EVALUATION OF HSI-NIR AND CHEMOMETRICS FOR IDENTIFICATION OF SEMEN ON COLORED COTTON FABRICS

Among several problems faced by forensic scientists, body fluid identification at crime scenes is one of the most common and relevant. The identification of presence and nature of body fluids residues is important, especially when those residues contain DNA information (such as blood, semen and saliva) that allows the identification of a victim, witness or even the lawbreaker, leading to important information to solve cases. Therefore, it is necessary to use nondestructive methods to analyze those samples, in order to preserve DNA information. Near Infrared Hyperspectral Image (HSI-NIR) is a new, nondestructive alternative technique and, when associated with chemometric tools, can provide reliable and objective analysis, that does not depends on analyst experience. In this work, six different pieces of cotton fabrics (white, black, blue, red, green and yellow) were used as substrates to deposit semen and possible false positives. Raw and diluted semen samples from goat and horse respectively, which were used with experimental model and 4 different brands of lubricants were acquired and placed on the fabrics creating a stain. Hyperspectral images were acquired in each fabric containing the stains using a SisuCHEMA imaging system from Specim. The spectral range employed was 928-2524 nm, the spectral sampling per pixel was 6.3 nm, and the spectral resolution of 10 nm. The images were acquired with 50 mm lens with pixel size of 156x156 µm. Savitzky-Golay (SG) 1st derivative (2nd order polynomial and window’s width of 9) was applied to preprocess the spectra. Principal Component Analysis (PCA) and Partial Least Squares – Discriminant Analysis (PLS-DA) were performed to analyze the images. All chemometric treatments were performed using Matlab® R2010a and Hypertools toolbox. PCA scores images plot shows the difference between semen and lubricants. Loading plots from the first PC shows the water contribution at 1400 and 1900 nm, that is mostly associated with the pixels from lubricant stains. At the second PC, significant contributions at 1900 and 2150 nm in the loading plots reveals information from semen, even in the diluted sample, the 2150 nm region is due to N–H stretching and might be related with compounds present in semen that contains N–H groups, such as urea and spermine. For the black fabrics, due to the high absorbance in the region from 900-1500 nm and the presence of noise in the spectra, PCA were performed using only the range 1800-2430 nm and 17 points window size for SG derivative (2nd order polynomial).The same conclusions were achieved. PLS-DA models were built using the white fabric and then employed to classify the stains in the other color fabrics. In all cases, the raw semen goat samples were correctly discriminated. The diluted semen horse samples were not identified in the images for all fabrics. Nevertheless, hyperspectral imaging associated with chemometrics data treatment shows potential to be applied in semen detection on different fabrics as a presumptive and confirmatory method.