傅里叶变换显微红外光谱结合机器学习算法鉴定电击伤  

Determination of Electrocution Using Fourier Transform Infrared Microspectroscopy and Machine Learning Algorithm

作　　者：托娅 黎世莹 张吉 邓恺飞 罗仪文 孙其然 董贺文 黄平 TUO Ya;LI Shi-ying;ZHANG Ji;DENG Kai-fei;LUO Yi-wen;SUN Qi-ran;DONG He-wen;HUANG Ping(School of Basic Medical Science,Shanghai University of Medicine&Health Science,Shanghai 201318,China;Shanghai Key Laboratory of Forensic Medicine,Key Laboratory of Forensic Science,Ministry of Justice,Shanghai Forensic Service Platform,Academy of Forensic Science,Shanghai 200063,China)

机构地区：[1]上海健康医学院基础医学院,上海201318 [2]司法鉴定科学研究院,上海市法医学重点实验室,司法部司法鉴定重点实验室,上海市司法鉴定专业技术服务平台,上海200063

出　　处：《法医学杂志》2020年第1期35-40,共6页Journal of Forensic Medicine

基　　金：国家自然科学基金资助项目(81601645);上海市法医学重点实验室资助项目(17DZ2273200);上海市司法鉴定专业技术服务平台资助项目(19DZ2292700)。

摘　　要：目的基于傅里叶变换显微红外光谱结合机器学习算法分析猪皮肤电击伤、烧伤及擦伤的差异,构建3种皮肤损伤鉴定模型,筛选电击伤特征性标志物,为皮肤电流斑鉴定提供新方法。方法建立猪皮肤电击伤、烧伤及擦伤的模型,使用传统HE染色检验不同损伤的形态学改变。运用傅里叶变换显微红外技术检测表皮细胞光谱,运用主成分、偏最小二乘法分析损伤的分类情况,运用线性判别和支持向量机构建分类模型,因子载荷筛选特征性标志物。结果与对照组相比,电击伤、烧伤及擦伤组的表皮细胞均呈现出极化现象,以电击伤、烧伤组更为明显。通过主成分和偏最小二乘法分析可区分不同类型损伤,线性判别、支持向量机模型均能够有效诊断不同损伤。2 923、2 854、1 623、1 535 cm-1吸收峰在不同损伤组显示出明显的差异,电击伤的2 923 cm-1吸收峰峰强最高。结论傅里叶变换显微红外光谱结合机器学习算法为诊断皮肤电击伤、鉴定电击死提供了新技术。Objective To analyze the differences among electrical damage, burns and abrasions in pig skin using Fourier transform infrared microspectroscopy(FTIR-MSP) combined with machine learning algorithm, to construct three kinds of skin injury determination models and select characteristic markers of electric injuries, in order to provide a new method for skin electric mark identification. Methods Models of electrical damage, burns and abrasions in pig skin were established. Morphological changes of different injuries were examined using traditional HE staining. The FTIR-MSP was used to detect the epidermal cell spectrum. Principal component method and partial least squares method were used to analyze the injury classification. Linear discriminant and support vector machine were used to construct the classification model, and factor loading was used to select the characteristic markers. Results Compared with the control group, the epidermal cells of the electrical damage group, burn group and abrasion group showed polarization, which was more obvious in the electrical damage group and burn group. Different types of damage was distinguished by principal component and partial least squares method. Linear discriminant and support vector machine models could effectively diagnose different damages. The absorption peaks at 2 923 cm-1, 2 854 cm-1, 1 623 cm-1, and 1 535 cm-1 showed significant differences in different injury groups. The peak intensity of electrical injury’s 2 923 cm-1 absorption peak was the highest. Conclusion FTIR-MSP combined with machine learning algorithm provides a new technique to diagnose skin electrical damage and identification electrocution.

关 键 词：法医病理学 电击伤 谱学 傅里叶变换红外 机器学习 皮肤 猪 

分 类 号：DF795.1[医药卫生—法医学]