Reasearch On Automated Detection Of Pulmonary Nodules Based On Chest X-ray Images

As is known to all, lung cancer is one of the most common and lethal kinds of malignant neoplasm. If lung cancer was detected early and treated aggressively, the survival rate of patients can be effectively improved. Therefore, early diagnosis of lung cancer has important significance. Early lung cancer in medical imaging usually presents solitary pulmonary nodules and the nodules manifests as a circular or oval dense shadow in chest X-ray images. It is very difficult to classify the pulmonary nodules from the lung soft tissue areas only by eyes.With the improving development of computer science and technology, applying the pattern recognition algorithms to the medical field can be effectively assisted medical diagnosis and treatment. The CAD(Computer-aided Diagnose) System can detect and mark the candidate nodules to doctor for diagnosing. It will rescue the radiologists from unquantifiable X-rays, overcome the insensitivity for gray scale of eyes and improve the detection accuracy. Therefore, more and more researchers and doctors pay close attention to the research of automated detection of pulmonary nodules.This study is focus on the research of detect pulmonary nodules automatically in chest X-ray images. Firstly the paper reviewed the research status and development of computer-aided lung nodule detection algorithm. Four major research steps are discussed, including lung region segmentation, detection of interested spots, feature extraction and selection, and nodule classification. The main contribution of the paper has four main aspects are:(1) The paper proposes a new automated and improved active shape model and applies it to lung region segmentation. The new method not only achieved a fully automated segmentation of lung area, but also got more accurate results and a faster speed.(2) By comparing some spot detection algorithms, the DoG operator tests proved the best. The paper combined multi-scale scheme and DoG operator to achieve an efficient and accurate spot detection.(3) Considering the characteristics of pulmonary nodules on chest radiograph, we extract sevral features including location, scale, grayscale statistics and difference, and the texure feature respectively based on Gray-level cooccurrence matrix, Hessian matrix and Guassian derivative filters. Then we used the improved F-score algorithm to select features.(4) To solve the classification offset problems caused by imbalance data, we explored the solution by data preprocessing and algorithm improving. At last, the method combining the balance factor and feature fusion with SVM algorithm resulted to the experiments got satisfactory outputs.