Pictorial Pattern Recognition Applied To Morphometric Cytology

Abstract

This paper discusses algorithms for an interactive picture-processing system which would gather quantitative morphological information on four organelles (mitochondria, glycogen, smooth endoplasmic reticulum and rough endoplasmic reticulum) from electron micrographs of liver cells. Many investigations have shown that quantitative data is more indicative of an alteration of function than qualitative data. Unfortunately, the techniques to obtain quantitative data currently in use are slow and sometimes unreliable. The algorithm is a ladder-structured decision tree in which processing technique image resolution is a function of the depth in the decision tree. The algorithm will generate chain-encoded descriptions of each organelle discovered in the current image from which both quantitative and qualitative data may be extracted.