When Mehul Shah, a 52-year-old teacher from Singapore, had surgery to remove his liver cancer, his doctors told him there was a good chance the disease would come back. But they could not say exactly how likely that was or where the cancer might return. Now, a new artificial intelligence tool developed by researchers in Singapore could give patients like Shah far clearer answers.
A team of scientists from the National Cancer Center Singapore, Duke-NUS Medical School, and the A*STAR Genome Institute of Singapore has built a machine-learning tool that predicts which liver cancer patients are likely to see their cancer return after surgery. The tool analyzes both genetic information from tumors and clinical data about patients, and researchers found it performed better than the standard TNM staging system, which doctors have used for decades to estimate cancer severity.
"Liver cancer frequently returns after surgery, yet we still lack effective therapies to prevent recurrence," said Pierce Chow, the study's co-senior author and a senior consultant surgeon at Singapore General Hospital.
Liver cancer is the third leading cause of cancer-related death worldwide. In Singapore specifically, it ranks as the third most common cancer death among men and fifth among women. About 72 percent of all cases globally occur in Asia, partly due to the high rates of hepatitis B virus infection in the region. Even when doctors catch the disease early and remove it surgically, the cancer still returns in roughly two out of three patients, most often within the liver itself.
To better understand why the disease comes back, the research team studied 106 patients. Sixty-eight of them, or 64.2 percent, experienced recurrence. The scientists discovered something unexpected: there are two biologically different ways liver cancer can return.
In the first pattern, called polyclonal seeding, multiple groups of cancer cells spread from the original tumor all at once. This type tends to recur within the liver and may respond better to certain immunotherapy drugs. In the second pattern, called monoclonal seeding, the returning cancer grows from just one group of cells that broke away. This pattern is linked to later recurrences and cancer spreading to distant organs like the lungs or bones.
This distinction matters enormously. Right now, clinical trials testing treatments to prevent recurrence enroll all patients the same way, without accounting for these biological differences, which has led to disappointing results. The new tool could help doctors select which patients should join which trials, making research more efficient and giving patients better odds of benefiting.
"Our discovery could change how clinical studies on adjuvant therapies are designed and executed, and provide new hope for patients," Chow said.
The findings, published in the journal Gut, represent a significant step toward personalized cancer care. Instead of a one-size-fits-all approach, doctors may soon be able to tailor follow-up care and additional treatments to each patient's specific tumor biology.