A hybrid neural-digital computer-aided diagnosis system for lung nodule detection on digitized chest radiographs

A hybrid neural-digital computer-aided diagnosis (N.CADx) system is proposed for early detection and classification of cancerous lung nodules of size 3-15 mm. Digital image processing techniques are used for noise reduction, image enhancement, and “suspect” search and localization. The neural classification concerns feature extraction and diagnosis of a particular pattern-class aimed at high degree of “true positive fraction” detection and low “false positive fraction”. Applying fuzzy linguistic concepts, we have developed a multi-label output encoding procedure, for purposes of neural training and for interpretation of the activity distributions in the output neurons. Over five output neurons, that were used in the nodule diagnosis problem, the activity distribution is interpreted by the centroid as the normalized disease index-NDI, or as a nodule detection probability-NDP value associated with a confidence level factor. The proposed N.CADx system has great potential in many medical applications