Local Wavelet Pattern: A New Feature Descriptor for Image Retrieval in Medical CT Databases

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Content-based image retrieval (CBIR) is crucial for medical imaging.
• Existing local image descriptors like Local Binary Patterns (LBP) have limitations in capturing complex texture information in CT images.
• There is a need for more robust and accurate feature descriptors for medical CT image analysis.

Purpose of the Study:

• To propose a new image feature description method, the Local Wavelet Pattern (LWP), for content-based computer tomography (CT) image retrieval.
• To enhance the characterization of medical CT images for improved retrieval accuracy.
• To introduce a novel approach for encoding local neighboring information using wavelet decomposition.

Main Methods:

• The Local Wavelet Pattern (LWP) is derived for each pixel by analyzing the relationship between the center pixel and its local neighbors using local wavelet decomposition.
• A center pixel transformation scheme is employed to normalize value ranges for accurate comparison.
• The method utilizes the relationships among neighboring pixels via wavelet decomposition before considering the center pixel.

Main Results:

• The proposed LWP descriptor demonstrated superior performance in terms of precision and recall over three CT image databases.
• Experimental results indicate that the LWP method outperforms other state-of-the-art local image descriptors for CT image retrieval.
• The centrally symmetric wavelet decomposition scheme proved suitable for CT image characteristics.

Conclusions:

• The Local Wavelet Pattern (LWP) offers a significant advancement in feature description for medical CT image retrieval.
• The novel encoding of local neighboring information via wavelet decomposition enhances retrieval accuracy.
• The LWP method provides a robust and effective solution for content-based retrieval of medical CT images.