Amplification Method of Lung Nodule Data Based on DCGAN Generation Algorithm

Early diagnosis of lung cancer can effectively reduce the mortality of patients. Doctors use low-dose spiral CT to detect lung nodules, which is time-consuming and prone to omissions. Deep learning has achieved good results in the field of medical image sub-processing, which can reduce the pressure of doctors to a certain extent. However, in the actual lung CT images, the images containing lung nodules account for less than 1% of the total images. The lack of data increases the difficulty of detecting lung nodules by using deep learning methods. This paper proposes an amplification method using deep convolutional anti-generation network (DCGAN) to generate lung nodule data. Compared with different amplification methods, and the effectiveness of this method is confirmed. Experiments can prove that the use of DCGAN to generate data can better solve the problems of high false positive rate and low sensitivity of lung nodule classification than the graphical data amplification mode. Compared with the existing methods, this experimental method greatly improves the accuracy, sensitivity and F1 score of lung nodule detection, and achieves good results of 99.98%, 99.15% and 99.55%, respectively.