Deep Feature Transfer Learning in Combination with Traditional Features Predicts Survival among Patients with Lung Adenocarcinoma

Abstract

Lung cancer is the most common cause of cancer-related deaths in the USA. It can be detected and diagnosed using computed tomography images. For an automated classifier, identifying predictive features from medical images is a key concern. Deep feature extraction using pretrained convolutional neural networks (CNNs) has recently been successfully applied in some image domains. Here, we applied a pretrained CNN to extract deep features from 40 computed tomography images, with contrast, of non-small cell adenocarcinoma lung cancer, and combined deep features with traditional image features and trained classifiers to predict short- and long-term survivors. We experimented with several pretrained CNNs and several feature selection strategies. The best previously reported accuracy when using traditional quantitative features was 77.5% (area under the curve [AUC], 0.712), which was achieved by a decision tree classifier. The best reported accuracy from transfer learning and deep features was 77.5% (AUC, 0.713) using a decision tree classifier. When extracted deep neural network features were combined with traditional quantitative features, we obtained an accuracy of 90% (AUC, 0.935) with the 5 best post-rectified linear unit features extracted from a vgg-f pretrained CNN and the 5 best traditional features. The best results were achieved with the symmetric uncertainty feature ranking algorithm followed by a random forests classifier.

Deep Transfer Learning and Radiomics Feature Prediction of Survival of Patients with High-Grade Gliomas

“…Previous literature on quantitative image-based prediction of OS has suggested that deep features play a complementary role to radiomics features. 12,14,22,23 Our study demonstrates that this remains true when using deep features extracted by VGG-19, an advanced model with documented excellent performance in image classification. Deep features are not limited to previously identified image attributes or even to those understandable by humans.…”

“…Under this circumstance, the transfer learning technique is introduced to apply CNN models from one field to another. For example, Paul et al 14,22 generated deep features from a pretrained CNN model to improve survival prediction accuracy for patients with lung cancer. Lao et al 12 adopted the pretrained CNN_S model for prediction of survival in GBM.…”

“…This practice, called "transfer learning," has been shown to be an effective way of using deep learning in many cases. 12,14,15 In this study, we integrated deep transfer learning and traditional radiomics techniques to explore a very large number of features in brain MR imaging of patients with HGGs. We then classified patients into longer term and shorter term survivors by training a machine learning model to predict OS on the basis of these image features.…”

Artificial intelligence in radiology

“…Given the growing number of applications of deep learning in medical imaging 14 , several efforts have compared deep learning methods with their predefined feature-based counterparts and have reported substantial performance improvements with deep learning 34,35 . Studies have also shown that deep learning technologies are on par with radiologists’ performance for both detection 36 and segmentation 37 tasks in ultrasonography and MRI, respectively.…”

Application of Deep Learning in Automated Analysis of Molecular Images in Cancer: A Survey

“…Liu et al [ 52 ] applied the CNN-F architecture [ 53 ] pretrained on ILSVRC-2012 with the ImageNet dataset for predicting survival time based on brain MR images, achieving the highest accuracy of 95.45%. Paul et al [ 54 ] predicted short- and long-term survivors of non-small-cell adenocarcinoma lung cancer on contrast CT images, with 5 postrectified linear unit features extracted from a VGG-F pretrained CNN and 5 traditional features. They obtained an accuracy of 90% and AUC of 0.935.…”