A hardware-based Hybrid Lung Nodule Detection System is proposed for improving diagnostic accuracy and speed for lung cancerous pulmonary radiology. The configuration of the proposed system includes the following processing phases: (1) data acquisition and pre-processing, in order to reduce and to enhance the figure-background contrast; (2) quick selection of nodule suspects based upon the most prominent feature of nodules, the disc shape; and (3) complete feature space determination and neural classification of nodules. Our R&D work is aimed on extending existing digital processing techniques, developing new ones, and introducing robust Artificial Neural Network (ANN) architectures in the detection and classification of nodules. In this project, we would like (1) to automate the feature extraction and lung texture analysis techniques in ways that will improve the speed and accuracy of selection of suspect nodules; (2) to analyze the detailed suspect nodules and to derive the additional relevant parameters and characteristic patterns, which subsequently are used for the classification task; and (3) to test and assess the performance of the developed hybrid (digital/neural) computation system. A multistage ANN architecture involving early supervised learning processing stage (e.g., a back propagation stage) followed by self-adaptive output stage (e.g., a Kohonen feature map stage) is currently being investigated to serve as a basis for testing the proposed Hybrid Lung Nodule Detection System. this project will eventually lead to an efficient, cost-effective, robust, and hardware-based system for improvement of the accuracy and speed in nodule detection. It will also provide the basis for further development in other areas of diagnostic radiology.
| Chiou, Y S; Lure, Y M (1994) Hybrid lung nodule detection (HLND) system. Cancer Lett 77:119-26 |
