Facilitating Neuro-Oncology Research: An Extensible Graph Neural Network Framework for Brain Tumour Classification

Abstract

This study addresses the necessity for precise and efficient brain tumour classification techniques, traditionally dependent on manual histopathology, a process prone to inaccuracies due to subtle differences in images. Edge-definition techniques in cell-graphs play a fundamental role in graph-based learning, as they encode the interaction between the tumours’ cells that can be crucial for capturing complex histopathological patterns. These patterns, when accurately identified, can provide valuable insights into tumour structure and potential malignancy, therefore enhancing the precision of cancer diagnosis and prognosis. The potential of Graph Neural Networks (GNNs) is further explored within this context. Recognizing the diversity and complexity of brain tumours, we leveraged the flexibility and extensibility of the GraphGym framework in our method, which allowed for a more comprehensive and nuanced approach to brain tumour classification. The resultant framework is used to evaluate both the performance of edge-definition approaches and the effectiveness of different GNN architectures. The objective is to identify the most effective combination for brain tumour classification. The results of this study aim to provide significant insights and make a substantial contribution to the advancement of diagnostic accuracy in neuro-oncology.