Image analysis for discrimination of cervical neoplasia

Colposcopy involves visual imaging of the cervix for patients who have exhi bited same prior indication of abnormality, and the major goals are to visu ally inspect far any malignancies and to guide biopsy sampling. Currently c olposcopy equipment is being upgraded in many health care centers to incorp orate digital image acquisition and archiving. These permanent Images can b e analyzed for characteristic features and color patterns which may enhance the specificity and objectivity of the routine exam. In this study a serie s of images from patients with biopsy confirmed cervical intraepithelia neo plasia stage 2/3 are compared with images from patients with biopsy confirm ed immature squamous metaplasia, with the goal of determining optimal crite ria for automated discrimination between them. All images were! separated i nto their red, green, and blue channels, and comparisons were made between relative intensity, intensity variation, spatial frequencies, fractal dimen sion, and Euler number. This study indicates that computer-based processing of cervical images can provide some discrimination of the type of tissue f eatures which are important for clinical evaluation, with the Euler number being the most clinically useful feature to discriminate metaplasia from ne oplasia. Also there was a strong indication that morphology observed in the blue channel of the image provided more information about epithelial cell changes. Further research in this field can lead to advances in computer-ai ded diagnosis as well as the potential for online image enhancement in digi tal colposcopy. (C) 2000 Society of Photo-Optical Instrumentation Engineers . [S1083-3668(00)00401-9].