Classification of cardiovascular tissues using LBP based descriptors and a cascade SVM

Area of Science:

• Histopathology
• Biomedical Image Analysis
• Computational Biology

Background:

• Histological images possess distinct features like texture, shape, and color for tissue differentiation.
• Texture analysis is a key feature for distinguishing between normal tissues in histology.
• Automatic classification of non-pathological tissues remains a challenge in digital pathology.

Purpose of the Study:

• To develop an automated method for classifying cardiovascular tissues using texture analysis.
• To evaluate the feasibility of recognizing cardiovascular organs through texture-based classification.
• To identify optimal texture descriptors and machine learning models for accurate tissue classification.

Main Methods:

• Utilized texture descriptors including Local Binary Patterns (LBP), LBP Rotation Invariant (LBPri), and Haralick features.
• Employed Support Vector Machines (SVM) with linear and polynomial kernels for classification.
• Compared SVM performance against Random Forest and Linear Discriminant Analysis.
• Validated the approach using 3000 blocks of 100x100 pixels across four tissue classes with 10-fold cross-validation.

Main Results:

• A combination of LBP and LBPri texture descriptors with a linear SVM achieved an Area Under the Curve (AUC) > 0.98 for classifying four main tissue types.
• Further refinement using a polynomial kernel improved AUC > 0.98 for differentiating elastic arteries and veins.
• The selected SVM classifier outperformed Random Forest and Linear Discriminant Analysis.

Conclusions:

• The proposed texture analysis and SVM approach enables highly precise automatic classification of cardiovascular tissues.
• The method successfully differentiates fundamental tissues of the cardiovascular system, including heart muscle, arteries, and veins.
• This technique offers a robust solution for automated histological analysis of cardiovascular structures.