Quantitative Autofluorescence Imaging Reveals the Potential for Cervical Cancer Detection

Autofluorescence imaging is an emerging method for detection of cancer associated risks, depending on the variation in autofluorescence property of normal and cancer cells. In the present study, autofluorescence images of cervical cell were assessed for early cervical cancer risk prediction. For classification of cervical epithelial cells collected from normal and clinically diagnosed cancer patients, a set of spectral texture features (SPTF) were extracted from 2-dimensional (2-D) Fourier transform of autofluorescence images. To discriminate the normal and the abnormal cells, SPTFs were evaluated by two 1-D functions, i.e., radial function and angular function, accumulated from the frequency spectrum. The classification was assessed by four different types of support vector machines (SVMs), namely, Linear SVM, Quadratic SVM, Cubic SVM, and Gaussian SVM. The best accuracy was achieved by Gaussian SVM (0.95) with a sensitivity of 0.90 and specificity of 0.88. The overall accuracy of all the classifiers was more than 0.82. This experiment was also evaluated by receiver operating characteristics (ROC) curve and area under ROC curve (AUC) of each classifier, which reveals the comparative results for this database.