DEEP-BNET: A NOVEL ARCHITECTURE OF DEEP-BONE NETWORK LAYER BASED BONE CANCER DETECTION AND CLASSIFICATION

Abstract

There are many types of cancer, but bone cancer is one of the deadliest and the rarest. Although its incidence is low, it has increased over the past years. An early diagnosis of bone cancer is important as it prevents the spread of cancerous cells and reduces mortality. Classical manual detection techniques are very tedious and require expert knowledge. To address these issues, the Deep-BNET model, a deep learning method using VGG19 for feature extraction, is presented. The system uses a process of transfer learning by using a pre-trained convolutional neural network (CNN), which processes pre-processed input images and extracts features of importance. The aforementioned features are applied to and utilized by an ensemble classifier for either cancer or healthy bone tissue differentiation. The virtue of CNN is that its capability of image learning can become enhanced with the increase of the depth of the network. Deep-BNET in particular employs the VGG19 model for feature extraction on X-ray images. Then a mutual information-based measure accesses the correlation between the features and selects the most informative ones, a new methodology never before used in the area of bone cancer detection. These features are then categorized into malignant and benign types by the ensemble model. Performance evaluations confirm that Deep-BNET is very efficient, 98.8% accuracy and better than other detection techniques.