P041 Automated image analysis as a diagnostic tool in discriminating between subgroups of microscopic colitis
A.-M. Kanstrup-Fiehn1, M. Kristensson2, U. H. Engel3, L. K. Munck4, 5, S. Holck3, P. J. H. Engel*1, 5
1Roskilde Hospital, Department of Pathology, Roskilde, Denmark, 2Visiopharm, Hoersholm, Denmark, 3Copenhagen University Hospital, Department of Pathology, Hvidovre, Denmark, 4Koege Sygehus, Department of Gastroenterology, Koege, Denmark, 5University of Copenhagen, Faculty of Health and Medical Sciences, Copenhagen, Denmark
The diagnosis of microscopic colitis (MC) rests on a triad of clinical symptoms, a normal or near-normal endoscopy, and characteristic histopathological findings. MC comprises the subgroups collagenous colitis (CC), lymphocytic colitis (LC), and incomplete microscopic colitis (MCi), that is incomplete collagenous colitis (CCi) and incomplete lymphocytic colitis (LC). The inter-pathologist agreement in discriminating MC subgroups from inflammatory bowel disease (IBD) and normal colonic mucosa using Haematoxylin-Eosin (HE) stained slides is very good, but agreement is lower when discriminating between the MC subgroups. Distinguishing between CC and CCi may be difficult, even with special staining of the subepithelial collagenous band.
Automated image analysis (AIA) has been found useful in quantifying fibrosis, for instance in liver biopsy specimens, and the aim of the present study was to develop software for AIA, measuring the thickness of the subepithelial collagenous band in colon biopsies with CC and CCi.
A training set of 10 colon biopsies, diagnosed as CC, CCi, and normal, was used to develop software for AIA to match the assessment by gastrointestinal pathologists. The study set consisted of biopsies from 75 patients; 25 each with CC, CCi, and normal or near normal. Four pathologists individually reassessed the biopsies and categorised them into 1 of the 3 above-mentioned categories. Diagnosis based on measurement of the collagenous band by AIA was compared with the pathologists’ diagnosis.
The Cohen kappa (κ) values for each pair of pathologists and the software is shown in Table 1. The overall inter-observer agreement between the 4 pathologists was κ = 0.69, whereas the overall agreement between the 4 pathologists and AIA was κ = 0.75.
Table 1. Cohen kappa (k) values for each pair of pathologists and the AIA software
|Pathologist 1||Pathologist 2||Pathologist 3||Pathologist 4||AIA software|
|Pathologist 1||κ = 0.81||κ = 0.63 (0.48-0.77)||κ = 0.75 (0.63-0.88)||κ = 0.79 (0.67-0.91)|
|Pathologist 2||κ = 0.65 (0.50-0.79)||κ = 0.73 (0.60-0.86)||κ = 0.77 (0.65-0.90|
|Pathologist 3||κ = 0.56 (0.41-0.72)||κ = 0.63 (0.63-0.88)|
|Pathologist 4||κ = 0.75 (0.62-0.88)|
Measuring the thickness of the subepithelial collagenous band by AIA is congruent with the diagnosis made by trained pathologists. AIA may be useful as a supplementary diagnostic tool in borderline cases of CC and in differentiating between CC and CCi. Being an objective and reproducible tool, eliminating inter-observer variability, AIA is also a tool to achieve uniform pathological diagnosis in multicentre studies. If applied to prospective cohort studies, AIA could help to redefine the histopathological criteria of CC and CCi.