Research On Computer Aided Diagnosis Of Solitary Pulmonary Nodules

With the exacerbation of environmental pollution, lung cancer and other pulmonary diseases continue to rank as the leading reason of death, and significantly debase the quality of lives. The early detection and treatment is the most effective way to prevent incurability of pulmonary diseases. Most pulmonary diseases usually behave as Solitary Pulmonary Nodules (SPNs) in imaging. Thus the detection and recognition of SPNs is the best way to diagnose the pulmonary diseases. On the other hand, with the extensive application of CT scan, which is one of the most common imaging methods, the number of CT images is increasing in an exploding mode. This fact has lead to burdensome work load for physicians and negative influence of diagnostic performance. However this problem can be solved with the help of computer, which is also called after Computer Aided Diagnosis (CAD), due to the maturation of Digital Image Processing (DIP), Pattern Recognition (PR) and Machine Learning (ML) theory and technologies.The Computer Aided Diagnosis of SPNs problem is studied in this thesis. The main aspects of research include:1. A reformative automatic detecting method to SPNs. The traditional methods always use segmentation methods to detect SPNs in terms of fixed rules. Based on the traditional methods, this dissertation proposes a method to extract features of the segmented regions, and then use AdaBoost algorithm, a typical statistical learning method, to automatically distinguish the SPNs from other tissues in lung field.2. A classification scheme to distinguish between benign and malignant SPNs. The ultimate objective of diagnosis is to identify which disease the patients take. Unfortunately even in the clinical context, the identification is still a complicated problem for physicians and the clinical physicians have not come to an agreement about this diagnostic problem until now. But for the basic classification of SPNs, there is a set of systematic rules to obey to distinguish benign SPNs from malignant ones. Aimed at this classifying problem, 30 features based on these clinical rules are extracted from the SPNs detected by the method mentioned above. And then the Support Vector Machine (SVM) classification algorithm is applied to these 30-D vectors to distinguish the benign and malignant SPNs.3. A diagnostic scheme for pulmonary CT images based on Gabor representation of key regions. Because the imaging diagnosis is still an assistant technique for...