The high false-negative rate from Papanicolaou (Pap) smears has motivated the use of colposcopy to perform visual inspection of the uterine cervix. It has made a tremendous impact on the ability of physicians to perform accurate punch biopsies for histological analysis. However, the difficulty in training gynecologists in the recognition of pathology has hindered the full utilization of colposcopy. Colposcopic images contain complex and confusing lesion patterns. Correctly analyzing and classifying different types of tissues require substantial training. The goal of this research is to develop image analysis software to help physicians classify various abnormal cervical tissue patterns in colposcopic examinations. Such software will enhance the capability of colposcopy by increasing the diagnostic accuracy and speed, and simplifying the use of colposcopy, making its utilization more widespread and accurate. To recognize various vascular patterns characterizing different stages of dysplasia, texture analysis techniques will be used. We will use the texture spectrum method to represent the vascular patterns and to design features. Our primary objectives in Phase I are to implement this technique, evaluate its performance using a series of test images, and improve it. The Phase I effort will result in prototype software for discriminating vascular patterns. In Phase II, we will refine the software, conduct clinical trails, compare its performance with results from biopsies, and enhance its user-interface capability.
