Fax +41 61 306 12 34 E-Mail karger@karger.ch www.karger.com Techniques Acta Cytologica 2012;56:554–559 DOI: 10.1159/000341546 Computer-Assisted Diagnosis in Colposcopy: Results of a Preliminary Experiment? Grit Mehlhorn a Christian Münzenmayer b Michaela Benz b Andreas Kage b Matthias W. Beckmann a Thomas Wittenberg b a Department of Gynecology, Erlangen University Hospital, and b Fraunhofer Institute for Integrated Circuits IIS, Erlangen, Germany sions: The CAD system may be able to play a supportive role in the further diagnosis of CIN, potentially paving the way for new and enhanced developments in colposcopy-based di- agnosis. Copyright © 2012 S. Karger AG, Basel Introduction In Germany, a substantial reduction of the incidence of cervical cancer as well as of the resulting morbidity and mortality could be achieved by the introduction of a cy- tological public screening system in the early 1970s [1]. The aim of the Pap smear is to identify precancerous le- sions in the genitalia as early as possible, clarify them in a differentiated way, and treat them when appropriate. Once an abnormal cytological finding leads us to believe that a precancerous lesion or even a carcinoma could be present, the diagnostic guidelines require a histological analysis to be carried out for confirmation. To enhance the diagnostic accuracy of the assessment, colposcopic examination is an absolute requirement [2–4]. With much-improved visual conditions and the use of various contrast-enhancing solutions (such as acetic acid), better diagnostic clarification is possible on the basis of a tar- geted biopsy [5]. This type of examination requires exten- Key Words Cervical intraepithelial neoplasia  Colposcopy  Computer-assisted diagnosis  Automated tissue characterization  Image analysis Abstract Purpose: Diagnosis of cervical intraepithelial neoplasia (CIN) is currently based on the histological result of an aiming bi- opsy. This preliminary study investigated whether diagnos- tics for CIN can potentially be improved using semiautomat- ic colposcopic image analysis. Methods: 198 women with unremarkable or abnormal smears underwent colposcopy examinations. 375 regions of interest (ROIs) were manually marked on digital screen shots of the streaming documenta- tion, which we provided during our colposcopic examina- tions (39 normal findings, 41 CIN I, and 118 CIN II–III). These ROIs were classified into two groups (211 regions with nor- mal findings and CIN I, and 164 regions with CIN II–III). We developed a prototypical computer-assisted diagnostic (CAD) device based on image-processing methods to auto- matically characterize the color, texture, and granulation of the ROIs. Results: Using n-fold cross-validation, the CAD sys- tem achieved a maximum diagnostic accuracy of 80% (sen- sitivity 85% and specificity 75%) corresponding to a correct assignment of abnormal or unremarkable findings. Conclu- Received: March 30, 2012 Accepted after revision: June 28, 2012 Published online: September 27, 2012 Correspondence to: Dr. Grit Mehlhorn Department of Gynecology, Erlangen University Hospital Universitätsstrasse 21–23 DE–91054 Erlangen (Germany) Tel. +49 9131 85 33553, E-Mail grit.mehlhorn  @  uk-erlangen.de © 2012 S. Karger AG, Basel 0001–5547/12/0565–0554$38.00/0 Accessible online at: www.karger.com/acy Downloaded by: Universitaetsbibliothek Erlangen-Nuernberg 131.188.201.35 - 5/23/2014 11:20:03 AM