CT Images Analysis for Early Detection of Lung Cancer
Sammouda, Rachid . 2008
Chest CT images Arti fi cial neural networks Segmentation Lung cancer diagnosis
An automatic Computer-Aided Diagnosis (CAD) system for early detection of lung cancer by analysis of chest 3D Computed Tomography (CT) images is proposed
in this paper. Our system differs from previous implementations of CAD systems in two major aspects: 1) It provides automatic extraction of the lung regions from 3D-CT images based on the bit-plane slicing technique, which has improved accuracy and sharpness. 2) Unsupervised segmentation of the extracted lung regions, our regions of interests (ROIs), using a modified version of Hopfield Neural Network (HNN), was able to segment the lung area into regions with crisp contours. These regions may include true lung nodules,and normal structures consisting mainly of blood vessels. To distinguish between true and false candidate nodules, we have adopted a rule-based approach consisting of two steps:
1) We calculate the area, the maximum drawable circle (MDC), and the mean intensity value of each segmented region.
2) A set of diagnostic rules has been formulated based on the extracted features, which aims at eliminating (as far as possible) non-cancerous
candidate nodules or false positives (FPs) without sacrificing cancerous candidates or true positives (TPs). We have evaluated our system using a database of 2668 CT slices from 11 patients. We have obtained 90% sensitivity, with 0.05 false positives per slice. The proposed CAD system is capable of detecting lung nodules with diameter equal to or greater than two millimeters.
Presents contributions to improve a previously published approach for the segmentation of magnetic resonance images of the human brain, based on an unsupervised Hopfield neural network.
The paper presents a method for automatic segmentation of sputum cells color images, to develop an efficient algorithm for lung cancer diagnosis based on a Hopfield neural network.