Advanced AI pipeline for histological classification of gliomas: a route toward the clinical implementation

Artificial intelligence (AI) is revolutionizing computational pathology by enabling cost-effective and accurate analysis of histopathological images. This thesis is focused on evaluating Multi-Attention Visual Transformer (MAViT), a technology that enables more complex and precise analysis compared to Convolutional Neural Networks (CNN). We have validated a pre-trained model using our own dataset, sourced from two distinct centres and annotated by our team, to classify adult-type gliomas into the three major histological types (glioblastoma, astrocytoma, oligodendroglioma) and to predict the MGMT promoter methylation status. We benchmarked the model against other state-of-art approaches such as Virchow2.