AI-based software to predict survival rates in brain tumour patients

15 March 2018 (Last Updated March 15th, 2018 11:44)

A new artificial intelligence (AI) based software developed by researchers at Emory and Northwestern Universities in the US can predict survival rates of patients with glioma, a type of brain tumour, by analysing tissue biopsy data.

AI-based software to predict survival rates in brain tumour patients
New AI software is trained to learn visual patterns in microscopic images of brain tumour samples. Credit: Emory University.

A new artificial intelligence (AI) based software developed by researchers at Emory and Northwestern Universities in the US can predict survival rates of patients with glioma, a type of brain tumour, by analysing tissue biopsy data.

This new technology is considered beneficial because the duration of survival of glioma patients is uncertain.

Whilst the disease can cause death within two years post-diagnosis, some patients are known to live for ten years and above.

By identifying the disease course at diagnostic stage, the AI software can help in determining suitable treatments and allow the planning of lives for both patients and their families.

A combination of genomic tests and microscopic tissue examination is currently used to predict the cancer’s clinical behaviours or treatment response.

“After training the software with images as well as genomic data, the researchers observed that it could predict patient survival duration more precisely than conventional human pathologist approaches.”

However, the reliable genomic testing does not entirely inform patient outcomes, and microscopic exams are so subjective that various pathologists often reach different conclusions for the same case.

Northwestern University Feinberg School of Medicine pathology chair Daniel Brat said: “Genomics have significantly improved how we diagnose and treat gliomas, but microscopic examination remains subjective.

“There are large opportunities for more systematic and clinically meaningful data extraction using computational approaches.”

The team used deep learning and microscopic images of brain tumour tissue samples to train the software on learning visual patterns associated with patient survival.

After training the software with images as well as genomic data, the researchers observed that it could predict patient survival duration more precisely than conventional human pathologist approaches.

Additionally, the software showed that it could learn to recognise several similar structures and patterns in tissues used by the pathologists during their examinations.

The researchers intend to conduct subsequent studies to determine if the new technology can be used to deliver better outcomes for newly diagnosed patients.