AI Course Project
Magnetic Resonance Imaging (MRI) of the knee uses a magnetic field, radio waves and a computer to produce detailed pictures of the structures within the knee joint. It can be used to help with the diagnosis, weakness, swelling, or evaluate pain in the joint. Determining which sequences/images are to be used for a proper diagnosis need careful research and expertise. Deep learning model like Generative Adversarial Networks (GANs) and their extensions have provided many exciting ways to tackle such problems to a great extent with various de-noising, reconstruction, segmentation, data simulation, detection and classification techniques. Also, we could use these models to alert the radiologists and physicians based on the severity of the condition thereby reducing the workload in radiology and help in providing on time diagnosis and treatment.So here we propose the use of a few deep learning models to help us with the classification of the MR images to one of the three categories:abnormalities, ACL tears and meniscal tears.Our first step wasto performthe image classification using Resnet. We then used DCGAN to generate the images and its discriminator to classify the imagesas real or fake. We were successful in classifyingthe abnormalities in axial images with good accuracy and able to train the DCGAN model with downsampled and upsampled images