Computational grading of hepatocellular carcinoma using multifractal feature description

Area of Science:

• Medical image analysis
• Computer-aided diagnosis
• Digital pathology

Background:

• Accurate cancer grading is crucial for diagnosis and treatment planning.
• Histopathological texture analysis is key in computer-aided diagnosis systems for cancer grading.
• Existing methods may not fully capture the complex textural characteristics of tumors.

Purpose of the Study:

• To propose a novel feature descriptor for histopathological texture analysis in cancer grading.
• To develop a robust classification model for hepatocellular carcinoma (HCC) grading.
• To compare the proposed descriptor's performance against established texture analysis methods.

Main Methods:

• Utilized fractal geometric analysis with four multifractal measures to create an eight-dimensional feature space.
• Employed a bag-of-feature classification model with feature selection for discriminating HCC images.
• Compared the proposed multifractal features against Gabor filters, LM-filters, local binary patterns, and Haralick features.

Main Results:

• The proposed multifractal feature descriptor significantly outperformed other textural feature descriptors.
• The classification model successfully discriminated non-neoplastic tissues from tumors.
• An average correct classification rate of approximately 95% was achieved for grading HCC into five classes.

Conclusions:

• Multifractal features are highly effective for describing histopathological image textures in cancer grading.
• The developed method provides a promising approach for accurate and automated hepatocellular carcinoma grading.
• This technique enhances computer-aided diagnosis systems for precision oncology.