Samsung Medical Center Team Develops AI Model to Assess 'Liver Fibrosis' During Surgery

Prediction Accuracy Reaches 94%—Higher Than Surgeons at 84% and Blood Tests at 68%

"Provides Objective Evidence for Determining Surgical Strategies in Liver Cancer Patients"

An artificial intelligence (AI) technology has been developed that can accurately predict the degree of liver fibrosis—a key factor in determining surgical strategies for liver cancer patients—in real time during surgery. The technology has even been evaluated as more accurate than surgeons performing liver cancer operations.

Choi Kyuseong - Oh Namgi - Yoo Hakje

On May 12, Samsung Medical Center announced that the research team led by Professors Choi Kyuseong and Oh Namgi from the Department of Transplantation Surgery, along with Dr. Yoo Hakje from the AI Research Center, developed an AI model to predict liver fibrosis. The team analyzed 103 patients who underwent laparoscopic liver resection for liver cancer between December 2019 and March 2022.

For liver cancer patients, it is extremely important to assess the degree of liver fibrosis because the extent of liver resection or even the surgical method may need to be adjusted if fibrosis or cirrhosis is present.

Traditionally, preoperative blood tests or CT and MRI scans were used to assess fibrosis, but their accuracy was limited. Invasive tissue biopsies not only posed risks but also made it difficult to assess the overall degree of fibrosis in the liver. For these reasons, surgeons have typically relied on direct visual assessment of the liver surface during surgery, but this method is subjective and its accuracy varies depending on the surgeon's experience.

Analyzing liver fibrosis using videos captured during laparoscopic surgery through AI predictive models. Samsung Medical Center

The research team analyzed HD-quality videos recorded during patients' laparoscopic surgeries. They retrained a deep learning model, initially trained on ImageNet, for liver fibrosis diagnosis so that it could automatically recognize features such as the unevenness, color changes, and irregular contours of the liver surface.

As a result, the deep learning-based AI model predicted severe liver fibrosis with an accuracy of 92.7%, surpassing the 80.8–84.4% accuracy achieved by surgeons and the 68.0% accuracy of blood tests.

According to the research team, while surgeons showed a high sensitivity—over 95%—in detecting patients with liver fibrosis, their specificity in distinguishing normal livers without fibrosis was relatively low, at 61.1–67.8%, due to a tendency to make conservative judgments for patient safety. In contrast, the AI model demonstrated balanced performance, with a sensitivity of 91.8% and a specificity of 91.0%.

Professor Choi Kyuseong stated, "This study extends the clinical value of liver fibrosis assessment by leveraging AI to address blind spots. By combining surgeons' extensive clinical experience with AI, we expect to lay an important foundation for developing precise surgical strategies for cancer patients."

The results of this study were published in a recent issue of Scientific Reports (IF=3.9), an international scientific journal in the Nature portfolio.

This content was produced with the assistance of AI translation services.