Improving HCC Prognostic Models after Liver Resection by AI-Extracted Tissue Fiber Framework Analytics.

Area of Science:

• Oncology
• Medical Imaging
• Computational Pathology

Background:

• Predicting patient outcomes in hepatocellular carcinoma (HCC) remains difficult.
• Tumor properties and underlying liver conditions (e.g., cirrhosis, NAFLD) complicate prognostic accuracy.
• Existing models often lack detailed microarchitectural insights.

Purpose of the Study:

• To investigate the prognostic value of reticulin and collagen microarchitecture in HCC liver resection samples.
• To develop AI-driven models for predicting HCC patient outcomes based on tissue microarchitecture.
• To compare the efficacy of microarchitecture-based models against conventional clinicopathologic parameters.

Main Methods:

• Analysis of 105 scanned liver resection tissue sections stained for collagen and reticulin.
• Utilized a convolutional neural network (CNN) for segmentation of collagen and reticulin fibers.
• Employed hexagonal grid subsampling, automated epithelial edge detection, and computational fiber morphometry.
• Developed two penalized Cox regression models using LASSO regression.

Main Results:

• AI models incorporating microarchitectural features achieved a concordance index (C-index) greater than 0.7.
• Key prognostic features included patient age, tumor multifocality, and fiber characteristics at the epithelial edge.
• Reticulin structure was significant at the tumor edge; collagen characteristics were significant at the peritumoral liver edge.
• AI-derived models demonstrated superior prognostic performance compared to conventional models.

Conclusions:

• AI-extracted microarchitectural features of reticulin and collagen offer significant prognostic value in HCC.
• These features, particularly at the tumor and peritumoral liver edges, enhance outcome prediction.
• Integrating AI-based tissue analysis into HCC management can improve patient prognostication and treatment strategies.