A team of international researchers has developed the world’s first artificial intelligence (AI) model that can accurately determine both the stage and risk category of thyroid cancer—offering over 90% accuracy and potentially saving doctors half the time they currently spend preparing for consultations.
This major breakthrough, published in npj Digital Medicine, was led by researchers from the University of Hong Kong’s Faculty of Medicine (HKUMed), the InnoHK Laboratory of Data Discovery for Health (InnoHK D24H), and the London School of Hygiene & Tropical Medicine.
Thyroid cancer is one of the most common cancers in Hong Kong and many other parts of the world.
Doctors rely on two main systems to understand the seriousness of the disease and plan treatment. One is the AJCC staging system, which helps determine how far the cancer has spread.
The other is the American Thyroid Association (ATA) risk classification, which helps predict the likelihood of the cancer returning. However, figuring out where each patient fits in these systems usually requires careful, time-consuming review of complex medical documents like pathology reports, operation notes, and clinical summaries.
To solve this problem, the research team developed an AI tool that uses large language models (LLMs)—a type of AI trained to understand human language.
The tool analyzes free-text clinical notes and can automatically classify the stage and risk level of thyroid cancer for each patient. To do this, it combines the strengths of four top open-source LLMs: Mistral (Mistral AI), Llama (Meta), Gemma (Google), and Qwen (Alibaba).
The AI was trained using pathology reports from 50 real thyroid cancer patients through the Cancer Genome Atlas Program (TCGA) in the United States. It was then tested with data from 289 additional TCGA patients and 35 simulated cases created by thyroid surgeons.
The results were impressive. The AI reached between 88.5% and 100% accuracy in ATA risk classification, and between 92.9% and 98.1% accuracy in AJCC staging. By merging the outputs of all four LLMs, the team improved performance even further. Compared to traditional manual review by doctors, the AI assistant cut preparation time by about 50%.
Professor Joseph T. Wu, who led the project, highlighted the significance of the model’s offline design. Because it doesn’t require internet access to function, hospitals can run it on their own systems without uploading any sensitive patient data, making it safer and more private.
The AI model was also tested against newer tools like DeepSeek R1, DeepSeek V3, and even GPT-4o from OpenAI, using a method known as “zero-shot” learning—where the AI is tested on new tasks it wasn’t specifically trained for. Remarkably, the offline model performed just as well as these powerful cloud-based tools.
Dr. Matrix Fung Man-him, who is also part of the team and a thyroid surgery expert at HKUMed, said the tool’s value extends beyond just accuracy. It streamlines a process that usually requires a lot of time and effort, freeing up doctors to focus more on direct patient care and communication.
The research team believes this tool could be easily adopted across hospitals and research institutions, both in Hong Kong and worldwide. It could be useful in both public and private healthcare settings, improving efficiency and possibly helping more patients receive faster, more accurate care.
Dr. Carlos Wong, another member of the team, emphasized that the AI aligns well with the government’s current focus on expanding the use of AI in healthcare. The next step, he said, is to test the tool using large-scale real-world patient data to prepare it for routine clinical use.
In conclusion, this AI breakthrough is not just a high-tech innovation—it’s a practical tool that could help doctors diagnose and treat thyroid cancer more efficiently, with fewer delays and greater accuracy. As healthcare systems around the world look for ways to improve care while reducing costs and workload, this kind of smart technology could make a big difference.
If you care about cancer, please read studies about how to fight cancer with these anti-cancer superfoods ,and a berry that can prevent cancer, diabetes and obesity.
For more health information, please see recent studies about how to harness the power of anti-cancer foods and supplements, and cancer-fighting foods and recipes.
The research findings can be found in npj Digital Medicine.
Copyright © 2025 Knowridge Science Report. All rights reserved.
