Singapore Scientists Build AI System to Predict Liver Cancer Relapse with Striking Accuracy
Singapore researchers have unveiled a powerful new AI tool that can predict whether a liver cancer patient is likely to suffer a relapse well before it happens. The technology, known as the Tumour Immune Microenvironment Spatial (TIMES) score, is designed to assess the risk of recurrence in patients with hepatocellular carcinoma (HCC), the most common type of liver cancer.
It recently earned a spot on the cover of Nature, one of the world's leading scientific journals — a signal of its potential global impact.
The scoring system was developed by scientists at the A*STAR Institute of Molecular and Cell Biology (IMCB) and Singapore General Hospital (SGH). It blends cutting-edge spatial biology with machine learning to pick up signs of possible cancer relapse that current diagnostic methods often miss.
Here's how it works. TIMES combines multiplex immunofluorescence imaging, spatial transcriptomics, and proteomics data to study tumour samples in remarkable detail, a Singapore General Hospital article stated. Singapore General Hospital is the clinical collaborator on the project.
The system uses the XGBoost machine learning algorithm to find subtle patterns in immune cell behaviour and gene expression patterns invisible to the human eye. This allows it to predict relapse with an accuracy of around 82%, significantly better than existing clinical tools.
One of the key breakthroughs is the system's ability to map the spatial distribution of natural killer (NK) cells inside the tumour, and analyse the expression of five critical genes. Based on this, the AI flags patients who are at high risk of cancer coming back after surgery, helping doctors act faster with personalised follow-ups or treatment adjustments.
“In Singapore, up to 70% of liver cancer patients see the disease return within five years,” said Dr Joe Yeong, the study's lead investigator. Yeong is a senior figure at both A*STAR IMCB and SGH, and heads ImmunoPathology at the SingHealth Duke-NUS Pathology Academic Clinical Programme. “TIMES is a game-changer in how we detect and manage recurrence early,” he said.
The researchers also identified a key biomarker called SPON2, expressed by NK cells, which closely links to relapse risk. Experiments showed that SPON2-positive NK cells are more active against tumors—they move more efficiently toward cancer cells and help trigger a stronger response from CD8+ T-cells. This finding opens the door to new immunotherapy approaches, potentially guided by AI-driven diagnostics.
“TIMES turns routine pathology slides into powerful predictive tools,” explained Denise Goh, co-first author and Senior Research Officer at A*STAR IMCB. “Our AI doesn't just improve prediction—it helps clinicians fine-tune how and when to act, which could ultimately save lives.”
To test its reliability, the TIMES system was validated using tumour tissue from 231 patients across five hospitals. The team also launched a free online portal, where researchers can upload tumour images and get AI-generated risk assessments. The algorithm has been patented, and efforts are underway to package it into a diagnostic kit for real-world clinical use.
Further trials are planned later this year at SGH and the National Cancer Centre Singapore, with diagnostic partners working to standardise lab workflows and bring the technology into mainstream cancer care.
This breakthrough reflects the broader goals of A*STAR's Institute of Molecular and Cell Biology, which focuses on using biology to develop next-generation treatments and diagnostic tools. It also underscores A*STAR's mission to drive public sector R&D that improves lives, not just in Singapore, but globally.
health