Automated Pulmonary Lung Nodule Detection Using an Optimal Manifold Statistical Based Feature Descriptor and SVM Classifier
Journal of Biomedical Engineering and Medical Imaging
The pulmonary lung nodule is the most common indicator of lung cancer. An efficient automated pulmonary nodule detection system aids the radiologists to detect the lung abnormalities at an early stage. In this paper, an automated lung nodule detection system using a feature descriptor based on optimal manifold statistical thresholding to segment lung nodules in Computed Tomography (CT) scans is presented. The system comprises three processing stages. In the first stage, the lung region is
... ted from thoracic CT scans using gray level thresholding and 3D connected component labeling. After that novel lung contour correction method is proposed using modified convex hull algorithm to correct the border of a diseased lung. In the second stage, optimal manifold statistical image thresholding is described to minimize the discrepancy between nodules and other tissues of the segmented lung region. Finally, a set of 2D and 3D features are extracted from the nodule candidates, and then the system is trained by employing support vector machines (SVM) to classify the nodules and non-nodules. The performance of the proposed system is assessed using Lung TIME database. The system is tested on 148 cases containing 36408 slices with total sensitivity of 94.3%, is achieved with only 2.6 false positives per scan. um e 4, No 4 , Au g 2 0 17 Moreover, machine learning based detection methods have also been used for false positive reduction based on genetic algorithm, neural networks [14, 19] , and genetic programming  . It should be noted that though the above classifiers comparatively large number of false positives are detected the still remain. In this paper we have used SVM classifier by radial basis function to classify pulmonary nodule detection to have improved accuracy and reduced false positive rate. The proposed system is evaluated using Lung TIME database  of thoracic CT scans with manually annotated pulmonary nodules. The experimental results have high degree of accuracy, sensitivity with reasonable specificity. The remainder of this paper is organized in three major sections: In sub-section 3.1segmentation of lungs from the raw DICOM CT images is described. Sub-section 3.2 presents the extraction of nodule candidates from the segmented lung tissues. Classification of true nodules from the segmented nodule candidates is discussed in sub-section 3.3. Results of a detailed evaluation on 148 cases containing 36408 slices are presented in section 4. Conclusions are given in section 5.