From Raw Pixels to Recurrence Image for Deep Learning of Benign and Malignant Mediastinal Lymph Nodes on Computed Tomography

Lung cancer causes the most deaths worldwide and has one of lowest five-year survival rates all types. It is reported that more than half patients with lung die within year being diagnosed. Because mediastinal lymph node status important factor for treatment prognosis cancer, aim this study to improve predictive value in assessing computed tomography (CT) lymph-node malignancy primary cancer. This paper introduces a new method creating pseudo-labeled images CT regions nodes by using concept recurrence analysis nonlinear dynamics transfer learning. Pseudo-labeled original are used as input into deep-learning models. Three popular pretrained convolutional neural networks (AlexNet, SqueezeNet, DenseNet-201) were implementation proposed classification benign malignant public database. In comparison use data, results show high performance transformed task classification. AlexNet, DenseNet201 trained tested images. Classification accuracies areas under receiver operating characteristic curve obtained from ten-fold cross-validation 93% 0.97, 96% 0.99, 100% 1 DenseNet201, respectively. The potential differentiating on CT, may provide way studying radiology imaging.