3D-local oriented zigzag ternary co-occurrence fused pattern for biomedical CT image retrieval

Area of Science:

• Medical Imaging
• Computer Vision
• Image Processing

Background:

• Computed Tomography (CT) image retrieval relies on effective feature descriptors.
• Conventional 2D local pattern methods capture limited spatial information.
• 3D texture analysis is crucial for accurate CT image retrieval.

Purpose of the Study:

• To propose a novel 3D feature descriptor for enhanced CT image retrieval.
• To improve the discriminative power and efficiency of CT image retrieval systems.
• To address limitations of existing 2D local pattern-based approaches.

Main Methods:

• Introduced the three dimensional local oriented zigzag ternary co-occurrence fused pattern (3D-ZTP) descriptor.
• Employed a 3D zigzag sampling structure on multiscale Gaussian filtered CT images.
• Calculated 3D local ternary patterns and fused co-occurrence information for texture representation.

Main Results:

• The 3D-ZTP descriptor effectively captures both uniform and non-uniform texture patterns.
• Multiscale analysis via Gaussian filtering preserves fine to coarse image details.
• Experiments on NEMA and TCIA-CT databases showed superior retrieval precision and recall compared to existing methods.

Conclusions:

• The proposed 3D-ZTP descriptor significantly outperforms traditional local pattern-based methods for CT image retrieval.
• The 3D zigzag sampling and fusion scheme enhance feature distinctiveness and reduce dimensionality.
• This descriptor offers a promising advancement for medical image retrieval applications.