Convolution Neural Network Based Bone Cancer Detection

Abstract

Bone cancer is an uncommon type of cancer in which cells in the bone start to grow out of control[1-4].  It destroys normal bone tissue. In a survey conducted [1] it was seen that out of 100 radiologists, 52 of them reported more than 10 cases of bone tumour per year. A benign tumour does not threaten life and will not spread to other body parts, whereas a malignant tumour can spread to other body parts. Fig.1-A depicts the 19 variants of bone cancers that occurs in the human body. Each type has unique characteristics and is seen in people of different age groups. According to Cancer Research UK [5], the survival rate for patients with bone cancer is 40%. Early detection of tumours can increase chances of survival by providing treatment at the initial stages of cancer. This paper explores various techniques of medical image processing and deep learning and applies them to detect and classify tumours into benign or malignant. Techniques used include image pre-processing using filtering methods, K-means segmentation and edge detection to detect cancerous regions in Computer Tomography (CT) images for Parosteal osteosarcoma, enchondroma and osteochondroma types of bone cancer. After segmentation of the tumour, classification of benign and cancerous cells is done with the help of a deep learning model-based Convolution neural network (CNN) classifier. The accuracy of the model is given by: Accuracy (TP+TN)/(TP+FP+FN+TN) as seen in fig.2. Table1 depicts the prediction percentage of the confusion matrix. The accuracy of the developed model is 98.6%. The tumour detected areas are displayed using a Graphical User Interface (GUI) as shown in fig.1 (B-C). This paper aims to give an overall idea about how image processing techniques and deep learning-based CNN classifier can be used to detect and classify bone cancer at earlier stages to prevent complications and fatalities.