Mobile Menu

AI speeds up brain tumour profiling

AI speeds up brain tumour profiling‘ – Written by Charlotte Harrison, Science Writer 

A team, led by Harvard Medical School scientists, has designed an AI tool that can determine the molecular makeup of a brain tumour much quicker than current methods. Although not yet clinically validated, the tool could one day be used to molecularly diagnose a tumour and guide treatment decisions during surgery.

 “Right now, even state-of-the-art clinical practice cannot profile tumours molecularly during surgery. Our tool overcomes this challenge by extracting thus-far untapped biomedical signals from frozen pathology slides,” said study author Kun-Hsing Yu in a press release.

Overcoming problems

The current approach for evaluating tumours during surgery involves freezing brain tissue and examining it with a microscope. But freezing the tissue tends to alter the appearance of cells, which can reduce the accuracy of the testing.

Moreover, widely used criteria to classify CNS tumours use molecular profiles as part of the diagnostic categories, which cannot be determined by visually examining cryosections.  

The new AI approach, dubbed CHARM – cryosection histopathology assessment and review machine – was developed to overcome these problems. It is freely available to other researchers.

Molecular changes in glioma

The researchers focused on glioma, a common form of brain cancer that has three main forms characterised by different molecular markers and propensities for growth and spread.

CHARM was developed using brain-tumour cryosections from 1,524 people with glioma. When tested on unseen brain samples, the tool distinguished tumours that carried an isocitrate dehydrogenase (IDH) mutation from wild-type tumours and identified the most prevalent subtypes of IDH-mutant tumours. It could also classify the three major types of gliomas.

Furthermore, the tool captured the visual characteristics of the tissue surrounding the malignant cells. It was capable of spotting areas with greater cellular density and more cell death within samples, which are signs of a more aggressive glioma type.

CHARM also found clinically relevant molecular alterations in a subset of low-grade gliomas, including ATRXTP53 and CIC mutations, the CDKN2A/B homozygous deletion and the 1p/19q co-deletion.

Other brain cancers

Finally, the authors showed that CHARM could link the appearance of the cells — namely the presence of oedema around the cells and the shape of their nuclei — with the molecular profile of the tumour. The authors note that this ability to assess the broader context around the cell image makes the tool accurate and akin to how a human pathologist would visually assess a tumour sample.

The researchers highlight that CHARM is not restricted to use in glioma; future studies will be able to retrain CHARM to identify other brain cancer subtypes and update it to reflect evolving diagnostic criteria.

Image Credit: Canva