Mapping Cancer

UND School of Medicine & Health Sciences researchers use AI to reveal ‘tumor neighborhoods’ to better predict cancer behavior

“A tumor is more than just a cluster of cells,” explained Sandeep Singhal, Ph.D., associate professor in the UND School of Medicine & Health Sciences (SMHS) Department of Pathology. “It’s a complex ecosystem – a ‘microenvironment’ – where cancerous cells interact with immune cells, connective tissue, blood vessels, and other material within a human body.”

This microenvironment shapes how a tumor grows, spreads, and responds to therapy, Singhal said. And because the treatment of tumors can vary from person to person – thanks to our genetic differences – the challenge for pathologists, oncologists, and other health providers is knowing exactly where and how to target different treatments.

With the help of AI, though, Singhal and his team just made this task easier.

AI-assisted diagnosis

“We developed an AI-based way of studying the tumor microenvironment,” continued Singhal, who integrates AI and data mining with human biology to improve cancer diagnosis and treatment. “In our recent work, more aggressive tumors showed stronger biological signals concentrated in tumor rich regions and fewer immune cells nearby.

That spatial pattern can help clinicians anticipate tumor behavior and support more targeted treatment decisions.”

In other words, while interpretation by a pathologist remains the clinical gold standard in cancer diagnostics, AI is lending a hand by adding a digital “map” to tissue samples. After a sample has been collected and reviewed by a pathologist, for example, AI can assist in diagnosis by identifying and color-coding regions of a sample that may warrant closer review. This review allows the observer to literally see patterns across the culture in question: where the cancer cells are clustering, where the immune cells are most active, and where healthy tissue is still actively fighting cancer.

This additional set of “eyes” means that a tumor’s microenvironment just got much easier to see, track, and compare, resulting in more consistent measurements and clearer treatment decision-making, said Singhal.

Cancer ‘neighborhoods’

In a study published in Nature-Scientific Reports, Singhal and his colleagues described an AI-driven workflow that turns a tissue slide into a clear map of what is happening across the entire sample. The map shows where tumor cells dominate, where immune cells are present or missing, and how tissue structure changes from one region to another.

To build these maps, the research team starts with high resolution digital scans of standard pathology tissue slides. The AI then reviews the entire slide and identifies different tissue and cell “neighborhoods,” including cancer dominated regions, inflammatory or immune-rich regions, connective tissue, and necrotic areas (dead tissue).

Instead of summarizing the tumor as a single average, the approach highlights the most important regions and creates consistent measurements that can be reviewed by specialists and compared across cases, said Singhal, whose study also demonstrates how this mapping can be used to understand why certain cancers may behave differently based on prior exposures.

“This research is a first step toward creating AI-powered tools that can scan a tissue slide, highlight the most informative regions, and support faster and more consistent interpretation,” Singhal said. “Over time, this can help personalize cancer care by linking tumor biology and individual immune response to treatment decisions.”

Heavy metal(s)

For the study in question, Singhal’s team used long term exposure to toxic metals such as arsenic and cadmium because exposures to these and other heavy metals have been linked to higher cancer risk. Harder to pin down, though, is how exposure connects to what physicians and lab scientists see inside the tumor microenvironment after cancer forms.

Using AI-guided mapping, though, research teams can now examine where biologic signals concentrate in tumor neighborhoods and how immune cells are distributed around those regions, helping connect exposure-linked biology to tissue-level behavior.

A key strength of this novel framework is its adaptability. While the team demonstrated the effectiveness of AI for bladder cancer, the same concept can be applied to many cancer types and tissues – as long as tissue samples can be digitized.

The technology may also be useful when a diagnosis is uncertain or when cancer is subtle. By generating a standardized map of suspicious regions and microenvironment patterns, the system can help focus attention on areas most likely to be “involved,” said Singhal, emphasizing that his goal is supporting – not replacing – pathologists.

“Pathologists provide the expert interpretation that guides diagnosis, and this work fits into UND’s broader effort to move advanced biomedical research closer to patient care through translational collaborations,” said Singhal, a computational biologist specializing in bioinformatics and epigenomics. “This system adds a quantitative layer that can highlight spatial patterns, support consistent reporting, and help clinical teams translate what is seen on the slide into clearer treatment decisions.

“Our goal,” he concluded, “is to move AI from the research lab into hospital workflows as decision support that makes care faster, clearer, and more personalized for patients.”