Introduction: In the lung cancers, a computer-aided detection system that is capable of detecting very small glands in high volume of CT images is very useful.This study provided a novelsystem for detection of pulmonary nodules in CT image.
Methods: In a case-control study, CT scans of the chest of 20 patients referred to Yazd Social Security Hospital were examined. In the two-dimensional and three-dimensional feature analysis algorithm, which were suspicious areas of pulmonary nodules and automatic diagnosis for evaluation, and the area segmentation results by active contour model, were compared with the results of the donation by the physician. Finally, to categorize the areas into two groups of cancerous and non-cancerous helping the MATLAB software Ver. 2014 b using Support Vector Machine (SVM) with three linear kernels, cubic polynomial and a kernel of the radial base function and repeated measurements test were analyzed at level of P≤0.05.
Results: The mean error for 10 cancer patients and 10 healthy individuals was 0.023 and 0.453, respectively and the best results were obtained using the RBF (Radial Basis Function) kernel algorithm and the σ = 0.28 parameter for it. Using the local area-based active contour model, the zoning time was reduced from 18.66 to 5 seconds on average and the calculated distances were calculated to be less than or equal to 0.75 mm; which indicates an increase in the speed of identification of high-precision pulmonary nodules.
Conclusion: In the proposed algorithm, the amount of false positive error and the time of identifying the nodules were significantly reduced and all areas suspected of being cancerous were identified with high accuracy and speed.
Feature Extraction of Ground-Glass Opacity Nodules using Active Contour Model for Lung Cancer Detection