Project Description: Detection of brain tumor subtypes is imperative to diagnosing patients and deciding between different treatment regimens. Typically, brain biopsies are the only way to officially confirm that a tumor is a specific type. If we can classify brain images as having no tumor, a benign tumor, or a malignant tumor using different neural network architectures, this could potentially reduce the need for such invasive diagnostic procedures. We will start with CNNs as our baseline architecture, build other models based on an RNN architecture, and possibly implement a variational autoencoder (VAE) model. We will present a comparative analysis of the different types of model architectures we tested.
Dataset: Brain MRI Images