The Research Of Key Techniques In Solitary Pulmonary Nodule Computer Aided Detection

Most of lung pathological changes will show as Solitary Pulmonary Nodules (SPN), whose inner structures, dimensions, edge features and relationships with tissues around them are closely related with the disease nature, benign or malignant. The application of SPN Computer Aided Detection (CAD) system is of advantage in discovering the doubtful pathological changes and determining the disease nature. SPN CAD system is very important to improving the rate of diagnosis accuracy and reducing the quantity of failed detection and become the important method of early lung cancer diagnosis.SPN CAD system is comprised of five parts: data collection, image pre-processing, image segmentation, image display and detection & analysis. Each of these parts has its own knowledge fields and difficult points. The technologies of first two parts are relatively mature. However, there are some key techniques which still not be solved nowadays in the last three parts. Three key points among them will be studied in this paper.Firstly, the problem of SPN precise segmentation has been studied. After analyzing the common segmentation methods and their advantages and disadvantages, an improved level set segmentation approach base on Rayleigh Mixture Distribution has been forwarded aiming at the specialty of SPN. This approach has efficiently solved the problem of border leakage in traditional level set method. Another hybrid approach integrating boundary-based and region-based segmentation methods based on the fuzzy connectedness and Voronoi diagram classification algorithms has also been forwarded to complete the segmentation task when the SPN is adhesion with the chest or abdominal cavity and of direct 3D segmentation. Good segmentation results have been obtained by both two approaches.Secondly, the problem of 3D visualization of CT images has been studied. After analyzing the common volume rendering methods and their advantages and disadvantages, especially the algorithm complexity of popular ray-casting volume rendering method, an improved rapid tri-linear interpolation method has been forwarded aiming at the problem of huge computation and low efficiency of traditional tri-linear interpolation algorithm in ray-casting volume rendering method.Thirdly, the problem of tumor volume measurement has been studied. After converting the volume measurement to the section area measurement of tomography images and deeply analyzing the popular polygonal approximation algorithm, an optimal polygonal approximation algorithm has been forwarded aiming at the shortage of traditional polygonal approximation algorithm. The vertexes of the approximating polygon have been decreased and the calculating efficiency has been improved.Lastly, medical images computer aided detection system has been designed and realized. Appearance, technical architecture and functional modules have been designed and the functions of each module in medical images computer aided detection software have been introduced one by one.