A new artificial intelligence system can examine a standard tumor biopsy and determine whether immunotherapy will be effective for a given patient, even when the AI has never encountered that specific cancer type or drug before. Developed by researchers at Memorial Sloan Kettering Cancer Center and Weill Cornell Medicine, the tool, named SCORPIO, relies solely on a widely available piece of medical data: a routine H&E-stained slide of a tumor sample.
The system functions as a kind of universal translator for cancer, learning the visual language of how tumors interact with the immune system and then applying that knowledge across cancer types and treatments outside its original training. Immunotherapy has become a cornerstone of cancer care, but it works for only a subset of patients.
Doctors currently have limited ways to predict who will benefit from these treatments, leaving many to endure side effects without guarantee of response. SCORPIO is designed to address that gap by offering a fast, affordable test that could work across the board.
How SCORPIO Works
SCORPIO analyzes the spatial arrangement of cells on a biopsy slide. Rather than simply identifying which cells are present, the AI examines how those cells are organized relative to one another, picking up patterns that are invisible to the human eye.
For example, a tumor with immune cells clustered at its edge might respond differently than one where immune cells have infiltrated the core. This approach marks a departure from earlier AI models in medicine, which are typically trained on a single cancer type and a single drug. Such narrow models break when either variable changes, a brittleness that has hindered widespread clinical adoption.
SCORPIO, by contrast, was built to be generalizable. Researchers trained it on a diverse set of tumors and treatments and then tested it on entirely new combinations.
It still worked. Dr. Travis Hollmann, a pathologist at Memorial Sloan Kettering and co-senior author of the study, described the tool’s potential in concrete terms. “This could help identify patients who are most likely to benefit from immunotherapy, potentially sparing others from unnecessary side effects and allowing them to pursue alternative treatments sooner,” he said.
From Research to Clinical Reality
The study, published in Nature Cancer, tested SCORPIO on more than 5,000 patients across 24 cancer types and 10 different immunotherapy regimens. The tool consistently identified which patients would respond and, crucially, which would have better overall survival. Researchers have made SCORPIO’s code publicly available so other institutions can test and validate it.
They caution, however, that the tool needs prospective clinical trials before it can be used in routine patient care. For patients, the practical upshot is significant.
A generalizable predictive system like SCORPIO would not require separate AI models for each cancer type or each drug. A single tool, trained on a broad range of examples, could be deployed even in hospitals that lack the resources to build customized artificial intelligence. This could bring consistent, evidence-based predictions from the lab to the clinic, reducing unnecessary treatments and enabling more informed decisions.
The Shift Toward Generalizable AI in Medicine
The development of SCORPIO reflects a broader shift in medical artificial intelligence. For decades, AI has been used in medicine, but early systems were rule-based and relied on explicit instructions from human experts. More recently, deep learning allowed computers to find patterns in vast amounts of visual data.
Yet most medical AI tools have remained narrowly focused on one disease, one drug, or one type of scan, making them unreliable when applied to new scenarios. This brittleness has been a barrier to clinical adoption.
A generalizable AI model, such as SCORPIO, extracts underlying principles that hold across different contexts. It does not need to see every possible scenario in advance. For cancer care, where immunotherapy does not work for everyone and doctors have limited predictive tools, such a system could change how treatment decisions are made.
It also addresses the deeper challenge of bridging research and real-world practice, where many AI models perform well in controlled studies but fail when faced with data from a different hospital or patient population. SCORPIO, if validated in prospective trials, could help close that gap.
Patients and their doctors should consult a medical professional for advice on treatment options.
