Automatic Evaluation of Multi-Parameter Fluorescence Micrographs with a Neural Network Architecture

A neural network architecture for the automatic evaluation of uorescence microscopy images of tissue samples is presented. One image is segmented into regions of uorescent lymphocyte cell bodies and background. The system uses two kinds of neural nets to (i) detect the uorescent cells and (ii) compute a one-pixel wide contour line for each detected cell. Even though the cells contours show considerable variation and the have nonhomogenous uorescence signals across the cell body the system detects a minimum of 95% of the cells with less than 10% false positives, using human expert evaluation of the images as a reference. Keywords: shape recognition, contour grouping, uorescence microscopy, functional proteomics 1 Introduction In this work we present a system for the rapid and exact evaluation of multi-parameter uorescence images. The platform of our study is the evaluation of images of uorescent lymphocytes in a tissue sample to prove the presence of n cell surface proteins at ..