Automated System for Lung Nodule Classification Based on ResNet50 and SVM

In this paper, a hybrid strategy is developed that combines deep learning for feature extraction with machine learning for feature categorization. The initial step starts with process for extracting the nodules from LUng Nodule Analysis 2016 (LUNA16) dataset from CT scans. Instead of full images, patches are used for feature extraction and classification. To evaluate the LUNA16 in this way, Residual Neural Network50 (ResNet50) is employed as a feature extractor. The deep features are classified using the Support Vector Machine (SVM). Further, this hybrid model is used for evaluation on test data. Performance metrics are computed based on the outcomes of the test dataset. MATLAB simulation is performed on test dataset which provides accuracy of 97.53%.