| Objective:To explore the application of Fourier transform infrared spectroscopy(FTIR)combined with machine learning methods in forensic pathology on diagnosis of death from fetal hyperthermia and postmortem interval inference.Methods:(?)Differential diagnosis of death from hyperthermia:42 male New Zealand white rabbits,weighting 2.2~2.5kg,were randomly divided into the heatstroke death group(21 rabbits)and the control group(a total of 21 rabbits,including 7 drowned,7mechanically suffocated and 7 hemorrhagic shock).FTIR spectral data of vitreous body fluid were collected and pretreated,and the discrimination model of heatstroke death was established by using the pretreated spectral data.(?)Postmortem interval estimation:30 healthy male SD rats with a mass of 240~260g were placed in 4?,20?and 30?environment(10 rats each group)after cervical dislocation executed.Rat liver tissues were taken at different time points,FTIR spectral data of liver samples were collected,and the spectral data were pretreated and a time of death inference model was established.Results:(?)Differential diagnosis of death from hyperthermia:(1)The mean spectra of vitreous humor of the rabbits with different causes of death were significantly different,and the results of t-test and PCA showed that these variations in spectra of hyperthermia and control groups were mainly related to the different structures or components of proteins,amino acids and carbohydrates in vitreous humor;(2)Multiple algorithms were used to establish the discriminant model of heatstroke death.Among them,genetic algorithm combined with partial least squares discriminant analysis(GA-PLS-DA)was used to establish a simple discriminant model(57variables were selected)with high performance,and the sensitivity,specificity and accuracy of the model reached 100,the rears under the ROC curves(AUCs)of the receiver operating characteristic curve(ROC)are 1.(?)Postmortem interval estimation:(1)The mean spectral absorption peak of liver tissues of rats at different time points of death changed significantly;(2)PCA(principle component analysis,PCA)showed that in 20?and 30?environment,the spectra at different death time points showed an obvious clustering trend,and the clustering trend of these samples were mainly related to the composition or structure differences of proteins,nucleic acids,carbohydrates and other substances in the tissues,while the distribution of the samples in 4?group was scattered,and no obvious distribution rules were found;(3)The spectral data in 4?,20?and 30?temperature was used to establish partial least squares(PLS)regression models to deduce the time of death and the external validation determination coefficient(R~2)of the three models were 0.55,0.82 and 0.96 respectively,the root mean square error of prediction(RMSEP)were 38.5 h,24.2 h and 4.6 h respectively;(4)The FTIR spectral data of 4?and 20?groups of rat liver tissues were divided into three time periods of 0~24 h,48~96 h and 120~168 h,then partial least squares discriminant analysis(PLS-DA)models were established to discriminate the time range of death.The external validation accuracy of the PLS-DA models in 4?and 20?groups were 69%and 86%respectively.Conclusion:(1)FTIR spectroscopy has sufficient ability to reflect the information of almost all the molecular components of the tested biological samples;(2)FTIR spectroscopy combined with machine learning algorithm was used to establish the differential diagnosis model of hyperthermia,which created a new research idea and method for forensic identification of complex causes of death;(3)FTIR spectroscopy combined with machine learning algorithm can be used to analyze rat liver tissue qualitatively and quantitatively after death,which is expected to be a simple and rapid technology for inferring the time of death. |
PDF Full Text Request