A Novel Hepatocellular Carcinoma Image Classification Method Based on Voting Ranking Random Forests.

Area of Science:

• Medical Imaging
• Computational Pathology
• Machine Learning in Oncology

Background:

• Hepatocellular carcinoma (HCC) diagnosis relies heavily on accurate pathological image analysis.
• Existing methods may lack the precision required for early and reliable HCC detection.
• The need for robust computational tools to assist pathologists is growing.

Purpose of the Study:

• To develop and evaluate a novel machine learning approach for classifying hepatocellular carcinoma (HCC) images.
• To introduce a Voting Ranking Random Forests (VRRF) method for enhanced HCC image classification.
• To identify and extract discriminative features from pathological images for improved diagnostic accuracy.

Main Methods:

• Preprocessing of hematoxylin-eosin (HE) images using bilateral filtering.
• Image segmentation to isolate cell structures for feature extraction.
• Definition and extraction of atypia, shape, fractal dimension, and texture features (LBP, GLCM, Tamura).
• Implementation of a Random Forests classifier optimized with a voting ranking strategy (VRRF).

Main Results:

• The proposed feature set demonstrated significant utility in HCC image classification.
• The VRRF method achieved good performance in distinguishing HCC from non-HCC images.
• The combination of novel features and the VRRF model proved effective for the classification task.

Conclusions:

• The developed VRRF method offers a promising tool for automated HCC image classification.
• The extracted atypia and texture features are valuable for identifying cancerous changes.
• This approach has the potential to aid in more accurate and efficient HCC diagnosis.