Related Experiment Videos
Automatic intraductal breast carcinoma classification using a neural network-based recognition system
A Reigosa1, L Hernández, V Torrealba
1Image Processing Center, University of Carabobo, Venezuela Anatomic Pathology Department, Oncological Institute "Dr. Miguel Perez Carreño;" Venezuela System Department, Technological Institute of Valencia, Venezuela Department of Pathology, Red Cross Hospital of Valencia, Valencia, Venezuela.
Abstract:
A contour-based automatic recognition system was applied to classify intraductal breast carcinoma into high nuclear grade and low nuclear grade in a digitized histologic image. The image discriminating characteristics were selected by their invariability condition to rotation and translation. They were acquired from cellular contours information. The totally interconnected multilayer perceptron network architecture was selected, and it was trained with the error back propagation algorithm. Forty cases were analyzed by the system and the diagnoses were compared with that of pathologist consensus, obtaining agreement in 97.5% (p < .00001 of cases). The system may become a very useful tool for the pathologist in the definitive classification of intraductal carcinoma.