Quality Assurance of Cervical Smear Slide Inspection Using a Novel Eye-Tracking Technique

Abstract

A novel objective quality assurance system for smear slide screening is investigated in this thesis. A method of data validation was developed that compares data from an eye tracked image display, machine image colour texture analysis and expert judgements in a statistical manner to identify salient areas of cervical cytological images. These data are used to construct screener performance profiles, which have been compared to screener experience. The experimental methodology is described and how the screener performance profile is constructed. Results from a study of 10 screeners, checkers and pathologists are presented showing predicted trends of human performance. Relations to experience and strategy are also shown, though these relationships are not statistically significant. A standardised quality assurance test is developed that profiles screeners across many performance measures. Highly significant correlations were found between fixation saliency and machine colour texture (maxima density), though fixation saliency suffers from a lack of a significant statistical basis. Further fixation data is needed, however if it conforms to the existing trends then the results would support the new data validation method as a framework from which image analysis techniques applied to cytology may be objectively tested. Furthermore, this new approach to cervical cytology quality assurance would have the potential to further reduce human errors in the cervical smear inspection process by lowering levels of observer variation found in all aspects of the cervical screening process.