Performance Analysis of Different Deep Learning Models for Bone Tumor Identification from MRI Images

Identification of bone tumor from MRI images faces many difficulties. For pathologists, manual classification is a complicated and time-consuming process. Typical bone tumor MRI images are classified as viable tumor, non-viable tumor, and nontumor images. However, variation within a class and similarity among classes create challenges. Therefore, development of deep learning based method for this purpose is important. This work explores convolutional neural networks (CNNs) such as VGGNet and InceptionNet for efficient classification of bone tumor MRI images. The VGG19 model achieved a classification accuracy of 91.98 % and AUC scores of 1.00,0.97, and 0.99 for non-tumor, non-viable, and viable classes, respectively.