Virtual Booth

Osteosarcoma is the most common type of bone cancer and usually affects children and young adults during periods of rapid growth. It often starts near the ends of long bones, such as around the knee or near the shoulder. Common symptoms include pain, swelling, and trouble moving the nearby joint. Accurate diagnosis, usually relying on Medical Resonance Imaging (MRI), is vital for treatment planning (surgery and chemotherapy). However, manual interpretation of MRI images by experts is time-consuming, subjective, and risks human error. To solve this, recent advances in deep learning, a type of artificial intelligence, show great potential for improving diagnostic accuracy. This study explores the effectiveness of three convolutional neural network (CNN) architectures which are ResNet50, VGG16, and Xception in classifying osteosarcoma MRI T1 sequence, axial plane images into Normal, Tumor, and Tumor with Inhomogeneity classes. To improve classification accuracy, MRI images are preprocessed to enhance image quality and highlight relevant features, thereby improving the learning performance of the deep learning models. Each model is trained and evaluated using the enhanced, labelled dataset. By analyzing the strengths and limitations of the three models, the most optimized model is selected for MRI T1 image classification. Finally, the selected model is implemented in a Graphical User Interface (GUI), enabling medical experts to input MRI images and obtain instant, objective classification results, thereby facilitating faster and more efficient clinical diagnosis.