A Feature Level Image Fusion Algorithm Based On Rough Sets For Lung Nodule Detection

Background Lung cancer has been the highest morbidity and mortality of malignant tumor around the world, early detection is an effective way to improve the efficiency of the lung cancer treatment, and due to lung nodule is the early form of lung cancer, so the importance of detection of lung nodules recognition in lung cancer treatment was highlighted. Computed Tomography(CT) imaging for clinical diagnosis is of multiple perspectives, visualization and high quality, but with the wide application of CT in lung nodules detection, the factors such as CT image data overload and subjective interpretation led to a high clinical misdiagnosis rate.Objectives Lung CT images were regard as research objects in this paper. According to three aspects such as the segmentation algorithm, the feature description and extraction of ROI, classifier design, in view of the CT images of lung nodule segmentation less consider the spatial distribution of lung nodules, irrational characteristics structure and non-tightness of feature expression, a lung nodule detection model was proposed based on rough set at feature level fusion.Methods 1.During the section of original CT image preprocessing, lung parenchyma and candidate ROIs contain different types of lung nodules were segmented. Firstly, the lung parenchyma is segmented using clustering algorithm based on the connection of three dimensionality, and then lung nodules are divided into three categories on the basis of fully considering the lung nodules in spatial distribution: isolated lung nodules, pleural adhesion lung nodules and vascular adhesion lung nodules, finally based on gray level distribution and the geometric structure of different types of lung nodules, different segment methods based on connectivity, gray level drops, the divergence difference algorithms are used respectively for segmentation of ROI. 2. To extract the shape features, intensity characteristics and texture features. In terms of the shape feature, three new 3D characteristics was put forward, namely the external spherical volume, surface-standard deviation and circumscribed cuboid center distance intersection distance; and with respect to intensity characteristics, three new 3D features was put forward, the intensity gradient, Laplace average divergence, Laplace divergence distance. 2D texture characteristics, 3D shape features, and three-dimensional intensity quantitative were used together to describe the features. 3. With respect to the high-dimensional redundancy problem of the feature sets, and in order to eliminate the correlation between the features, rough set model with no prior knowledge was used to select feature subsets, and the grid optimization algorithm was used to optimize the SVM with 10-cross validation to choose the optimal kernel function, and which was applied to the recognition of ROI.Results In order to verify the validity, stability, advantages of the model, four groups of comparative experiments are performed in this paper, i.e., model validation experiments before and after rough set reduction, model stability experiments before and after rough set reduction, validation experiments of the superiority of the rough set feature level fusion model, and comparative experiments with other lung nodule detection models to compare the performance. The experimental results show that the method proposed in this paper can improve, to a certain extent, the rationality of feature structure and compactness of feature expression, thereby improving the detection accuracy of lung nodules.