Discovering Genetic Subtype of Glioblastoma from MRI Scans using Sparse Autoencoders

Glioblastoma is a highly aggressive and deadly form of brain cancer. It is characterized by the presence of multiple genetic subtypes, each with its own unique set of molecular and clinical features. These subtypes can have a significant impact on patient prognosis and treatment response, making it important to accurately identify and classify them. Magnetic resonance imaging (MRI) images are a widely used modality for detecting and staging brain tumors, including glioblastoma. The genetic analysis using imaging has not been widely explored due to glioblastoma’s heterogeneity and aggressive growth patterns. The heterogeneous appearance of MGMT signatures in an MRI image is difficult to characterize using MRI-specific features, as the presence of MGMT is associated with specific tumoral locations and enhancement patterns. In this work, we propose a method for discovering genetic subtypes of glioblastoma from MRI images using sparse autoencoders. Sparse autoencoders are a type of unsupervised deep learning algorithms that can learn to represent high-dimensional data in a lower-dimensional space while preserving important features and patterns. We use an autoencoder algorithm to learn a compact representation of the MR images, and then apply sparsity parameters to classify the images into different genetic subtypes based on their representation. We evaluate the performance of our method using a large dataset of MRI images from glioblastoma patients and show that it can accurately identify and classify the different genetic subtypes of the disease. Our findings suggest that the proposed approach has the potential to improve the diagnosis and treatment of glioblastoma by providing a more nuanced understanding of the underlying biology of the tumor, which ultimately can provide more personalized and targeted treatment options for patients.