An interdisciplinary research team of the LKS Faculty of Medicine of the University of Hong Kong (HKumed), the Discovery Data Laboratory for Health (INNOHK D24H) of INNOHK and the London School of the Hygiene School and Tropical Medicine (LSHTM) has presented the first category of cancer cancer cancer cancer cancer cancer cancer cancer. precision greater than 90%. This innovative IA HKumed model promises to significantly reduce the preparation time prior to the consultation of first -line doctors in approximately 50%, aligning with the initiative of the Hkysar government to take advantage of AI technology in medical care. The findings were published in the magazine NPJ digital medicine.
Thyroid cancer is among the most frequent cancers in Hong Kong already worldwide. The precision management of the disease often depends on two systems: (1) The eighth edition of the American Cancer Committee (AJCC) or the Tumor Node Cancer Stading System (TNM) to determine the cancer stage; and (2) the Risk Classification System of the American Thyroid Association (ATA) to classify the risk of cancer. These systems are crucial to predict patient survival and guide treatment decisions. However, the manual integration of complex clinical information in these systems can be slow and lack efficiency.
The research team developed an AI assistant who takes advantage of large language models (LLM), such as Chatgpt and Deepseek, which are designed to understand and process human language, to analyze clinical documents and improve the precision and efficiency of staging and risks classification of thyroid cancer.
The model takes advantage of four llm of open source outside-line (Mistral AI), flame (finish), Gemma (Google) and Qwen (Alibaba)-to analyze free text clinical documents. The AI model was trained with open access data based in the United States with pathology reports of 50 thyroid cancer patients from the Atlas Atlas (TCGA) program, with the subsequent validation against pathology reports of 289 patients with CGA and 35 pseudo cases created by endocrine surgeons.
By combining the production of the four LLM, the team improved the general yield of the AI model, achieving a general precision of 88.5% to 100% in the ATA risks classification and 92.9% to 98.1% in the AJCC cancer staging. Compared to traditional manual documents reviews, this progress is expected to reduce the time that doctors spend on pre -consulting preparation.
Professor Joseph T Wu, Professor of Sir Kotewall in Public Health and Managing Director of INNOHK D24H in HKumed, emphasized the remarkable performance of the model. “Our model achieves more than 90% precision in the classification of AJCC cancer stages and the ATA risk category,” he said. “A significant advantage of this model is its outside line, which would allow local implementation without the need to share or load confidential information of the patient, thus providing the maximum privacy of the patient.”
‘In view of the recent Deepseek debut, we carry out more comparative tests with a “zero shooting approach” against the latest versions of Deepseek-R1 and V3-assi as GPT-4O. We were pleased to discover that our model worked along with these powerful LLM online, “added Professor Wu.
Dr. Matrix Fung Man-Him, clinical assistant professor and head of Endocrine Surgery, Department of Surgery, School of Clinical Medicine, HKumed, declared: “In addition to providing high precision in the extraction and analysis of the information of the complex pathology reports, the registers of operations and clinical notes, our model of AI also drastically reduces the preparation time of the doctors human interpretation could simultaneously provide cancer staging and stratification of clinical risk based on two internationally recognized clinical systems.
“The AI model is versatile and could easily integrate into several environments in the public and private sectors, and both local and international research and research institutes,” said Dr. Fung. ‘We are optimistic that the implementation of the real world of this AI model could improve the efficiency of frontline doctors and improve the quality of care. In addition, doctors will have more time to advise their patients.
‘In line with the strong defense of the government of adoption of AI in medical care, as evidenced by the recent launch of the LLM -based medical writing system in hospital authority, our next step is to evaluate the performance of this AI assistant with a large amount of real -world patients in the real world. Once validated, the AI model can be easily implemented in real clinical environments and hospitals to help doctors improve operational and treatment efficiency, “said Dr. Carlos Wong, an honorary associate professor in the Department of Family Medicine and Primary Care, School of Clinical Medicine, HKumed.
The study was directed by Professor Joseph Wu TSZ-KEI, Sir Robert Kotewall professor in Public Health at the School of Public Health and main managing director of INNOHK D24H; Dr. Matrix Fung Man-Him, Clinical Assistant Professor and Head of Endocrine Surgery in the Department of Surgery, Faculty of Clinical Medicine; and Dr. Carlos Wong King-Ho, Honorary Associate Professor in the Department of Family Medicine and Primary Care, School of Clinical Medicine and Director of Senior Research at Innohk D24H; All under Hkumed. The first authors were Dr. Eric Tang Ho-Man and Dr. Tingting Wu of INNOHK D24H.