Parallel factor analysis of ovarian autofluorescence as a cancer diagnostic

Background and Objectives Endogenous fluorescence from certain amino acids, structural proteins, and enzymatic co-factors in tissue is altered by carcinogenesis. We evaluate the potential of these changes in fluorescence to predict a diagnosis of malignancy and to estimate the risk of developing ovarian cancer. Study Design/Materials and Methods Ovarian biopsies were interrogated over 270-550nm excitation and fluorescence was collected from 290 to 700nm. Two hundred forty-nine measurements were performed on 49 IRB-consented patients undergoing oophorectomy. Data are analyzed using parallel factor analysis to determine excitation and emission spectra of the underlying fluorophores that contribute to the total detected fluorescence intensity. Results Using multivariate normal distribution fits and cross-validation techniques, sensitivity and specificity of 88% and 93%, respectively, are achieved when classifying malignant samples versus others, while 88% and 80%, respectively, are achieved when classifying normal post-menopausal patients as being either at high- or low-risk of developing ovarian cancer based on their personal and family history of cancer. Performance of classifying cancer increases when the normal group does not include benign neoplasm and endometriosis samples. Performance of high- versus low-risk classification decreases when normal samples include both pre- and post-menopausal women. Excitation over 270-400 and 380-560nm, respectively, have the best diagnostic performance for cancer detection and risk-status assessment. Conclusions Assessing the endogenous fluorescence could be useful in screening women at increased risk of developing ovarian cancer.