An Adaptive Learning Model for Multiscale Texture Features in Polyp Classification via Computed Tomographic...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Texture analysis from computed tomography (CT) is crucial for polyp classification.
• Integrating diverse texture descriptors presents challenges due to variation and redundancy.

Purpose of the Study:

• To propose an adaptive learning model for integrating multi-scale texture features in polyp classification.
• To address challenges of feature variation and redundancy in texture-based polyp analysis.

Main Methods:

• A novel adaptive learning model employing geometric splitting and hierarchical framework for feature integration.
• Utilized traditional (random forest + support vector machine) and convolutional neural network (CNN) classifiers.
• Employed extended Haralick measures and gray-level co-occurrence matrix (GLCM) as texture features.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.925 with traditional classifiers and 0.902 with CNN.
• The proposed adaptive learning framework significantly outperformed nine established classification methods.
• Demonstrated robust performance on a dataset of 63 polyp masses (32 adenocarcinomas, 31 benign adenomas).

Conclusions:

• The adaptive learning model effectively manages feature variation and redundancy for improved polyp classification.
• The proposed method shows significant utility and feasibility for texture feature integration in medical imaging analysis.
• Outperformed comparative methods, highlighting the potential of this approach for clinical application.