EMBC Special Issue: SAM-HovNet: SAM-Enhanced Multi-Class Nucleus Segmentation and Classification in Hepatocellular...

Area of Science:

• Digital pathology
• Computational oncology
• Artificial intelligence in medicine

Background:

• Hepatocellular carcinoma (HCC) is a prevalent primary liver cancer with high mortality.
• Manual analysis of HCC nuclei in histopathology is time-consuming and inconsistent.
• Current deep learning models struggle with accurate nucleus segmentation and classification in liver tissue.

Purpose of the Study:

• To develop an advanced deep learning framework, SAM-HovNet, for precise multi-class nucleus segmentation and classification in HCC histopathology.
• To improve upon existing methods by integrating the SAM foundation model, Squeeze-and-Excitation modules, and multi-scale features.

Main Methods:

• Proposed SAM-HovNet, a unified framework enhancing HoverNet with SAM, Squeeze-and-Excitation modules, and multi-scale features.
• Utilized the HCCNuc dataset for training and evaluation.
• Validated generalizability on the public PanNuke dataset.

Main Results:

• SAM-HovNet achieved an image-level mean F1-score of 0.577 and a mean panoptic quality (PQ) of 0.521 on the HCCNuc dataset.
• Demonstrated superior performance and generalizability compared to existing methods on the PanNuke dataset.

Conclusions:

• SAM-HovNet offers a robust and accurate solution for automated nucleus segmentation and classification in HCC.
• The framework shows significant potential for advancing downstream nuclear characterization in liver cancer pathology.