Intraoperative diagnosis of thyroid diseases by fourier transform infrared spectroscopy based on support vector machine

Background: The recent development of attenuated total internal reflectionFourier transform infrared (ATR-FTIR) spectroscopy has provided a new avenue for distinguishing cancerous tissue from normal one. The focus of this investigation is to develop a novel spectral discriminant method for thyroid malignant and benign samples, intraoperatively. Methods: A total of 112 cases of human thyroid tissues, were obtained and underwent ATR-FTIR spectroscopy scanning intra-operatively. The average ATR-FTIR spectra of nodular goiter and thyroid carcinoma was built. Standard normal variate (SNV) method was applied to cut down scatter effect, and support vector machine (SVM) discrimination model was used to discriminate spectra of benign thyroid diseases from malignant one. Leaveone-out cross validation (LOOCV) was exploited to evaluate SVM discriminant effects. Results: 67 nodular goiter (benign) and 45 thyroid carcinoma (malignant) were pathologically diagnosed. The average ATR-FTIR spectra of nodular goiter was significantly different from thyroid carcinoma group. The sensitivity, specificity, and accuracy rate of the SVM algorithm’s discriminants were 84.4%, 88.0%, and 86.6%, respectively. Conclusion: A novel approach to distinguish nodular goiter from thyroid carcinoma intraoperatively by using the ATR-FTIR technique combined with mathematical procedures of SVM, was established and demonstrated.