This article was published on May 27, 2020

Scientists are using AI to predict which lung cancer patients will relapse

Their tool could help doctors tailor treatment plans


Scientists are using AI to predict which lung cancer patients will relapse Image by: Oregon State University

A new AI tool could predict which lung cancer patients will suffer a relapse by analyzing genetic data and pathology images.

Pathologists trained the tool to differentiate between immune cells and cancer cells in tumors. This revealed that while some parts of the tumor were packed with immune cells — which they describe as “hot” regions — others appeared completely devoid of them.

The research team, led by Dr Yinyin Yuan of London’s Institute of Cancer Research, found that patients with a lot of these “cold” regions were more likely to relapse.

After investigating the genetic make-up of the patients, they discovered that cancer cells in cold regions may have evolved more recently than those found in hot regions. They suspect this is because the tumor develops a “cloaking” mechanism to hide from the body’s natural defenses.

[Read: AI helps eliminate radiation exposure in breast cancer screening]

Their tool was able to spot how many regions with this cloaking mechanism exist within a tumor.

Dr Yuan said in a statement: 

We’ve gained new insight into how lung cancers can cloak themselves to escape the attention of the immune system – and in doing so can continue to evolve and develop. Cancer’s ability to evolve and to come back after treatment is one of the biggest challenges facing cancer researchers and doctors today.

The researchers envision doctors using the tool to predict which patients will suffer a relapse and tailor treatments to their individual needs.

Ultimately, it could improve survival rates for the disease, which currently kills over 35,000 people in the UK every year — making it the most common cause of cancer death in the country.

Get the TNW newsletter

Get the most important tech news in your inbox each week.

Also tagged with