In a new study led by researchers of the Karolinska Institutet and SciLifeLab in Sweden, researchers have used a new approach to artificial intelligence (AI) to analyze and interpret huge data sets from tumor tissue. This could support the development of more personalized treatments for cancer patients in the future.
Healthy tissues function because cells develop different characteristics – or phenotypes – and organize themselves spacially within the organ. In the liver, for example, the position of liver cells along an artery-vein axis determines their function. And, antibody-mediated immunity in the lymph nodes can only occur if B cells move from the B cell zone to the T cell zone.
However, disease can occur when this spatial organization and cooperation between cells is disrupted. This knowledge can be used to make clinical decisions.
For example, “In cancer, tumors can be stratified by the density and distribution of cytotoxic T cells: tumors with dense and uniform T cell infiltration respond best to immune checkpoint inhibitors whereas tumors in which T cells are segregated away from cancer cells show poorer response,” the researchers wrote in their paper published in Nature Communications.
Current advancements in imaging allow the observation of tumors on a microscopic level, including the simultaneous measurement of hundreds of molecules and thousands of cells.
However, these tumor imaging techniques come with their own challenges. First, the amount of data generated is enormous. Second, with so many measurements made at the same time, researchers often struggle to find the right molecules and cells to focus on.
In their paper, the researchers provide an example: “Visualizing the spatial architecture of cellular phenotypes in a multiplex histology dataset of 15 cell types with 15 phenotypic markers in 40 samples requires surveying 9000 images…, each with 10,000 to 1,000,000 cells depending on the imaging technology.”
While AI can help with these large datasets, current AI processes remain difficult to understand and inaccessible to humans. In their study, the researchers therefore sought AI solutions in the areas of satellite imaging and community ecology.
“We realized that the interpretation of tumor images is similar to the interpretation of satellite images and that the relationships between cells in a tissue are similar to the relationships between species in ecology,” explained Jean Hausser, senior researcher at the department of cell and molecular biology, Karolinska Institutet, who led the research, in a press release.
“By combining techniques used in satellite imaging and ecology and adapting them for the analysis of tumor tissue, we have now been able to turn complex data into new insights into how cancer works.”
The researchers called the combination of AI, satellite imaging, and community ecology niche-phenotype mapping (NIPMAP). NIPMAP enables the analysis of cellular characteristics and reveals how cells are spatially driven in healthy and diseased tissues. It can provide insights into the architecture and fundamental properties of tissues.
“With our new method, we can reveal important details in tumor tissue that can determine whether a cancer treatment works or not. The long-term goal is to be able to tailor cancer treatments to individual needs and avoid unnecessary side effects,” added Hausser.
In the near future, the researchers want to test their AI-based method in clinical trials. Together with a major cancer-focused hospital in Lyon, France, the researchers want to investigate why some patients respond to cancer immunotherapy and others don’t. The researchers are also collaborating with the Mayo Clinic in the United States to discover why some breast cancer patients don’t require chemotherapy.
