Lung Nodule Classification via Deep Transfer Learning in CT Lung Images

Lung cancer corresponds to 26% of all deaths due to cancer in 2017, accounting more than 1.5 million deaths globally. Considering this challenging situation, several computer-aided diagnosis systems have been developed to detect lung cancer at early stages, which increases the patients' survival rate. Motivated by the success of deep learning in natural and medical image classification tasks, the proposed approach aims to explore the performance of deep transfer learning for lung nodules malignancy classification. For this, convolutional neural networks (CNN), such as VGG16, VGG19, MobileNet, Xception, InceptionV3, ResNet50, Inception-ResNet-V2, DenseNet169, DenseNet201, NASNetMobile and NASNetLarge, were used as features extractors to process the Lung Image Database Consortium and Image Database Resource Initiative (LIDC/IDRI). Next, the deep features returned were classified using Naive Bayes, MultiLayer Perceptron (MLP), Support Vector Machine (SVM), Near Neighbors (KNN) and Random Forest (RF) classifiers. Additionally, to compare the classifiers performance with themselves and with other ones in literature, the evaluation metrics Accuracy (ACC), Area Under the Curve (AUC), True Positive Rate (TPR), Precision (PPV), and F1-Score were computed. Finally, the best combination of deep extractor and classifier was CNN-ResNet50 with SVM-RBF, which achieved ACC of 88.41% and AUC of 93.19%. These results are equivalent to related works, even just using a CNN pre-trained on non-medical images. For this reason, deep transfer learning proved to be a relevant strategy to extract representative imaging biomarkers for lung nodule malignancy classification in chest CT images.