Researchers at UT Southwestern Medical Center have tapped AI to help develop a way to predict which skin cancers are highly metastatic.

According to the summary of the report, which was published in Cell Systems, the results illustrate “how the application of artificial intelligence can support the identification of cellular properties that are predictive of complex phenotypes and integrated cell functions but are too subtle to be identified in the raw imagery by a human expert.”

Put more broadly, the team said study demonstrated the potential for AI-based tools to revolutionize pathology for cancer and a variety of other diseases.

“We now have a general framework that allows us to take tissue samples and predict mechanisms inside cells that drive disease, mechanisms that are currently inaccessible in any other way,” study leader and Professor and Chair of the Lyda Hill Department of Bioinformatics at UTSW Gaudenz Danuser, PhD, said in a press release.

According to Danuser, with deep learning-based methods, AI technology can distinguish differences in images invisible to the human eye. Researchers have previously recommended using AI technology to look for differences in disease characteristics to offer insight into diagnoses or guide treatment plans, but according to Danuser, the differences distinguished by AI are typically not interpretable in terms of specific cellular characters, making AI difficult for clinical use.

For their study, Danuser and his team used AI to search for differences in images of melanoma cells with both high and low metastatic potential. The team then reverse-engineered the AI findings to discover which features in the image were responsible for the differences.

“To further prove the utility of this tool, the researchers first classified the metastatic potential of cells from human melanomas that had been frozen and cultured in petri dishes for 30 years, and then implanted them into mice. Those predicted to be highly metastatic formed tumors that readily spread throughout the animals, while those predicted to have low metastatic potential spread little or not at all,” the press release stated.

Specifically, using tumor samples from seven patients, as well as available information on their disease progression, the researchers filmed a video of about 12,000 random cells living in petri dishes, which generated around 1,700,000 raw images. The research team then used an AI algorithm to find 56 different abstract numerical features from the images.

The researchers found one feature that accurately differentiated between cells with high and low metastatic potential, and by manipulating the abstract numerical feature, the researchers created artificial images that exaggerated visible characteristics inherent to metastasis undetectable by the human eye.

Photo by thodonal/Getty Images