Automatic method of colposcopic multi-spectral images analysis for television systems diagnostics of cervical cancer

Abstract

Automated method of fluorescence images analysis obtained by excitation radiation with a wavelength of 360 and 390 nm is proposed. The method allows to detect the status of tissues of cervix: normal, chronic nonspecific inflammation (CNI) and cervical intraepithelial neoplasia (CIN), and build differential map pathology. For the border CIN/CNI achieved a sensitivity of 87 % and specificity 71 %. The method includes a specific preprocessing of the original images: combining images taken in different lighting conditions and highlight the area of interest. Features of the method are the use of a combination of features calculated for images of different types, and decision rule for classification based data mining techniques.

Keywords

• multi-spectral images processing
• multi-spectral images combining
• region of interest selection
• classification
• medical images processing