Development and evaluation of the virtual pathology slide: a new tool for understanding inter-observer variability in diagnostic microscopy
Costello, Sean (2004) Development and evaluation of the virtual pathology slide: a new tool for understanding inter-observer variability in diagnostic microscopy. PhD thesis, Dublin City University.
Full text available as:
The VPS (Virtual Pathology Slide) is a microscope emulator enabling the examination of pathology slides via the Internet or CD-Rom. A novel feature of the VPS is the ability to record the migratory traces (image viewed and magnification) of pathologists examinations on a remote relational database located in Dublin City University.
In order to evaluate the VPS, Ten breast needle core biopsies were randomly selected and presented to 17 pathologists or trainee pathologists with at least 2 years experience in pathology practice. Participants were required to examine each case online and provide a diagnostic classification using online feedback forms, based on the Core Biopsy Reporting Guidelines for Non-operative Diagnostic Procedures and Reporting in Breast Cancer screening as used by the British National Co-ordinating Committee for Breast Screening Pathology. The recorded data permitted examination of interobserver variability and user satisfaction.
The study demonstrated that Pathologists can make a correct diagnosis using the VPS. Consensus glass diagnosis agreed with consensus VPS diagnosis in 9 out of 10 cases. Consensus diagnosis for Slide 8 differed from glass slide diagnosis by one classification grade. Several Participants using the VPS achieved strong individual performance, with 10 of the 17 participants displaying “good” to “exce//e«i” (>0.6) agreement with VPS consensus, based on a weighted Kappa rating.
Modification of diagnostic classification based on a review of text diagnosis resulted in VPS consensus diagnosis for Slide 8 concurring with glass slide diagnosis and demonstrated a lack o f familiarity and understanding amongst participants in the application of the applied diagnostic guidelines, particularly in the diagnosis of Intraductal Pappilloma. Modification of diagnostic classification based on text diagnosis increased average overall slide consensus from 66.5% to 69.4% but decreased individual Kappa performance by 0.76 to 0.72.
Participants diagnostic performance was found to be unrelated to their confidence in making a diagnostic decision using the VPS. Perception of image quality was demonstrated to be clearly dependent on participants screen resolution and colour depth, but was shown not to influence diagnostic performance.
Perception of download speed was found to be unrelated to individual diagnostic performance. However, it was demonstrated that there is an increase in the number of fields of view examined by participants as their perception of download speed improves.
The number of fields of view examined per slide was found to be representative of the histological difficulty in interpreting a case. In general, as slide consensus decreases, the number of fields view examined for that slide increases. The number of fields of view examined at a particular magnification was found to be unique for each slide and dependent on the histological complexity of each slide.
To elucidate reasons for diagnostic inconsistency, a software application called ‘Bitmapper’ was developed. This generates a graphical representation of a diagnostic trace using data stored on the VPS database. This takes the form of 128x128 pixel bitmap image, where each pixel is representative of an individual field of view on a VPS slide at the highest magnification available. The colour value of each pixel is determined by whether the field of view it represents has been viewed, and if so, at what magnification.
This diagnostic trace was used to locate hotspot regions of potential diagnostic importance within a slide. For each of the slides a pathologist, specialist in breast disorders, examined images from these hotspots and successfully deduced a reason for diagnostic inconsistencies. This demonstrated that Bitmapper is an extremely useful tool for determining reasons for observer variation.
The development of the VPS and ancillary software tools was successful in that pathologists were willing to use the system. Pathologists could make a correct diagnostic decision using the system. The degree of observer variation could be quantified and using Bitmapper, reasons for observer variation could be determined. This technology has applications in determining the cause of observer variability and will prove a useful tool in external quality assurance studies (EQA) in pathology.
Archive Staff Only: edit this record