Study On Laser Raman Spectra Of Nude Mice Models With Human Gastric Cancer In Vivo

Gastric cancer is one of the common malignant tumors in China, impairing people's health seriously. It requires a doctor to have a very rich clinical experience with visual identification of gastric cancer and normal tissue during operation. While the traditional pathological examination is not only time consuming, but also cause some pain and injuries to patients. In recent years, Raman spectroscopy attracted more and more attention to domestic and foreign scholars on detection and diagnosis in medicine because of its advantages such as non-invasive, high resolution, high sensitivity and high automation. This experiment is intended to explore the new methods and technologies of detecting cancer accurately, on real-time and in situ which can lay the foundation for helping physicians distinguish cancer during operation in the future.The near-infrared Raman spectroscopy system in our lab was set up for collecting the Raman spectra of the tissues, using 785nm near-infrared laser as the light source. A total of 205 Raman spectra in vivo were collected from cancerous and normal tissues by the near-infrared Raman spectroscopy system during simulated abdominal surgery on nude mice injected with SGC-7901.The mean Raman spectra of normal and malignant tissues were compared with each other after pretreatment. The results showed that there were significant differences in Raman spectra between normal and malignant tissues. Comparing the spectral intensity of normal tissues, the spectral intensity of malignant tissues were lower at 870cm-1(C-C stretching vibration of Hydroxyproline), 1450cm-1 (CH2 stretching or bending vibration) and 1660cm-1 (C=O stretching vibration of AmideⅠand H-O-H angle vibration of the water), and were higher at 1007cm-1(C-C symmetric stretching ring breathing of Phenylalanine),1050cm-1 (C-O stretching vibration of Deoxyribose),1093cm-1 (PO2- symmetric stretching vibration of Phosphodiester groups) and 1209cm-1 (C-C6H5 stretching vibration of Phenylalanine). In addition, the spectra of malignant tissues had two peaks at 1297cm-1 and 1331cm-1, while the spectra of normal tissues only had one peak at 1330cm-1 (C-H chain of Phospholipids and CH2 vibration of nucleic acid).The differences are attributed to the intensities of the nucleic acid, protein and water stretching bands. The increased DNA content in cancerous tissues may lead to raise at 1050 and 1093 cm-1, and the collagen is decomposed or its Raman spectra are covered that may be the reason of decrease at the 870cm-1 in cancerous tissue. These features can be the sign of cancer.The support vector machine (SVM) algorithm was used to classify the spectra of normal and malignant tissues. The kernel functions used in constructing SVM classifiers were the linear function, the radial basic function and the polynomial function. The SVM was trained on different parameter C and other parameters of the kernel function. And the performance of the SVM algorithms was evaluated with the 10-Fold Cross-Validation. The results of classification was optimal when the kernel function was the radial basic function, the parameter C=10 andσ=5. And the sensitivity, specificity and accuracy were 95.73%,70.73% and 90.73%, respectively.In order to understand of the process of carcinogenesis further, the SVM algorithm was used to classify the spectra of different growth stages of cancer. The results of classification was optimal when the kernel function was the radial basic function, the parameter C=100 andσ=15. And the sensitivity, specificity and accuracy were 98.82%,98.73% and 98.78%, respectively.From the results of the classification, we can see that the SVM method, with a good commendable sensitivity, accuracy and objectivity, is of great significance for identification of cancerous tissue in the surgical operation.