| Abstract [eng] |
Our work focused on using artificial intelligence (AI) methods to predict outcomes for patients with hepatocellular carcinoma (HCC), the most common form of primary liver cancer. We studied 134 HCC patients who underwent either liver resection or transplantation at Vilnius University Hospital Santaros Klinikos between 2007 and 2020. We developed three complementary AI approaches to analyze the tissue samples. First, we examined the spatial distribution of immune cells (CD8 lymphocytes) at the tumor and surrounding liver interfaces. We found that irregular density of these cells at the tumor edge was associated with better patient survival after liver resection. Interestingly, in transplanted patients, immune cell density patterns in the peritumoral liver were more informative than those within the tumor interface. Second, we applied machine learning to assess the microarchitecture of collagen fibers. The certain arrangement of these fibers, particularly their high lacunarity at the tumor edge, was linked to poorer prognosis. Third, we developed an AI system to identify distinct tissue patterns (histological fingerprints) associated with patient survival. By combining these computational biomarkers with conventional clinical factors, we improved prediction models for both HCC recurrence and overall survival. Our research contributes to improved risk stratification for precision medicine in patients with liver cancer, although larger studies are needed to validate our findings. This work also highlights the potential of machine learning to utilize subvisual spatial tissue pathology patterns to extract clinically relevant information. |
