Neural Classifier Systems for Histopathologic Diagnosis

Neural network and statistical classification methods were applied to derive an objective grading for moderately and poorly differentiated lesions, based on characteristics of the nuclear placement patterns. Using a multilayer network after abbreviated training as a feature extractor followed by a quadratic Bayesian classifier allowed grade assignment agreeing with visual diagnostic consensus in 96% of fields from the training set of 500 fields, and a 77% of 130 fields of a test set