Korean medical researchers have developed an artificial intelligence (AI) technology that accurately predicts the degree of liver fibrosis in real time during surgery, a critical piece of information for determining surgical strategy in liver cancer patients. The AI was even rated more accurate than surgeons who have performed more than 1,000 liver cancer operations.
Samsung Medical Center announced Wednesday that a research team led by Professors Choi Kyu-sung and Oh Nam-kee of the Department of Transplantation Surgery, together with Dr. Yoo Hak-je of the AI Research Center, developed the liver fibrosis prediction AI by analyzing 103 patients who underwent laparoscopic liver resection for liver cancer at Samsung Medical Center between December 2019 and March 2022.
Liver fibrosis is a process in which the liver is replaced by scar tissue and hardens due to chronic liver damage. If left untreated, it can progress to cirrhosis and liver cancer. It is considered a major factor that increases the risk of post-operative liver function decline and complications in liver cancer patients. Because the extent of liver resection or surgical method must be adjusted when liver cancer patients have fibrosis or cirrhosis, identifying the degree of liver fibrosis before surgery is essential. Previously, blood tests or imaging examinations such as computed tomography (CT) and magnetic resonance imaging (MRI) were used before surgery to assess the degree of fibrosis, but their accuracy was low. Tissue biopsies were not only invasive but also made it difficult to assess the fibrosis level of the entire liver. Ultimately, the best approach has been for surgeons to directly observe the liver surface during surgery, but this method has limitations as it is subjective and varies depending on experience.
The research team analyzed HD-quality video footage taken during the laparoscopic surgeries of 103 patients. The team retrained a deep learning model pre-trained on ImageNet to fit liver fibrosis diagnosis, enabling it to automatically recognize the roughness of the liver surface, color changes, and irregularities in the contour.
The results showed that the deep learning-based AI model could predict severe liver fibrosis with 92.7% accuracy. This figure is higher than the prediction accuracy of blood tests (68.0%) and even surgeons (80.8% to 84.4%).
The research team explained that surgeons showed high sensitivity of over 95% in identifying patients with liver fibrosis, but their specificity in distinguishing normal livers without fibrosis was relatively low at 61.1% to 67.8%, reflecting conservative judgment for patient safety. In contrast, the AI model recorded a sensitivity of 91.8% and a specificity of 91.0%, demonstrating balanced performance.
"This research is an achievement that expanded clinical value by using AI to address blind spots in liver fibrosis assessment," Professor Choi said. "We expect that adding AI to the rich clinical experience of surgeons will serve as an important foundation for establishing precision surgical strategies for cancer patients."
The findings were published in a recent issue of Scientific Reports, an international scientific journal in the Nature family.
