Computer-assistierte Diagnostik (CAD) für die Kolposkopie - Evaluation einer Pilotstudie

Abstract

The diagnosis of cervical intraepithelial neoplasias (CIN) of the Cervix uteri is currently based on cytology, individual experience of the colposcopic examiner and finally on the histological diagnosis. A novel developed prototypic system of computer-assisted diagnosis (CAD) for colposcopy yields high mean classification results of 80 % (sensitivity: 85 %, specifity: 75 %) based on an image based analysis (compared to the common diagnostic practice in colposcopy). Within this pilot study, the question was examined, if such a CAD system is in principle apt to enhance and support the assessment of the examiner and thus possibly improves the total diagnosis. With the context of an observer study 24 digitized colposcopic images of patients from the dysplasia consultation were assessed by 21 study participants. Among them were 8 experts with a colposcopic experience of more than 4 years and 13 non-experts with an experience less than 4 years. All study participants had acquired or refreshed the same colposcopic fundamentals in a workshop. All images were assessed in an organized way without and with the support of the prototypic CAD-system. As result, a higher diagnostic variance was seen for non-experts in comparison to the experts. An increase of confidence could be observed for the use of the CAD-system (Non-experts: mean sensitivity from 76 % to 82 %, mean specifity: from 62 % to 65 %; experts: mean sensitivity from 73 % to 79 %, mean specifity: 69 %). 85 % of the participants felt a higher confidence during the assessment while using the CAD-system. 76 % of the participants stated, they were able to perform the assessment in a shorter time. In its current development state, the applied CAD-System could have a supportive function within the continuative colposcopy of he cervical intra-epithelial neoplasias (CIN) and could thus potentially support and improve the colposcopic assessment.