Glioma Classification Using Multimodal Radiology and Histology Data

Gliomas are brain tumours with a high mortality rate. There various grades and sub-types of this tumour, the treatment procedure varies accordingly. Clinicians oncologists diagnose categorise these based on visual inspection radiology histology data. However, process can be time-consuming subjective. The computer-assisted methods help clinicians to make better faster decisions. In paper, we propose pipeline for automatic classification gliomas into three sub-types: oligodendroglioma, astrocytoma, glioblastoma, using both histopathology images. proposed approach implements distinct models radiographic histologic modalities combines them through an ensemble method. algorithm initially carries out tile-level (for histology) slice-level radiology) via deep learning method, then tile/slice-level latent features combined whole-slide whole-volume sub-type prediction. was evaluated data set provided in CPM-RadPath 2020 challenge. achieved F1-Score 0.886, Cohen’s Kappa score 0.811 Balance accuracy 0.860. ability model end-to-end diverse enables it give comparable prediction glioma tumour sub-types.