Cascade-correlation neural networks for the classification of cervical cells
Stephen James McKenna, Ian W. Ricketts, Alistair Y. Cairns, K. A. Hussein · 1992
As part of its programme of exploring the medical applications of computer vision, the Computer Vision Research Group at the University of Dundee have been researching into the automatic inspection of cervical smears. The authors have assembled a database containing over 2000 expertly verified cervical cell images and have investigated the performance of a number of competing techniques applied to this very demanding inspection task. The authors examined the performance of cascade-correlation neural networks, as developed by Fahlman and Lebiere (1990), to classify isolated cervical cells as either normal (benign) or abnormal (indicative of pre-cancerous changes). A set of 80 features extracted from cell images' 2D discrete Fourier transforms has been used as a basis for classification.