Texture features in the Shearlet domain for histopathological image classification.

Area of Science:

• Digital pathology and computational imaging.
• Machine learning applications in medical diagnostics.
• Biomedical signal and image processing.

Background:

• Histopathological slide image analysis is the gold standard for cancer diagnosis and staging.
• Traditional methods often rely on handcrafted texture analysis for supervised machine learning classification.
• Existing computational techniques for histopathology image analysis show varying success rates.

Purpose of the Study:

• To develop a novel feature space for automated classification of histopathological tissues.
• To integrate diverse texture features within the complex Shearlet domain for enhanced representation.
• To improve the accuracy of cancer classification in histology images using machine learning.

Main Methods:

• Constructed a new texture feature representation by integrating multiple descriptors in the complex Shearlet domain.
• Investigated the effectiveness of both magnitude and relative phase (RP) coefficients.
• Extracted four texture-based descriptors: co-occurrence features, Local Binary Patterns, Local Oriented Statistic Information Booster, and segmentation-based Fractal Texture Analysis.
• Reduced feature dimensionality using principal component analysis (PCA).
• Evaluated the proposed feature space using Support Vector Machine (SVM) and Decision Tree Bagger (DTB) classifiers.

Main Results:

• The novel feature representation achieved high performance on four public datasets.
• Achieved maximum accuracies of 92.56% (Kather), 91.73% (BreakHis), 98.04% (Epistroma), and 96.29% (Warwick-QU).
• The integrated feature set demonstrated superior classification accuracy compared to individual descriptors.

Conclusions:

• The proposed Shearlet domain method is effective for classifying histopathological images.
• The approach demonstrated robustness across datasets with varying complexity.
• This automated classification method shows significant potential for cancer diagnosis and staging.