A Novel Lung Tumor Classification Framework Using Deep Feature Fusion
Syeda Irum Gillani, Department of Computer Sciences, University of Engineering and Technology, Taxila, Pakistan.
Farrukh Zeeshan Khan, Department of Computer Sciences, University of Engineering and Technology, Taxila, Pakistan.
Momina Masood, Department of Computer Sciences, University of Engineering and Technology, Taxila, Pakistan.
Corresponding Author:
Syeda Irum Gillani (syedairum77@gmail.com)
Abstract:
Lung cancer is one of the most frequent cancers and a leading cause of cancer deaths globally. The tumors are traditionally diagnosed by health care physicians via optical inspection which can be prone to error and time consuming due to similarity of tumors with each other. In this study, we propose a novel lung tumor classification method using deep feature fusion-based methodology. Initially, the deep features are extracted from Computer Tomography (CT) images using various Convolutional Neural Network (CNN) architectures such as ResNet18, ResNet50, AlexNet and DenseNet201. These features are then classified via Support Vector Machines, K-Nearest Neighbors, Linear Discriminant and Naïve Bayes. The highest performing independent feature vectors such as ResNet50 and DesneNet201 are the fused together to form a fused feature vector that is more discriminate and informative compared to the individual vector. The proposed vector, thus, resulted in much better accuracy than the independent vectors i.e. 95%. The performance of the proposed strategy is also assessed with existing state-of-the-art methods. Hence, the results indicate the robustness of proposed framework for lung tumor classification and can be effectively utilized by health care physicians to diagnose the disease at earliest.
Keywords:
Lung Cancer; Deep Learning; Machine Learning; Feature Fusion; Medical Image Diagnosis