Artificial-intelligence-based Molecular Classification Of Diffuse Gliomas Using Rapid, Label-free Optical Imaging

This study aimed to develop a rapid molecular diagnostic screening tool for classifying adult-type diffuse gliomas into the taxonomy defined by the WHO CNS5 using clinical SRH and deep-learning-based computer vision methods. The results showed that key molecular diagnostic mutations produce learnable spectroscopic, cytologic and histoarchitectural changes in SRH images that allow for accurate molecular classification. The trained diagnostic system, DeepGlioma, was able to robustly and reproducibly screen fresh diffuse gliomas and accurately classify them into the WHO CNS5 taxonomy.