a confidence interval of 95% given by (0.85;1.15) for the male gender and 0.96 years, with a confidence interval of 95% given by (0.84;1.08) for the female gender. It is concluded that the Method of Nicodemo, Moraes and Médici tends to underestimate the real age and that as the real age increases this underestimate becomes higher. Now, the method of Demirjian tends to over- estimate the real age in 8.2 months for the male gender and in 7.1 for the female one in average.

Vitality and wound-age estimation in forensic pathology:review and future prospects

Determining the age of a wound is challenging in forensic pathology,but it can contribute to the reconstruction of crime scenes and lead to arrest of suspects.Forensic scholars have tended to focus on evaluating wound vitality and determining the time elapsed since the wound was sustained.Recent progress in forensic techniques,particularly high-throughput analyses,has enabled evaluation of materials at the cellular and molecular levels,as well as simultaneous assessment of multiple markers.This paper provides an update on wound-age estimation in forensic pathology,summarizes the recent literature,and considers useful additional informa-tion provided by each marker.Finally,the future prospects for estimating wound age in foren-sic practise are discussed with the hope of providing something useful for further study.

Use of deep learning in forensic sex estimation of virtual pelvic models from the Han population

Accurate sex estimation is crucial to determine the identity of human skeletal remains effectively.Here,we developed convolutional neural network(CNN)models for sex estimation on virtual hemi-pelvic regions,including the ventral pubis(VP),dorsal pubis(DP),greater sciatic notch(GSN),pelvic inlet(PI),ischium,and acetabulum from the Han population and compared these models with two experienced forensic anthropologists using morphological methods.A Computed Tomography(CT)dataset of 862 individuals was divided into the subgroups of training,validation,and testing,respectively.The CT-based virtual hemi-pelvises from the training and validation groups were used to calibrate sex estimation models;and then a testing dataset was used to evaluate the performance of the trained models and two human experts on the sex estimation of specific pelvic regions in terms of overall accuracy,sensitivity,specificity,F1 score,and receiver operating characteristic(ROC)curve.Except for the ischium and acetabulum,the CNN models trained with the VP,DP,GSN,and PI images achieved excellent results with all the prediction metrics over 0.9.All accuracies were superior to those of the two forensic anthropologists in the independent testing.Notably,the heatmap results confirmed that the trained CNN models were focused on traditional sexual anatomic traits for sex classification.This study demonstrates the potential of AI techniques based on the radiological dataset in sex estimation of virtual pelvic models.The excellent sex estimation performance obtained by the CNN models indicates that this method is valuable to proceed with in prospective forensic trials.

Recent advancements in the analysis of bone microstructure: new dimensions in forensic anthropology

Bone is a mechanically active,three-dimensionally(3D)complex,and dynamic tissue that changes in structure over the human lifespan.Bone tissue exists and remodels in 3D and changes over time,introducing a fourth dimension.The products of the remodelling process,secondary and fragmentary osteons,have been studied substantially using traditional two-dimensional(2D)techniques.As a result,much has been learned regarding the biological information encrypted in the histomorphology of bone,yielding a wealth of information relating to skeletal structure and function.Three-dimensional imaging modalities,however,hold the potential to provide a much more comprehensive understanding of bone microarchitecture.The visualization and analysis of bone using high-resolution 3D imaging will improve current understandings of bone biology and have numerous applications in both biological anthropology and biomedicine.Through recent technological advancements,we can hone current anthropological applications of the analysis of bone microstructure and accelerate research into the third and fourth dimensional realms.This review will explore the methodological approaches used historically by anthropologists to assess cortical bone microstructure,spanning from histology to current ex vivo imaging modalities,discuss the growing capabilities of in vivo imaging,and conclude with an introduction of novel nonhistological modalities for investigating bone quality.

Contributions of anatomy to forensic sex estimation: focus on head and neck bones

This study sought to provide an up-to-date review of the importance of anatomy to human identification,focusing on the usefulness of anatomical knowledge about the head and neck bones and teeth to sex estimation in routine forensic anthropology methods.A detailed search of osteology applications in forensic sex estimation was conducted through the electronic databases for the 10 years prior to July 2020.Relevant articles and classic literature on the subject were gathered and outlined in this review.Among the available literature,several metric analyses showed accuracy superior to 80%in sexual diagnosis.Angles measured from the inclination of glabellae and analysis of the external frontal bone surface through three-dimensional computer-aided design emerge as reliable cranial indexes for sex estimation.in the mandible,the condylar and coronoid height,bigonial width,and condylion-gonion distance express significant sexual dimorphism.Measurements of the canine are the best option for sex estimation using teeth,as well as the thickness of the dentine or enamel of incisors.The axis vertebra surpasses other neck bones for sex estimation due to its atypical shape and the presence of the odontoid process.Metric analyses based on anatomy can provide reliable accuracy in sexual diagnosis.Adequate training and anatomical knowledge can reduce bias and interobserver differences,and the use of three-dimensional models and computed tomography images can enhance the accuracy of these methods for sex estimation.However,every method should be validated before being applied to a different population.

基于磁共振成像技术的牙龄推断研究进展

牙龄推断是法医学年龄推断的重要内容和实现路径之一,影像技术是牙龄推断的重要技术手段。近年来,有研究初步证实了MRI在评估牙发育方面的可行性,该技术为牙发育评估提供了新的视角和可能性,有望成为更安全且准确的牙龄推断技术手段,但仍需进一步验证其准确性和适应性。本文综述了MRI技术在牙发育和年龄推断中的研究现状、研究瓶颈和局限性,为基于MRI技术的牙龄推断研究提供参考和借鉴。

基于机器学习和颅缝CT-MPR技术的北方成人年龄推断

目的运用CT和多平面重组(multiplanar reformation,MPR)技术获取颅缝断层图像,建立中国北方汉族成人年龄推断模型,探讨颅缝闭合规律在中国北方汉族人群年龄推断中的适用性。方法回顾性收集29~80岁健康北方汉族成人头部CT样本132例。对颅骨进行容积重组(volume reconstruction,VR)和MPR,每例样本生成160张颅缝断层图像。根据颅缝闭合分级标准对颅缝MPR图像进行评分,分别计算矢状缝、左右侧冠状缝和左右侧人字缝的平均闭合等级。以上述等级为自变量,建立北方汉族成人年龄推断的线性回归模型和梯度提升回归、支持向量回归、决策树回归和贝叶斯岭回归4种机器学习模型,并评估各模型推断年龄的准确性。结果各颅缝闭合等级均与年龄呈正相关,其中矢状缝相关性最高。4种机器学习模型年龄推断的准确性均高于线性回归模型,其中支持向量回归模型的准确性最高,平均绝对误差为9.542岁。结论机器学习模型和颅骨CT-MPR技术可联合用于中国北方汉族成人的年龄推断,但在法医学实践中仍需与其他成人年龄推断指标联合使用。

Demirjian法结合机器学习算法推断北方汉族儿童及青少年牙龄

目的探讨Demirjian法结合机器学习算法在北方汉族儿童及青少年牙龄推断中的应用价值。方法收集10256例我国北方汉族5~24岁人群的口腔全景片,运用Demirjian法对左下颌8颗恒牙的发育进行分期,并结合支持向量回归、梯度提升回归、线性回归、随机森林回归和决策树回归等多种机器学习算法,分别基于总样本、女性样本和男性样本构建年龄推断模型,并评价不同机器学习算法在3组样本中的拟合性能。结果对于总样本和女性样本,推断准确率最高的模型均为支持向量回归模型;对于男性样本,推断准确率最高的模型为梯度提升回归模型。最佳年龄推断模型在总样本、女性样本和男性样本的平均绝对误差分别为1.2463、1.2818和1.1538岁。最佳年龄推断模型对各年龄区间的推断准确率不同,对于18岁以下人群的年龄推断相对准确。结论本研究构建的年龄推断机器学习模型在我国北方汉族儿童及青少年中具有较好的准确率,但在成年人群中的推断效果不理想,可以考虑联合其他变量以提高年龄推断的准确性。

基于第三磨牙根管可见度判断华东地区个体是否年满18周岁

目的研究华东地区个体下颌第三磨牙根管可见度的增龄性变化,探讨应用其判断个体是否年满18周岁的可行性。方法共收集1280例华东地区15~30周岁人群的口腔全景片,应用OLZE等提出的方法(0~34个阶段)评估所有口腔全景片中下颌第三磨牙根管可见度,对各阶段的样本年龄分布进行描述性统计分析。结果男性首次出现阶段0、1、2、3的年龄分别为16.88、19.18、21.91、25.44岁,女性分别为17.47、20.91、22.01、26.01岁。所有样本中,阶段1~3的个体年龄均超过18周岁。结论基于口腔全景片中下颌第三磨牙根管可见度判断华东地区个体是否年满18周岁的方法具有一定的可行性。

基于CBCT中牙髓和牙体体积的机器学习用于青少年儿童年龄推断

目的利用锥形束计算机体层成像(cone beam computed tomography,CBCT)影像中左上颌中切牙与左上颌尖牙的牙髓体积和牙体体积,采用逐步回归法和机器学习方法分别推断青少年儿童年龄,并对推断效果进行比较分析。方法收集498例上海市汉族青少年儿童口腔颌面CBCT影像,测量左上颌中切牙与尖牙的牙髓体积和牙体体积并加以运算,运用K-最近邻、岭回归和决策树3种机器学习算法以及逐步回归法建立4个年龄推断模型,计算并比较决定系数、平均误差、均方根误差、均方误差和平均绝对误差等指标。绘制相关性热图,对参数间的单调关系进行可视化分析。结果K-最近邻模型(R^(2)=0.779)和岭回归模型(R^(2)=0.729)相对于逐步回归法(R^(2)=0.617)表现更为优越,而决策树模型(R^(2)=0.494)的拟合效果较差。相关性热图显示,年龄和牙髓体积、牙髓与牙体硬组织的体积比以及牙髓与牙体的体积比之间呈单调负相关。结论牙髓体积及牙髓体积占比与年龄之间存在密切关系,采用基于CBCT的机器学习方法能够提供更为准确的年龄推断结果,为进一步开展基于CBCT的深度学习牙龄推断研究奠定基础。

医学统计和机器学习方法在活体年龄推断中的应用

活体年龄推断研究中通常需要对大量的数据进行数理统计分析,合理的医学统计方法在数据整理和分析中发挥着重要作用,选择准确、恰当的统计方法是影响研究结果质量的关键因素之一。本文综述了活体年龄推断研究中描述性统计、差异性分析、一致性检验、多元统计分析等较为常用的医学统计方法以及浅层学习、深度学习等机器学习方法的原理和适用原则,并概括介绍了医学统计方法和机器学习方法之间的关联性和应用前景,旨在为活体年龄推断研究获得更为科学、精准的结果提供技术指引。

Estimation of the age of human semen stains by attenuated total reflection Fourier transform infrared spectroscopy:a preliminary study

Semen stain is one of the most important biological evidence at sexual crime scenes.Age estimation of human semen stains plays an important role in forensic work,and it is rarely studied due to lack of well-established methods.In this study,the technique called attenuated total reflection Fourier transform infrared spectroscopy(ATR-FTIR)coupled with advanced chemometric methods was employed to determine the age of semen stains on three different substrates:glass slides,tissues and fabric made of regenerated cellulose fibres up to 6 d.Partial least squares regression(PLSR)was used in conjunction with spectral analysis for age estimation,and the results generated high R2 values(cross-validation:0.81,external validation:0.74)but a narrow margin of error for root mean square error(RMSE)(RMSE of cross-validation:0.77 d,RMSE of prediction:1.02 d).Additionally,our results indicated the robustness of PLSR model was not weaken by the influence of different substrates in this study.Our results indicate that ATR-FTIR,combined with chemometric methods,shows great potential as a convenient and efficient tool for age estimation of semen stains.Moreover,the method could be applied to routine forensic investigations in the future.

Stature Estimation Using Hand and Foot Metrics among the Meena Tribal Population of the North-western Indian Region

Background:Studies on stature estimation models for North West Indian populations are limited.The present study attempts to develop stature estimation models for the Indian population using hand and foot dimensions.Materials and Methods:Hand and foot measurements of 280 individuals(140 males and 140 females)were recorded for the estimation of stature in the present study.The stature and nine hand and foot anthropometric measurements were obtained using standard methods and relevant landmarks.For bilateral and sex differences,independent and paired t-tests were used to statistically analyze the data.Results:The findings showed statistically significant sex differences(P<0.05)in stature,foot,and hand measurements.However,there were no significant bilateral differences.The Pearson correlation coefficient assessed the association between stature and hand/foot measurements.All the studied parameters in both sexes demonstrated a statistically significant(P<0.001)and strong positive correlation with stature.Males and females had the strongest correlation coefficients for right-foot length(r=0.70)and right-hand length(r=0.64),respectively.Each sex’s regression equations were calculated separately.Stature predictions were accurate to within 4.02-6.64 cm.Conclusion:Stature estimated from hand and foot measurements can help in personal identification by creating an individual’s biological profile.Morphological differences in feet and hands demonstrate the significance of developing ethnic group-specific models to determine stature.According to the study’s findings,the stature of a person belonging to the Indian subcontinent can be estimated using hands and feet for anthropological and forensic purposes.

CT三维重建技术结合深度学习算法推断成人坐骨年龄

目的探索适用于中国西部汉族人群的CT三维重建图像年龄自动推断深度学习模型,评估其可行性与可靠性。方法收集20.0~80.0岁中国西部汉族人群骨盆CT回顾性影像学数据1200例(男性600例,女性600例),重建为三维虚拟骨骼模型,区分性别、左右截取坐骨结节特征区域图像建立样本库。使用ResNet34模型,随机抽取不同性别各500例样本作为训练及验证集,剩余样本作为测试集,使用初始学习及迁移学习对区分性别、左右的图像进行训练,以平均绝对误差(mean absolute error,MAE)、均方根误差(root mean square error,RMSE)作为主要指标评价模型。结果不同性别组成中预测结果存在差异,双侧模型预测结果优于左、右单侧模型,迁移模型预测结果优于初始模型。不同性别组成的双侧迁移模型预测结果中,男性MAE为7.74岁、RMSE为9.73岁,女性MAE为6.27岁、RMSE为7.82岁,混合性别MAE为6.64岁,RMSE为8.43岁。结论基于中国西部汉族人群坐骨结节图像应用ResNet34结合迁移学习算法构建的骨龄推断模型可以有效推断成人坐骨骨龄。

Screening criteria of mRNA indicators for wound age estimation

Wound age estimation is a crucial and challenging problem in forensic pathology.Although mRNA is the most commonly used indicator for wound age estimation,screening criteria are lacking.In the present study,the feasibility of screening criteria using mRNA to determine injury time based on the adenylate-uridylate-rich element(ARE)structure and gene ontology(GO)categories were evaluated.A total of 78 Sprague-Dawley male rats were contused and sampled at 4,8,12,16,20,24,28,32,36,40,44,and 48 h after inflicting injury.The candidate mRNAs were classified based on with or without ARE structure and GO category function.The mRNA expression levels were detected using qRT-PCR.In addition,the standard deviation(STD),mean deviation(MD),relative average deviation(d%),and coefficient of variation(CV)were calculated based on mRNA expression levels.The CV score(CVs)and the CV of CV(CV’CV)were calculated to measure heterogeneity.Finally,based on classic principles,the accuracy of combination of candidate mRNAs was assessed using discriminant analysis to construct a multivariate model for inferring wound age.The results of homogeneity evaluation of each group based on CVs were consistent with the MD,STD,d%,and CV results,indicating the credibility of the evaluation results based on CVs.The candidate mRNAs without ARE structure and classified as cellular component(CC)GO category(ARE-CC)had the highest CVs,showing the mRNAs with these characteristics are the most homogenous mRNAs and best suited for wound age estimation.The highest accuracy was 91.0%when the mRNAs without ARE structure were used to infer the wound age based on the discrimination model.The accuracy of mRNAs classified into CC or multiple function(MF)GO category was higher than mRNAs in the biological process(BP)category.In all subgroups,the accuracy of the composite identification model of mRNA composition without ARE structure and classified as CC was higher than other subgroups.The mRNAs without ARE structure and belonging to the CC GO category were more homogenous,showed higher accuracy for estimating wound age,and were appropriate for rat skeletal muscle wound age estimation.

Comparing the accuracy of Demirjian and Nolla methods and establishing a new method for dental age estimation in northeastern Chinese children

Dental age estimation plays an important role in the field of clinic medicine and forensic medicine.The Demirjian and Nolla methods are common scoring methods for dental age estimation but there was no research about the comparison of accuracy of these two methods in northeastern Chinese children.Hence,in this study,we compared the accuracy of these two methods to explore more suitable method for our studied population.We collected 535 orthopantomograms from northern Chinese children aged from 6 to 15 years and divided them into training dataset and testing dataset according to the ratio of 7:3.The dental age of training dataset were estimated using Demirjian and Nolla methods,respectively.The results suggested that the mean differences of these two methods were 0.24 and−0.40 years,and mean absolute difference were 0.65 and 0.59 years.Then to further improve the accuracy of dental age assessment,the new improved formulas and dental age conversion tables were established after analyzing the relationship between the sum scores based on Nolla method and chronology age in training dataset.According to the new method used in testing dataset,the minimum value of mean difference(0.00)and mean absolute difference(0.49)were obtained,which are largely smaller than that of Demirjian and Nolla methods.The new developed method and dental age conversion scales may be more suitable dental age estimation method for northeastern Chinese children.

Age estimation of individuals aged 5-23 years based on dental development of the Indonesian population

Dental development can be used to estimate age for forensic purposes.However,most of the currently available methods are less reliable for the Indonesian population due to population variability.This study presents a new method and evaluates other methods that utilize dental development to estimate the age of Indonesian people.Panoramic radiographs of 304 young Indonesian people aged 5–23 years old were analysed for deciduous tooth root resorption,permanent tooth calcification,and eruption.The extent of tooth root resorption was determined based on AlQahtani’s modified Moorrees et al.method.Tooth calcification was classified based on a modified Demirjian et al.method.Tooth eruption was evaluated based on AlQahtani’s modified Bengston system.The sequence of tooth root resorption,and permanent tooth calcification and eruption were grouped into 19 age categories(from 5–23 years old)in an atlas.The differences between males and females,between maxillary and mandibular teeth,and between right and left teeth were also analysed.There were minimal significant differences of tooth development between males and females,and between the right and left teeth(P>0.05),while the maxillary and mandibular dental development was significantly different(P<0.05).The newly developed atlas showed the development of the right side of maxillary and mandibular tooth of combined sex of Indonesian population.Another 34 panoramic radiographs of known-age and sex individuals from Indonesia were assessed using the newly developed Atlas of Dental Development in the Indonesian Population,Ubelaker’s Dental Development Chart,The London Atlas of Human Tooth Development and Eruption by AlQahtani,and the Age Estimation Guide-Modern Australia population by Blenkin-Taylor.Accuracy was assessed by comparing estimated age to actual chronological age using the Bland-Altmand test.Results show that the smallest range of error was found in the Atlas of Dental Development in the Indonesian Population(−0.969 to 1.210 years),followed by The London Atlas of Human Tooth Development and Eruption by AlQahtani(−2.013 to 1.990 years),the Age Estimation Guide-Modern Australia population by Blenkin-Taylor(−2.495 to 2.598 years),and the Dental Development Chart by Ubelaker(−2.960 to 3.289 years).These findings show that the Atlas of Dental Development constructed in this study performs better than the other three methods and presents greater accuracy of age estimation in the Indonesian population.

Can canines alone be used for age estimation in Chinese individuals when applying the Kvaal method?

Due to the secondary dentin formation,the dental pulp undergoes changes in shape throughout life.Based on this phenomenon,the Kvaal method has been applied to various populations for age estimation,and its usefulness has been verified.When applying the Kvaal method to Chinese subjects,we observed a relatively strong correlation between mandibular canines and age.This study notes the correlation between canines and chronological age and is the first to identify which canine is most closely related to chronological age.In addition,a new,simpler formula is determined based on canines according to Kvaal’s methodology.The radiographs of 360 individuals from northern China were selected,from which the widths and lengths of the pulp from four canines were measured according to the Kvaal method.Next,inter-and intra-observer reliabilities were analyzed in order to assess the repeatability of these measurements.The correlation between measurements and age was examined,and Chinese-specific age estimation formulae were derived.The results revealed that the ratios from the left maxillary canine exhibited the strongest correlation with age compared to the other canines,whereas the left mandibular canine showed the weakest correlation,which may contribute to the overall poor correlation of mandibular canines with age.What’s more,the formula derived from the left maxillary canine in this study displayed the highest coefficients of determination,and the formula derived from all canines showed the lowest residuals.Both of these formulae performed better than the Chinese-specific formula derived from six different types of teeth in our previous study,which had formerly possessed the highest coefficients of determination and the lowest residuals.Thus,we concluded that canines do play an important role in age estimation in the Chinese population,and the correlation between maxillary canines and chronological age is stronger than that of mandibular canines,although no distinct trend as to which side is better correlated with age was established.Going forward,we recommend the analysis of additional samples from different geographical regions and populations to further verify the importance of canines in age estimation.