Background and Objective The diagnostic contribution of cervical tissue fluorescence spectra acquired in vivo at 380 and 460 nm excitation were analyzed using a general multivariate statistical algorithm. Materials and Methods The primary steps of the algorithm are to: (1) preprocess data to reduce interpatient and intrapatient variation of tissue spectra from the same diagnostic category, without a priori information, (2) dimensionally reduce the preprocessed spectral data using Principal Component Analysis, and (3) develop a probability based classification scheme based on logistic discrimination using the diagnostically useful principal components. The algorithm was tested on cervical tissue spectra acquired from 165 sites at 380 nm excitation and from 147 sites at 460 nm excitation. A retrospective and prospective estimate of the algorithm's performance was determined. Results At 460 nm excitation, (1) SILs can be differentiated from normal squamous tissues with an average sensitivity and specificity of 91% ± 1.3 and 75.5% ± 1, respectively; furthermore, (2) high grade SILs can be differentiated from low grade SILs with an average sensitivity and specificity of 80% ± 4 and 76% ± 5, respectively. In addition, using tissue spectra at 380 nm excitation, SILs can be differentiated from normal columnar epithelia and inflammation with an average sensitivity and specificity of 77% ± 1 and 72% ± 9, respectively. Conclusions Fluorescence spectra at multiple excitation wavelengths are essential for the detection and differential diagnosis of SILs at colposcopy. © 1996 Wiley-Liss, Inc.
Read full abstract