Research Of Classification Between Benign And Malignant Solitary Pulmonary Nodules Based On PET/CT

Nowadays, the incidence rate of the lung cancer is the highest mortality rate of malignant tumors in the world. Solitary pulmonary nodule (SPNs), which is a single, oval and maximum diameter, it is no more than 30 mm nodule in lung parenchyma, not accompanied by swollen lymph nodes, atelectasis and pneumonia lesions. Patients usually have no clinical symptoms, mostly founded by accident or a medical examination. Solitary pulmonary nodules differential diagnosis of benign and malignant has been a focus of clinical attention, as related to the choice of treatment and prognosis. Early resection in patients with malignant lung nodules five-year survival rate can be higher than 60%, but because the diagnosis is not clear enough, half of the removal nodules are benign. Therefore, accurate diagnosis of benign and malignant solitary pulmonary nodules is very critical.There are several imaging methods used to distinguish pulmonary nodules of benign and malignant, but the qualitative diagnosis is still difficult. CT is the conventional method, with a high sensitivity (up to 97%) and low specificity (only 70%). PET has a high accurate rate on diagnosis of pulmonary nodules, and low diagnostic sensitivity and specificity, CT diagnosis is mainly based on the location of the nodule, size, shape, edge density and adjacent pleura and vascular changes in morphological characteristics, PET is mainly the metabolic characteristics of the nodule. PET/CT which combine both of them, can acquire better diagnosis effect. But extract the CT images and PET images at the same time making the amount of image data to a sharp growing. Moreover, there are so many types of lung disease, this lead to complex changing in PET/CT images. In this case, doctors rely on strong subjectivity experience when diagnosing. Therefore, it is necessary to resort to the computer-aided diagnosis. In order to improve the accuracy of the solitary pulmonary nodule diagnosis with medical signs in medical imaging diagnostics, a novel computer-aided classification method is proposed. The method first extracted multiple key CT features of discrimination between benign and malignant pulmonary nodules by the experienced physician, and then combined with SUV values in PET images of solitary pulmonary nodules to establishment of a support vector machine classifier model, using particle swarm optimization on support vector machine parametric search and thus choose the most appropriate parameters, then get the appropriate SVM classification model.The experimental results show that, by using the improved particle swarm optimization to optimize parameters in SVM, the artificial selection randomness can be avoided. And it performs well in solving classification problems. With the input feature set extracted from our method above, the final classification model which used for classifying benign and malignant of pulmonary nodules can reach a high average accuracy (up to 85%). This method is also provides a theoretical basis for the main features selected when doctors diagnosis of pulmonary nodules.