6th ESACP Congress, Heidelberg, April 7-11, 1999

A131
TECHNICAL PROBLEMS RELATED TO SYNTACTICAL STRUCTURAL ANALYSIS (CELLULAR SOCIOLOGY) IN TISSUE SECTIONS OF THE INVASIVE FRONT OF CARCINOMS
Sudbo J, Reith A

Dept. Pathology, Section for Digital Pathology, The Norwegian Radium Hospital, Oslo, Norway

It is a well established observation among pathologists that tumour morphology in the most invasive parts (invasive front) of a malignant carcinoma often differs from more central and/or superficial parts of the same tumour. At the invasive front the tumour frequently shows a lesser degree of differentiation and a higher grade of cellular dissociation than the remaining areas of the tumour. With emphasis on reproducibility, simple semiquantitative malignancy grading systems have been developed specifically for evaluation of the invasive front of carcinomas. Such a grading system has proved to have additional prognostic value over established prognostic factors. Reproducibility is however, still a matter of debate. As one possible approach of increasing the reproducibility of the classification/prognostication of this crucial area of epithelial tumour, we have employed methods of syntactic structural analysis for the diagnosis of epithelial lesions in the form of oral squamous cell carcinomas. These are methods based on the mathematical models derived from graph theory as mathematical tools for describing the texture on tissue level. We have been able to demonstrate consistent morphological differences between normal epithelium and tumor tissue and, moreover, to distinguish between the profound parts of the epithelial lesions and the invasive front. However, several shortcomings of the methods that have to be taken into consideration when evaluating the results. These are technical problems which arise as a result of incomplete HE-staining of the nucleus/cytoplasm in the invasive front of some of the carcinomas and general problems related to nuclear staining (Feulgen) on squamous cell epithelium. Segmentation algorithms require a minimum of contrast in order to distinguish objects of interest in a reliable manner. Cases will be presented where the problems of automated procedures for tissue morphology assessment are discussed in further detail.