Medical X-ray Image Hierarchical Classification Using a Merging and Splitting Scheme in Feature Space

Area of Science:

• Medical Imaging
• Computer Science
• Artificial Intelligence

Background:

• Medical X-ray images are produced daily in large volumes with significant variations.
• Efficient classification is crucial for searching and retrieving these images, particularly in content-based medical image retrieval (CBMIR) systems.

Purpose of the Study:

• To propose a novel hierarchical classification structure for medical X-ray images.
• To enhance the performance of CBMIR systems through improved image classification.

Main Methods:

• A hierarchical classification structure employing a merging and splitting scheme based on shape and texture features.
• Utilizing an orthogonal forward selection algorithm with Mahalanobis class separability for feature selection and reduction.
• Supervised merging and splitting applied at each level to form the hierarchical classification based on class complexity and inter-class distance.

Main Results:

• The proposed structure achieved a classification accuracy rate of 93.6% for an 18-class problem on the IMAGECLEF 2005 database.
• The hierarchical approach effectively groups similar classes initially and then refines them into distinct categories.
• Feature optimization using orthogonal forward selection improved classification performance.

Conclusions:

• The developed hierarchical classification structure is effective for medical X-ray image organization and retrieval.
• The combination of merging/splitting schemes and advanced feature selection significantly enhances classification accuracy.
• This approach offers a robust solution for managing and accessing large datasets of medical X-ray images.