Research And Implementation Of Automated Detection Algorithm Of Lung Nodules Based On CT Images

Lung Cancer is one of the malignant tumours which have most damaging to human life, and mortality rate is higher than other cancer and increasing year by year. Computed tomography is the most common imaging in imaging diagnosis of thoracic and is widly used for the detection of lung cancer. And lung cancer most commonly manifests itself as pulmonary nodules. Since pulmonary nodules are characterized by various shapes, different size, uncertain location, easy to link with other organizations, similar density with some pulmonary organizations, manifesting itselt as round or oval dense shadow in CT image, It is difficult to separate pulmonary nodules and lung tissue with eyes only. Even experiend doctors do an objective and accurate analysis is quite difficult. In addition, the process of doctor analyze thoracic CT image is boring and tedious, and large amount of data result in difficult to avoid anylisis error. Accordingly, computer aided detection and diagnosis(CAD) is developed, and it can analyze CT images of thoracic automatically and present pulmonary nodules to help doctor more effectively analyze data security and overcome some objective factors. Whereas, how to properly detect pulmonary nodules is the key technology of CAD.This paper does some research on the detection algorithm of pulmonary nodules and presents an automatic detection method of lung nodules based on the special gray distribution of the organizations in CT images cf thoracic. First, according to the special chest structure, a new segmentation algorithm is proposed to extract lung parenchyma. Second, lung parenchyma is divided into nodule section and background section. Nodule section includes pulmonary nodules, pulmonary vessels and airways with similar density with pulmonary nodules, and the section is recorded as regions of interest (ROI). Then, the charactors of pulmonary nodules, including area, average gray, variance, spherical degree, shape descriptor, fourier descriptor, are selected as the features of classifier. Third, ROI are classified by the classifier. Among, the classifier is designed by the nearest neighbor algorithm and trained based on the pulmonary nodules in LIDC as the sample data. The last, the corresponding regions of ROI sentenced pulmonary nodules are marked in red.The automatic detection software of pulmonary nodules is completed, and reaches the requirements. Experimental result shows that the algorithm generally considers characters of lung nodule and arrives at higher accuracy and lower false rate positive.