PREDCGAN: An approach to synthetic lung nodule generation with the use of Pre-Training.
Early treatment and detection of lung cancer is important. However, the classification of nodules by neural convolutional network using few real computed tomography images is a difficult process. To work around this problem, this work proposes PREDCGAN. In it we add a pre-training in the pipeline of a generative adversarial network for the generation of pulmonary nodules. As a result, the use as a base increase of these synthetic images in conjunction with classical techniques to classify pulmonary nodules was found to be a value of 0.791 AUC, being the best value compared to other methods found in this work.
Generetantive Adversarial Network, PREDCGAN, Pre-Training, Lung Cancer