Research On Several Key Problems Of Lung Cancer Qualitative Diagnosis Based On CT Images

The mortality of lung cancer ranks the highest among most common malignant tumors. The incidence of lung cancer gradually increased in China. Using CT imaging technology, early discovery, early diagnosis and early treatment are the primary means to improve the survival rate of patients. A lot of clinical experiences indicate that qualitative diagnoses based on medical signs lead to a higher rate of false positive and false negative results. Due to the complexity of illness type and diversity of the pathological changes, the efficient improvement of accuracy in qualitative diagnosis based on medical images is a big challenge for existing knowledge and techniques.Computer-aided diagnosis (CAD) plays an active role in improving the diagnostic accuracy, popularizing exams, and reducing the missed lung cancer. CT as the representative imaging diagnosis is based on medical signs, such as size spatial location and external shape, to speculate their pathology. At present, most studies are focused on positioning algorithms, quantitative calculation and sorting models based on visual information. These studies pay so much attention to morphologic analysis that limits its contribution to accuracy of qualitative diagnosis.This paper, with the lung cancer CT images scanned by multi-slice spiral CT as the research objectives, starts from density change information of lung cancer CT image as research point. Deep studies in lung cancer early diagnosis with CT images are mainly about:First, based on the principle of CT, it is indicated that density values and their change in CT images reflect density distribution of human tissues. A hypothesis about the correlation between density distribution change information in CT images and tissue type is proposed, which provides the theoretical basis for finding features of CT images directly relevant to the pathological tissue type of lung cancers.Aiming to the limitation of qualitative diagnosis of lung cancers based on morphological features in CT images, a computer-aided qualitative identification method, based upon spatial features of density change in CT images, is proposed. This method constructs the multi-resolution histogram eigenvector fit for extraction and analysis of spatial information of density in CT images, which overcomes lacks of methods imitating doctors through quantifying medical signs, breaks the limitation of morphological information in medical images, and provides a new way to obtain information hiding in those images.To avoid unilaterally seeking local validity in past studies, a systemic construction method of the high-dimensional histogram eigenvector based upon the SVM algorithm was developed. This method takes full advantage of SVM for dealing with high-dimensional data sets. It can identify and classify different pathological changes, avoiding multiple choices of eigen sets for different pathological changes. The method accords with the principle of finding more efficient characteristics through clinic experience, and has better generalization capacity.Aiming to diagnosing solitary pulmonary nodules (SPN) in CT images qualitatively, a new SPN benign and malignant classification method combining high-dimensional multi-resolution histogram with SVM algorithm was developed. Results of 240 experiments on SPN images showed that the method can effectively express image spatial information the characteristics of SPN, the classification accuracy of benign and malignant SPNs is up to 71.67% without medical signs. It can also be used to classify malignant SPN and is a convenient objective aided means.In order to solve the low accuracy diagnosis of metastases and non-metastases tumid lymph nodes in the lung cancer N stage with chest CT images, effective image features of lymph nodes is extracted for quickly and accurately differentiating metastases and non-metastases tumid lymph nodes. Results of 100 experiments on lymph nodes images showed that image spatial information can effectively express the characteristics of lymph nodes. It provides an effective new method for improving the accuracy of the lung cancer N stage in medical imaging diagnosis.To solve the low accuracy diagnosis of central lung cancer and tumid lymph nodes with chest CT images, 20-dimensional texture, 100-dimensional histogram features and 100-dimensional multi-resolution histograms of lesions were extracted to be compared. Experimental results showed that the high-dimensional multi-resolution histogram can express the characteristics of lesions better, and the classification speed is faster than traditional texture methods. The classification speed and accuracy basically meet the requirements of central lung cancer and tumid lymph nodes in medical imaging diagnosis.