OBJECTIVE: To evaluate the clinical usefulness of a computer based system that stores and analyzes high-resolution digital colposcopic images. METHODS: Three hundred patients referred to our institutions for evaluation of abnormal cervical cytology underwent colposcopy with development of digital images. The first 70 patients were used to standardize the system. The images obtained from 188 evaluable patients were analyzed and scored by software according to internationally recognized colposcopic criteria. The results were compared with traditional colposcopic diagnoses and with corresponding histology results using kappa statistics for inter-observer agreement and McNemar test for significance. RESULTS: The exact concordance rate between computer-aided impressions and histology (85.1%, kappa = 0.77) was significantly higher (P < .001) than that observed between traditional colposcopic findings and histology (66%, kappa = 0.40). The computer-assisted colposcopy was much more accurate than traditional colposcopy in diagnosing high-grade lesions (91.2 versus 61.8% of exact concordance, P < .001) as well as normal histology (74.1 versus 34.5%, P < .001), but not significantly different when evaluating low-grade lesions (89.6 versus 86.5%). CONCLUSION: The integration of computer imaging and colposcopy can improve the colposcopic diagnostic accuracy. An inexperienced colposcopist may benefit from computerized support to obtain the most appropriate histologic specimen, and eventually access to distant consultation via modem or through on-line services. An additional advantage is the ability to develop a space-saving permanent record of digitized images readily available to review a patient's cervical history or perform effective programs of quality control in colposcopy.
OBJECTIVE: To evaluate the clinical usefulness of a computer based system that stores and analyzes high-resolution digital colposcopic images. METHODS: Three hundred patients referred to our institutions for evaluation of abnormal cervical cytology underwent colposcopy with development of digital images. The first 70 patients were used to standardize the system. The images obtained from 188 evaluable patients were analyzed and scored by software according to internationally recognized colposcopic criteria. The results were compared with traditional colposcopic diagnoses and with corresponding histology results using kappa statistics for inter-observer agreement and McNemar test for significance. RESULTS: The exact concordance rate between computer-aided impressions and histology (85.1%, kappa = 0.77) was significantly higher (P < .001) than that observed between traditional colposcopic findings and histology (66%, kappa = 0.40). The computer-assisted colposcopy was much more accurate than traditional colposcopy in diagnosing high-grade lesions (91.2 versus 61.8% of exact concordance, P < .001) as well as normal histology (74.1 versus 34.5%, P < .001), but not significantly different when evaluating low-grade lesions (89.6 versus 86.5%). CONCLUSION: The integration of computer imaging and colposcopy can improve the colposcopic diagnostic accuracy. An inexperienced colposcopist may benefit from computerized support to obtain the most appropriate histologic specimen, and eventually access to distant consultation via modem or through on-line services. An additional advantage is the ability to develop a space-saving permanent record of digitized images readily available to review a patient's cervical history or perform effective programs of quality control in colposcopy.