Research On Solitary Lung Nodule Detection Of Low-dose CT Images

Lung cancer is the most ordinary malignant tumor nowadays, and early detection and treatment can significantly increase the chance of survival for patients. Lung nodule is the major indication of lung cancer on CT images, which is approximate circular sickness region whose diameter is less than 3cm. It always shows as compact regions on CT images. HRCT scanning has been the most important method of lung cancer examination, however, reading the mass of the CT images requires a lot of time, energy and considerable experience. Because doctor's different clinic levels and fatigue often lead to misdiagnosis, the use of computer-aided detection(CAD) is a trend. The CAD for lung nodules is currently the hot spot of research. High detection accuracy and low false positive is the key indicators.The research around lung nodule CAD includes the following aspects:1. A novel MIP-based lung nodule detection algorithm is proposed. Specifically, local three-dimensional information has been integrated into the maximum intensity projection(MIP) images, where the region of interest has been detected. This method is effective to remove the false positives of the blood vessels and tracheas, greatly improves detection efficiency.2. In order to future reduce false positives, a MIP-based three-dimensional lung nodule detection algorithm in HRCT images is proposed. Using the algorithm before to make a two-dimensional detection of lung nodule, extracting the seeds to do adaptive three-dimensional growth. Extracting the volume of interest's features and classifying them so as to achieve three-dimensional detection of lung nodule. This method detects more accurately.3. In order to solve automate detection problems of lung nodules, a pattern recognition method based on two-dimensional medical signs of lung nodule is proposed. According to the characteristics of lung images, systematically extracting gray, geometry, texture and location features. In response to these medical signs, designing the multi-dimensional feature-based SVM classifier to distinguish the lung nodules from normal tissues and achieving good results.The experimental results in a fifty-three cases database show that, true positive rate of the proposed system achieves 87.8% and false positives limit 0.08 per lay. Compared with conventional detection algorithm, the proposed system has the advantages including higher precision, lower false positives and also can be performed in real-time, and solves the problem of the solitary lung nodules computer-aided detection.