New Opportunities and Challenges For Forensic Medicine in The Era of Artificial Intelligence Technology
New Opportunities and Challenges For Forensic Medicine in The Era of Artificial Intelligence Technology
New Opportunities and Challenges For Forensic Medicine in The Era of Artificial Intelligence Technology
·综 述·
人工智能技术时代法医学科面临的新机遇与挑战
方雅婷 1,兰琼 1,解通 1,刘艳芳 1,梅书燕 1,朱波峰 1,2
(1. 南方医科大学法医学院,广东 广州 510515; 2. 西安交通大学口腔医学院 陕西省颅颌面精准医学研
究中心,陕西 西安 710049)
摘 要:传统的法医学鉴定依赖于法医专家提取信息,运用医学、生物学等多领域知识结合工作经验作出
鉴定意见,不仅耗时耗力,而且存在难以克服的主观因素。在大数据时代,人工智能的蓬勃发展为法医学
带来了新思路。近些年,国内外法医学者们基于人工智能技术开展了人脸识别、年龄及性别鉴定、DNA 分
析、死亡时间推断、损伤以及死亡原因鉴定等多方面的研究,显示出运用人工智能技术解决法医学鉴定问
题的可行性与优越性。人工智能技术作为一种适应时代发展的崭新技术手段,为法医学带来新活力的同
时,也带来了新挑战。如何科学应对这些挑战,形成人工智能和法医学协同发展的“人工智能+法医学”新
模式,是大数据时代下法医学发展的新方向。
关键词:法医学;人工智能;机器学习;深度学习;综述
中图分类号:DF795.1 文献标志码:A doi:10.12116/j.issn.1004-5619.2020.01.016
文章编号:1004-5619(2020)01-0077-09
New Opportunities and Challenges for Forensic Medicine in the Era of Artificial
Intelligence Technology
FANG Ya-ting1, LAN Qiong1, XIE Tong1, LIU Yan-fang1, MEI Shu-yan1, ZHU Bo-feng1,2
(1. School of Forensic Medicine, Southern Medical University, Guangzhou 510515, China; 2. Research Center
of Shaanxi Province for Craniofacial Precision Medicine, College of Stomatology, Xi’an Jiaotong University,
Shaanxi 710049, China)
Abstract:Traditional forensic identification relies on forensic experts to manually extract information and
provide identification opinions based on medicine, biology and other fields of knowledge combined with
personal work experience, which is not only time-consuming and require great effort, but also affected
by subjective factors that are difficult to overcome. In the era of big data, the booming development
of artificial intelligence brings new ideas to forensic medicine. In recent years, forensic researchers at
home and abroad have conducted many studies based on artificial intelligence technology, such as face
recognition, age and gender identification, DNA analysis, postmortem interval estimation, injury and
cause of death identification, showing the feasibility and advantages of using artificial intelligence tech-
nology to solve forensic identification problems. As a new means of technology that has adapted to
the development of the times, artificial intelligence has brought new vitality to forensic medicine, but
at the same time also some new challenges. How to deal with these challenges scientifically and form
a new mode of ‘artificial intelligence plus forensic medicine’ with artificial intelligence and forensic
medicine developing collaboratively is a new direction for the development of forensic medicine in the
era of big data.
Keywords:forensic medicine; artificial intelligence; machine learning; deep learning; review
人工智能技术是一门多学科间广泛交叉的前沿 智能技术飞速发展,并被广泛应用于医学研究等各个
科学,自 1956 年的达特茅斯会议后,其概念和研究领 领域[2-4]。近年来快速发展的法医学同样面临着获取、
[1]
域不断扩展 。随着人工神经网络理论的提出,人工 分析和应用大数据,以应对复杂鉴定问题的挑战[5],很
基金项目:国家自然科学基金资助项目(81525015)
作者简介:方雅婷(1992—),女,博士研究生,主要从事法医物证学研究;E-mail:
[email protected]
通信作者:朱波峰,男,教授,博士,主要从事法医学教学及法医物证学科研;E-mail: [email protected]
· 78 · Journal of Forensic Medicine, February 2020, Vol.36, No.1
多传统的学习算法已难以适应发展的新需求[6]。如何 骨遗骸的年龄预测有助于推断其死亡时的年龄。传
将人工智能技术融会贯通到法医学各个领域,消除人 统的骨龄鉴定方法是以人工测定的方法对骨骺生长
工分析时产生的主观偏差,形成“人工智能 + 法医学” 进行分级,不仅在骨骺分类上存在一定的技术困难,
的新模式,是当前解决疑难复杂案件、提高实际检案 而且无法消除主观性偏差。为克服这些难题,WANG
工作的效率与科研能力的新策略。本文对 2015 年以 等 [12] 采用支持向量机对中国青少年的腕关节 X 线图
来人工智能技术在法医学中最新的研究进展及应用 像进行训练,构建年龄预测模型,实现了对桡骨远端
前景进行了系统梳理、总结与展望。 和尺骨远端骨骺的自动分类。 ZHANG 等 [13] 考虑到性
别差异,运用体绘制技术提取 280 名女性和 282 名男
1 人工智能在法医学中的运用
性第 5 颈椎至第 1 腰椎区域的 CT 图像信息后,分别采
1.1 人脸识别 用简单线性回归、多元线性回归、梯度增强回归、支持
人的面部容貌虽然在生长发育中会发生变化,但 向量机和决策树等算法构建了男性特异性和女性特
是个人特征是相对固定的,因此是同一认定和个体识 异性年龄预测模型。 SPAMPINATO 等 [14] 用公共数据
别 理 想 的 生 物 学 特 征 。 PENG 等 [7] 在 马 尔 可 夫 网 络 集 测 试 并 评 估 了 OverFeat 、GoogLeNet 和 OxfordNet 三
(Markov networks)的基础上,
自适应地学习组合图像的 种卷积神经网络模型在骨龄自动评估中的应用,并基
多个表征,并采用最小误差边界切割(minimum error 于处理后的 X 线图像,建立了全新的用于骨龄推断的
boundary cut)算法对重叠区域进行拼接,建立基于多 BoNet 卷积神经网络模型。 LEE 等 [15] 基于手部及腕关
重表征的自适应人脸素描合成方法,实现了高效的人 节 X 线图像,创建了一个可以自动检测并分类的卷积
脸识别。 SAJID 等 测量并评估了面部不对称特征的
[8]
神经网络模型,能够在放射科医生的指导下自动生成
尺寸及其随年龄变化的特点,发现非对称性面部特征 结构化报告,实现了精准的骨龄评估。 NAVEGA 等 [16]
是预测年龄的一个强有力指标,基于支持向量机算法 以欧洲女性的股骨为研究对象,构建了基于骨密度测
针对量化的面部不对称特征估计给定人脸图像的年 量法的人工神经网络模型用于骨龄评估。
龄组,并采用卷积神经网络提取非对称的面部特征, 除骨骼外,牙齿的发育情况也是年龄预测的重要
通过分析人脸图像年龄组估计误差对人脸识别算法 指标之一。 de TOBEL 等 [17] 先对口腔全景 X 线图像对
的影响,发现将从年龄组估计中提取到的信息嵌入人 比度设置进行优化,评估了下颌第三磨牙的特征,并
脸识别算法中,可以大大提高人脸识别的准确性。 使用 Adobe Photoshop CC 2017 软件以标准化的方式
下颌骨是颌面部重要的组成结构,因其仅以软组 在其周围放置矩形边框,随后利用 MATLAB R2017a
织连接于上颌部,所以在尸体白骨化中易遗失。为实 软件中的机器学习算法对下颌第三磨牙的特征进行
现 丢 失 了 下 颌 骨 的 颌 面 部 重 建 ,NINO-SANDOVAL 自动识别,研发了一种基于下颌第三磨牙的年龄预测
等 在 2016 年采用支持向量机监督学习算法选取哥伦
[9]
自 动 检 测 技 术 。 STEPANOVSKY 等 [18] 比 较 了 从 简 单
比亚人头颅侧位 X 线图像中 10 个颅颌关节变量对下颌 数 学 模 型 到 复 杂 机 器 学 习 算 法 的 22 种 年 龄 预 测 模
骨相对于上颌骨的矢状位置关系进行分类,发现骨性 型,证明基于机器学习算法的数据挖掘方法优于基于
Ⅱ 类和骨性 Ⅲ 类的颅颌面结构有不依赖于下颌骨的 平均年龄的传统数学模型方法,并提出了用具有代表
独特骨性关系。 2017 年他们采用人工神经网络等深 性的中位年龄代替牙齿发育阶段的方法与缺失牙病
度学习算法,基于同样的数据集,选取了 17 个颅颌面 例的缺失数据替换技术,构建了基于列表多元线性回
形态的解剖相关变量预测上、下颌骨的位置关系,发 归模型(tabular multiple linear regression model)的年
[10]
现人工神经网络的分类预测能力优于支持向量机 。 龄预测模型。
另一方面,随着人脸识别的相关研究逐步报道, 当缺乏完整的骨骼证据时,用分子生物学方法分
如何有效地收集和管理与人脸有关的数据成了另一 析案发现场遗留的血液、组织等生物学物证所包含的
项 挑 战 。 KAUR 等 采 用 基 于 文 本 挖 掘 的 人 工 智 能
[11]
信息,是法医学年龄推断的另一重要手段。继多元回
技术对文献自动注释,尝试开发一个可以提取面部形 归、反向传播神经网络、支持向量机以及随机森林模
态学特征以及面部疾病相关信息的人脸数据库。虽 型 被 应 用 于 基 于 DNA 甲 基 化 的 年 龄 预 测 模 型 构 建
然这项工作目前仍处于初始研究阶段,但其为人工智 后 [19-20],VIDAKI 等 [21] 选出 16 个 CpG 位点,首次应用人
能在人脸识别中的应用开辟了新思路与方向。 工神经网络构建年龄预测模型,不仅提高了预测的精
1.2 年龄推断 准 度 ,还 排 除 了 未 成 年 以 及 不 同 种 族 背 景 的 影 响 。
骨骼的发育过程具有连续性和阶段性特征,对尸 FENG 等 [22] 使用逐步向后的多元线性回归分析和穷举
法 医 学 杂 志 2020 年 2月 第 36 卷 第1期 · 79 ·
chine learning models[J]. Forensic Sci Int Genet, [35] MARCIANO M A, WILLIAMSON V R, ADEL-
2018,37:215-226. doi:10.1016/j.fsigen.2018.09.003. MAN J D. A hybrid approach to increase the in-
[24] SPOLNICKA M, POSPIECH E, PEPLONSKA B, formedness of CE- based data using locus- specific
et al. DNA methylation in ELOVL2 and C1orf132 thresholding and machine learning[J]. Forensic Sci
correctly predicted chronological age of individuals Int Genet,2018,35:26-37. doi:10.1016/j.fsigen.2018.
from three disease groups[J]. Int J Legal Med,2018, 03.017.
132(1):1-11. doi:10.1007/s00414-017-1636-0. [36] HANED H,PENE L,LOBRY J R,et al. Estima-
[25] KRISHAN K,CHATTERJEE P M,KANCHAN T, ting the number of contributors to forensic DNA
et al. A review of sex estimation techniques during mixtures: Does maximum likelihood perform better
examination of skeletal remains in forensic anthro- than maximum allele count?[J]. J Forensic Sci,2011,
pology casework[J]. Forensic Sci Int,2016,261:161- 56(1):23-28. doi:10.1111/j.1556-4029.2010.01550.x.
165. doi:10.1016/j.forsciint.2016.02.007. [37] HANED H,PENE L,SAUVAGE F,et al. The pre-
[26] MUSILOVA B,DUPEJ J,VELEMINSKA J, et al. dictive value of the maximum likelihood estimator of
Exocranial surfaces for sex assessment of the hu- the number of contributors to a DNA mixture[J].
man cranium[J]. Forensic Sci Int,2016,269:70- 77. Forensic Sci Int Genet,2011,5(4):281-284. doi:10.
doi:10.1016/j.forsciint.2016.11.006. 1016/j.fsigen.2010.04.005.
[27] BEJDOVA S,DUPEJ J,KRAJICEK V,et al. Sta- [38] TAYLOR D,BRIGHT J A,BUCKLETON J. Inter-
bility of upper face sexual dimorphism in central preting forensic DNA profiling evidence without spe-
European populations (Czech Republic) during the cifying the number of contributors[J]. Forensic Sci
modern age[J]. Int J Legal Med,2018,132(1):321- Int Genet,2014,13:269- 280. doi:10.1016/j.fsigen.
330. doi:10.1007/s00414-017-1625-3. 2014.08.014.
[28] GAO H,GENG G,YANG W. Sex determination of [39] SWAMINATHAN H, GRGICAK C M, MEDARD
3D skull based on a novel unsupervised learning M, et al. NOCIt:A computational method to infer
method[J]. Comput Math Methods Med,2018,2018: the number of contributors to DNA samples ana-
4567267. doi:10.1155/2018/4567267. lyzed by STR genotyping[J]. Forensic Sci Int Genet,
[29] CAVALLI F, LUSNIG L, TRENTIN E. Use of 2015,16:172-180. doi:10.1016/j.fsigen.2014.11.010.
pattern recognition and neural networks for non- [40] MARCIANO M A,ADELMAN J D. PACE:Proba-
metric sex diagnosis from lateral shape of calvarium: bilistic assessment for contributor estimation- a ma-
An innovative model for computer- aided diagnosis chine learning- based assessment of the number of
in forensic and physical anthropology[J]. Int J Le- contributors in DNA mixtures[J]. Forensic Sci Int
gal Med,2017,131(3):823-833. doi:10.1007/s00414- Genet,2017,27:82- 91. doi:10.1016/j.fsigen.2016.11.
016-1439-8. 006.
[30] TOSI L L, BOYAN B D, BOSKEY A L. Does [41] MONSALVE T,HEFNER J T. Macromorphoscopic
sex matter in musculoskeletal health? A workshop trait expression in a cranial sample from Medellin,
report[J]. Orthop Clin North Am,2006,37(4):523- Colombia[J]. Forensic Sci Int,2016,266:571- 574.
529. doi:10.1016/j.ocl.2006.09.011. doi:10.1016/j.forsciint.2016.07.014.
[31] KOTEROVA A,VELEMINSKA J,DUPEJ J,et al. [42] POSPIECH E, KARLOWSKA- PIK J, ZIEMKIE-
Disregarding population specificity:Its influence on WICZ B, et al. Further evidence for population
the sex assessment methods from the tibia[J]. Int specific differences in the effect of DNA markers
J Legal Med,2017,131(1):251- 261. doi:10.1007/ and gender on eye colour prediction in forensics[J].
s00414-016-1413-5. Int J Legal Med,2016,130(4):923-934. doi:10.1007/
[32] TAYLOR D,POWERS D. Teaching artificial intel- s00414-016-1388-2.
ligence to read electropherograms[J]. Forensic Sci Int [43] WALSH S, LIU F, BALLANTYNE K N, et al.
Genet,2016,25:10- 18. doi:10.1016/j.fsigen.2016.07. IrisPlex:A sensitive DNA tool for accurate predic-
013. tion of blue and brown eye colour in the absence
[33] TAYLOR D,HARRISON A,POWERS D. An arti- of ancestry information[J]. Forensic Sci Int Genet,
ficial neural network system to identify alleles in 2011, 5(3): 170-180. doi:10.1016/j.fsigen.2010.02.004.
reference electropherograms[J]. Forensic Sci Int Ge- [44] JUHÁSZ Z,DUDÁS E,PAMJAV H. A new self-
net,2017,30:114- 126. doi:10.1016/j.fsigen.2017.07. learning computational method for footprints of ear-
002. ly human migration processes[J]. Mol Genet Ge-
[34] TAYLOR D,KITSELAAR M,POWERS D. The ge- nomics, 2018, 293(6): 1579-1594. doi: 10.1007/s00438-
neralisability of artificial neural networks used to 018-1469-7.
classify electrophoretic data produced under different [45] JUHÁSZ Z, DUDÁS E,VÁGÓ-ZALÁN A, et al.
conditions[J]. Forensic Sci Int Genet,2019,38:181- A simultaneous search for footprints of early human
184. doi:10.1016/j.fsigen.2018.10.019. migration processes using the genetic and folk mu-
· 84 · Journal of Forensic Medicine, February 2020, Vol.36, No.1
sic data in Eurasia[J]. Mol Genet Genomics,2019, pathological observation[J]. Am J Forensic Med Pa-
294(4):941-962. doi:10.1007/s00438-019-01539-x. thol,2016,37(4):279-283. doi:10.1097/PAF.0000000
[46] 王琪,林汉成,徐纪茹,等 . 死亡时间推断最新研究与 000000275.
展望[J]. 法医学杂志,2018,34(5):459-467. doi:10. [57] WEI F,BUCAK S S,VOLLNER J M,et al. Clas-
12116/j.issn.1004-5619.2018.05.002. sification of porcine cranial fracture patterns using
WANG Q,LIN H,XU J,et al. Current research a fracture printing interface[J]. J Forensic Sci,2017,
and prospects on postmortem interval estimation[J]. 62(1):30-38. doi:10.1111/1556-4029.13231.
Fa Yi Xue Za Zhi,2018,34(5):459-467. [58] YILMAZ R,ERKAYMAZ O,KARA E,et al. Use
[47] LI Z, HUANG J,WANG Z, et al. An investiga- of autopsy to determine live or stillbirth:New ap-
tion on annular cartilage samples for post- mortem proaches in decision- support systems[J]. J Forensic
interval estimation using Fourier transform infrared Sci,2017,62(2):468- 472. doi:10.1111/1556- 4029.
spectroscopy[J]. Forensic Sci Med Pathol,2019:1- 7. 13277.
doi:10.1007/s12024-019-00146-x. [59] MUJTABA G,SHUIB L,RAJ R G,et al. Predic-
[48] ZHANG J,WEI X, HUANG J, et al. Attenuated tion of cause of death from forensic autopsy reports
total reflectance Fourier transform infrared (ATR- using text classification techniques: A comparative
FTIR) spectral prediction of postmortem interval study[J]. J Forensic Leg Med,2018,57:41-50. doi:
from vitreous humor samples[J]. Anal Bioanal Chem, 10.1016/j.jflm.2017.07.001.
2018,410(29):7611-7620. doi:10.1007/s00216-018- [60] MUJTABA G,SHUIB L,RAJ R G,et al. Classi-
1367-1. fication of forensic autopsy reports through concep-
[49] COBAUGH K L,SCHAEFFER S M,DEBRUYN tual graph- based document representation model[J].
J M. Functional and structural succession of soil J Biomed Inform,2018,82:88-105. doi:10.1016/j.jbi.
microbial communities below decomposing human 2018.04.013.
cadavers[J]. PLoS One,2015,10(6):e130201. doi:10. [61] EBERT L C,PTACEK W,NAETHER S,et al. Vir-
1371/journal.pone.0130201. tobot--a multi-functional robotic system for 3D sur-
[50] FINLEY S J,PECHAL J L,BENBOW M E,et al. face scanning and automatic post mortem biopsy[J].
Microbial signatures of cadaver gravesoil during de- Int J Med Robot,2010,6(1):18-27. doi:10.1002/rcs.
composition[J]. Microb Ecol,2016,71(3):524- 529. 285.
doi:10.1007/s00248-015-0725-1. [62] EBERT L C,PTACEK W, BREITBECK R, et al.
[51] HAUTHER K A,COBAUGH K L, JANTZ L M, Virtobot 2.0:The future of automated surface doc-
et al. Estimating time since death from postmortem umentation and CT- guided needle placement in fo-
human gut microbial communities[J]. J Forensic Sci, rensic medicine[J]. Forensic Sci Med Pathol,2014,
2015,60(5):1234-1240. doi:10.1111/1556-4029.12828. 10(2):179-186. doi:10.1007/s12024-013-9520-9.
[52] METCALF J L,XU Z Z,WEISS S,et al. Microbial [63] EBERT L C,FURST M,PTACEK W,et al. Au-
community assembly and metabolic function during tomatic entry point planning for robotic post- mor-
mammalian corpse decomposition[J]. Science,2016, tem CT-based needle placement[J]. Forensic Sci Med
351(6269):158-162. doi:10.1126/science.aad2646. Pathol,2016,12(3):336- 342. doi:10.1007/s12024-
[53] JOHNSON H R,TRINIDAD D D,GUZMAN S, 016-9798-5.
et al. A machine learning approach for using the [64] EBERT L C,HEIMER J,SCHWEITZER W,et al.
postmortem skin microbiome to estimate the post- Automatic detection of hemorrhagic pericardial effu-
mortem interval[J]. PLoS One,2016,11(12):e167370. sion on PMCT using deep learning - a feasibility
doi:10.1371/journal.pone.0167370. study[J]. Forensic Sci Med Pathol,2017,13(4):426-
[54] MOORE H E, BUTCHER J B,ADAM C D, et 431. doi:10.1007/s12024-017-9906-1.
al. Age estimation of Calliphora(Diptera:Calliphori- [65] ELIAERTS J, MEERT N,Van DURME F, et al.
dae) larvae using cuticular hydrocarbon analysis and Practical tool for sampling and fast analysis of large
Artificial Neural Networks[J]. Forensic Sci Int,2016, cocaine seizures[J]. Drug Test Anal,2018,10(6):
268:81-91. doi:10.1016/j.forsciint.2016.09.012. 1039-1042. doi:10.1002/dta.2364.
[55] MOORE H E,BUTCHER J B,DAY C R,et al. [66] MOLLERUP C B,MARDAL M, DALSGAARD P
Adult fly age estimations using cuticular hydrocar- W, et al. Prediction of collision cross section and
bons and Artificial Neural Networks in forensically retention time for broad scope screening in gradient
important Calliphoridae species[J]. Forensic Sci Int, reversed- phase liquid chromatography- ion mobility-
2017,280:233- 244. doi:10.1016/j.forsciint.2017.10. high resolution accurate mass spectrometry[J]. J Chro-
001. matogr A,2018,1542:82- 88. doi:10.1016/j.chroma.
[56] REN G,ZOU D,HUANG P,et al. Identifying dif- 2018.02.025.
fuse axonal injury by matrix- assisted laser desorp- [67] WENDT F R,NOVROSKI N,RAHIKAINEN A L,
tion/ionization time- of- flight: A new method for et al. Supervised classification of CYP2D6 genotype
法 医 学 杂 志 2020 年 2月 第 36 卷 第1期 · 85 ·
and metabolizer phenotype with postmortem tramadol- imaging predicts psychopathic traits in adolescent
exposed finns[J]. Am J Forensic Med Pathol,2019, offenders[J]. Neuroimage,2017,145(Pt B):265-273.
40(1):8-18. doi:10.1097/PAF.0000000000000447. doi:10.1016/j.neuroimage.2015.12.013.
[68] MIGUEL- HURTADO O,GUEST R,STEVENAGE [79] 崔雍浩, 商聪, 陈锶奇, 等 . 人工智能综述: AI 的发展[J].
S V,et al. Comparing machine learning classifiers 无线电通信技术,2019,45(3):225-231.
and linear/logistic regression to explore the relation- CUI Y,SHANG C,CHEN S,et al. AI overview:
ship between hand dimensions and demographic cha- The development of AI[J]. Wu Xian Dian Tong
racteristics[J]. PLoS One,2016,11(11):e165521. doi: Xin Ji Shu,2019,45(3):225-231.
10.1371/journal.pone.0165521. [80] 胡婷鸿,万雷,刘太昂,等 . 深度学习在图像识别及骨
[69] HEINRICH A,GUTTLER F,WENDT S,et al. Fo- 龄评估中的优势及应用前景[J]. 法医学杂志,2017,
rensic odontology: Automatic identification of per- 33(6):629- 634,639. doi:10.3969/j.issn.1004- 5619.
sons comparing antemortem and postmortem panora- 2017.06.013.
mic radiographs using computer vision[J]. Rofo,2018, HU T, WAN L, LIU T, et al. Advantages and
190(12):1152-1158. doi:10.1055/a-0632-4744. application prospects of deep learning in image re-
[70] TACKMANN J, ARORA N, SCHMIDT T S B, cognition and bone age assessment[J]. Fa Yi Xue
et al. Ecologically informed microbial biomarkers Za Zhi,2017,33(6):629-634,639.
and accurate classification of mixed and unmixed [81] BEWES J,LOW A,MORPHETT A,et al. Artifi-
samples in an extensive cross- study of human body cial intelligence for sex determination of skeletal re-
sites[J]. Microbiome,2018,6(1):192. doi:10.1186/ mains:Application of a deep learning artificial neu-
s40168-018-0565-6. ral network to human skulls[J]. J Forensic Leg Med,
[71] DENG M,WANG C,TANG M,et al. Extracting 2019,62:40-43. doi:10.1016/j.jflm.2019.01.004.
cardiac dynamics within ECG signal for human iden- [82] 胡婷鸿,火忠,刘太昂,等 . 基于深度学习实现维吾尔
tification and cardiovascular diseases classification[J]. 族青少年左手腕关节骨龄自动化评估[J]. 法医学杂志,
Neural Netw,2018,100:70-83. doi:10.1016/j.neunet. 2018,34(1):27- 32. doi:10.3969/j.issn.1004- 5619.
2018.01.009. 2018.01.006.
[72] HAN I. Analysis of vehicle collision accidents based HU T,HUO Z,LIU T,et al. Automated Assess-
on qualitative mechanics[J]. Forensic Sci Int,2018, ment for Bone Age of Left Wrist Joint in Uyghur
291:53-61. doi:10.1016/j.forsciint.2018.08.004. Teenagers by Deep Learning[J]. Fa Yi Xue Za Zhi,
[73] CAN Y A,ACI C,AYDIN K. Traffic accident re- 2018,34(1):27-32.
construction and an approach for prediction of fault [83] HONG S R,SHIN K J,JUNG S E,et al. Plat-
rates using artificial neural networks:A case study form- independent models for age prediction using
in Turkey[J]. Traffic Inj Prev,2016,17(6):585-589. DNA methylation data[J]. Forensic Sci Int Genet,
2019,38:39-47. doi:10.1016/j.fsigen.2018.10.005.
doi:10.1080/15389588.2015.1122760.
[84] DAMAS S,WILKINSON C, KAHANA T, et al.
[74] DAMASO N, MENDEL J, MENDOZA M, et al.
Study on the performance of different craniofacial
Bioinformatics approach to assess the biogeographi-
superimposition approaches ( Ⅱ ): Best practices
cal patterns of soil communities:The utility for soil
proposal[J]. Forensic Sci Int,2015,257:504-508. doi:
provenance[J]. J Forensic Sci,2018,63(4):1033-1042.
10.1016/j.forsciint.2015.07.045.
doi: 10.1111/1556-4029.13741.
[85] IBANEZ O , VICENTE R , NAVEGA D , et al.
[75] WARMLANDER S,VARUL L,KOSKINEN J, et
MEPROCS framework for craniofacial superimposi-
al. Estimating the temperature of heat- exposed bone
tion:Validation study[J]. Leg Med (Tokyo),2016,
via machine learning analysis of sci color values:
23:99-108. doi:10.1016/j.legalmed.2016.10.007.
A pilot study[J]. J Forensic Sci,2019,64(1):190-
[86] IBANEZ O, VALSECCHI A, CAVALLI F, et al.
195. doi:10.1111/1556-4029.13858.
Study on the criteria for assessing skull-face corre-
[76] ZAILER E, DIEHL B W, MONAKHOVA Y B.
spondence in craniofacial superimposition[J]. Leg
Blood species discrimination using proton nuclear
Med(Tokyo), 2016, 23:59-70. doi:10.1016/j.legalmed.
magnetic resonance spectroscopy[J]. Int J Legal Med,
2016.09.009.
2017,131(3):723- 729. doi:10.1007/s00414- 016-
[87] GORDON G M,STEYN M. An investigation into
1500-7.
the accuracy and reliability of skull-photo superim-
[77] VITIELLO A,Di NUNZIO C,GAROFANO L,et
position in a South African sample[J]. Forensic Sci
al. Bloodstain pattern analysis as optimisation prob-
Int,2012,216(1/2/3):191-198. doi:10.1016/j.forsciint.
lem[J]. Forensic Sci Int,2016,266:e79-e85. doi:10.
2011.09.008.
1016/j.forsciint.2016.06.022.
[78] STEELE V R, RAO V, CALHOUN V D, et al. (收稿日期:2019-05-23)
Machine learning of structural magnetic resonance (本文编辑:黄平)