Applying image features of proximal paracancerous tissues in predicting prognosis of patients with hepatocellular...

Area of Science:

• Digital pathology
• Oncology
• Medical imaging analysis

Background:

• Hepatocellular carcinoma (HCC) prognosis prediction often overlooks the paracancerous tissue microenvironment (PTME).
• PTME plays a significant role in tumor initiation and metastasis.
• This study investigates the prognostic value of PTME image features in HCC.

Purpose of the Study:

• To identify the role of PTME image features in predicting HCC prognosis and recurrence.
• To develop a robust prognosis prediction model incorporating PTME.
• To compare the performance of PTME-based models against existing methods.

Main Methods:

• Collected whole slide images (WSIs) from 146 HCC patients (SYSM dataset).
• Manually annotated regions of interest (ROIs) in PTME and tumors.
• Developed a deep learning model for automatic WSI segmentation and prognosis prediction, validated on 225 HCC patients (TCGA-LIHC).

Main Results:

• PTME image features improved HCC prognosis prediction (C-index 0.668) compared to tumor-only features (C-index 0.648).
• Integrated PTME and tumor features achieved a C-index of 0.693.
• The deep learning model using automatically segmented PTME and tumor ROIs showed superior performance (C-index 0.665) on the TCGA-LIHC dataset compared to existing methods.
• Key PTME features included texture analysis related to immune cell infiltration and desmoplastic reaction.

Conclusions:

• Image features of PTME are critical for enhancing HCC prognosis prediction.
• PTME's immune cell infiltration and desmoplastic reaction features are key prognostic indicators.
• The developed model demonstrates the significant role of PTME in predicting HCC recurrence.