Z Gastroenterol 2005; 43 - 24
DOI: 10.1055/s-2005-869671

Automated histological classification of colon biopsy samples using image analysis on digital slides

L Ficsór 1, B Molnár 1, P Gombás 2, Zs Tulassay 1
  • 12nd Dept of Medicine, Semmelweis University, Budapest, Hungary
  • 2MI Central Hospital, Budapest, Hungary

Aims: Digital slides and virtual microscopy was shown to be alternative techniques for routine gastroenterological biopsy specimen analysis. Moreover automated, quantitative, reproducible classification of gastric slides into major histological diagnostic groups has been achieved by image analysis techniques.

Aims: Our aim was to adopt our gastric experiences to the colon cases detecting cell nucleus, glands, and surface epithelium; and classify the samples to different disease classes using statistical data.

Methods: Altogether 69 pieces of routine chronic colon well oriented mucosal biopsy specimen were selected (24 normal, 11 aspecific colitis, 25 colitis ulcerosa and 9 Crohn disease). To digitalize the selected H/E stained slides 3DHistech's Hi-Scope slide scanner system was used. Automatic histological evaluation modules were developed in C++. Altogether 45 parameters described the area, cell density and cellular characteristics of the basic tissue components as the surface epithelium, glands, connective tissue and the inflammatory cell compartment. In each tissue component cells' morphometrical features were calculated. Area and contained cell number ratios of different tissue compartments were also calculated which we named to tissue cytometric features.

Results: Significant differences were found by several ratios (biopsy/lamina cell number (BLCN); biopsy/epithelium cell number (BECN); cell concentration in lamina/biopsy L/B); glands shape GS)), the most significant parameters we found were the BLCN and the L/B (BLCN: healthy 1.57±0.17; aspecific colitis 1.34±0.16; colitis ulcerosa 1.18±0.09; Crohn disease 1.28±0.11, p<0.01:–-: L/B: healthy 1.13±0.09; aspecific colitis 1±0.03; colitis ulcerosa 0.97±0.02; Crohn disease 0.96±0.03, p<0.01).

Conclusions: This preliminary study proved again that the development and evaluation of quantitative tissuemetric features can be used in the automated classification of histological colon biopsy specimen.