Novel Approach to Computer-Aided Detection of Lung Nodules of Difficult Location with Use of Multifactorial Models and Deep Neural Networks

The questions of processing and analysis of complex digital signals are considered. Such signals are represented in a form of digital lung images, which were made with use of computed tomography. The main aim of processing and analysis of input images is the identification of potential lung nodules. To solve such a problem an appropriate procedure has been suggested, which is based on the binarization of input image and its fragmentation into energy levels, the use of methodology of neural networks for identification of lung nodules. It was shown that suggested procedure allows to identify the intraparenchymal location of lung nodule with a probability of 98.5%, subpleural location with a probability of 75.4%. The proportion of false detections is 8%.