Syntactic structure analysis of the arrangement of nuclei in dysplastic epithelium of colorectal adenomatous polyps

Anal Quant Cytol Histol. 1992 Dec;14(6):491-8.

Abstract

In search of new objective classifiers of the degree of dysplasia in colorectal adenomatous polyps, we conducted syntactic structure analysis on hematoxylin-eosin-stained tissue sections of 59 colorectal adenomas. The adenomas were subjectively graded as showing mild (n = 20), moderate (n = 20) or severe (n = 19) dysplasia in a double-blind examination by two pathologists. Using an interactive video-overlay measuring system, a minimum spanning tree was computed in 10 fields per specimen according to a strict measurement protocol. From each minimum spanning tree, such features as total line length, length of line segments, total number of points and number of points with one, two, three or four neighbors were derived. Simple descriptive statistics of these minimum spanning tree features were used to analyze differences between the three grades of dysplasia. Univariate analysis indicated that most of the variables used were suitable for discriminating between mild and moderate as well as between mild and severe grades of dysplasia. Only median minimum line length also showed significant differences between moderate and severe dysplasia. By means of a jackknifed stepwise discriminant analysis, an overall correct classification of mild versus moderate and severe dysplasia of 81.4% was achieved. It is concluded that syntactic structure analysis may have additional value in describing, in an objective and rapid way, the morphologic changes of dysplasia in colorectal adenomatous polyps.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adenoma / classification
  • Adenoma / pathology*
  • Cell Nucleus / pathology
  • Colonic Polyps / classification
  • Colonic Polyps / pathology*
  • Colorectal Neoplasms / classification
  • Colorectal Neoplasms / pathology*
  • Double-Blind Method
  • Epithelium / pathology
  • Image Processing, Computer-Assisted
  • Models, Biological
  • Neoplasm Staging
  • Reproducibility of Results
  • Statistics as Topic