An analysis of the variation of human interpretation: Papnet a mini-challenge

The Papnet was given a mini-challenge of 200 cervical smears loaded to 50% with varying degrees of abnormality as interpreted by the originating laboratory. The range of abnormality extended from 'atypia' to invasive cancer and a few 'glandular' lesions were included as were a few smears which had been reported as 'inadequate'. Three cytologists (two cytopathologists and one cytotechnologist) read and analysed the 128 monitor pictures per slide, selected by the Scanning Algorithm and Neural Network systems. These results were compared with a 'gold standard' report on the glass slide produced by two cytopathologists. The analysis was done for each individual cytologist, for cases in which all three agreed, for a consensus between two of the three and for the 'best of three'. The latter gave an error rate of 4% false negative (Papnet scan negative) and 10% 'false positive' (referred for glass slide examination). Individual cytologists had higher error rates demonstrating that errors could be due to human interaction and not necessarily to the Scanner. This also indicated that wide experience in interpretation of monitor images is needed to achieve high quality results.