Convolutional Neural Networks for Automated Detection and Classification of Bone Tumors in Magnetic Resonance Imaging

Today, medical imaging techniques are useful diagnostic tools in every specialty. The images are analyzed for diagnosis and treatment planning. Furthermore, medical imaging analysis is performed by specialized medical staff who, depending on work conditions tend to be subjective. Malignant bone tumors, like osteosarcoma, destroy the cortex of the bone and extend into surrounding soft tissues. So, it is important to detect and classify the bone tumor in an early stage with high accuracy. This work introduces a convolutional neural network approach along with image processing techniques to detect and classify bone magnetic resonance imaging scans into a malignant or benign tumor. Using transfer learning techniques, we compared the performance of two pre-trained CNN models VGG-16 and ResNet-50.