Brain medical image diagnosis based on corners with importance-values

Area of Science:

• Medical Imaging Analysis
• Neurology
• Computer Vision

Background:

• Brain disorders are a leading cause of death, necessitating accurate diagnosis through medical image analysis.
• Current corner detection and matching methods lack domain-specific information for brain images, leading to inefficiencies.
• Existing methods are often unsuitable for 2D brain image diagnosis due to differing mechanisms and lack of structural consideration.

Purpose of the Study:

• To develop a novel corner-based method for brain medical image classification.
• To address limitations of existing methods by incorporating domain-specific information and image structure.
• To improve the accuracy and efficiency of brain disorder diagnosis using medical imaging.

Main Methods:

• Automatic extraction of multilayer texture images (MTIs) capturing diagnostic information.
• Development of a corner matching technique utilizing brain image uncertainty and structure with a bipartite graph model.
• Implementation of a similarity calculation method tailored for brain image diagnosis.

Main Results:

• The proposed classifier demonstrated superior performance on both brain CT and MRI datasets compared to existing methods.
• Achieved at least 8% higher accuracy and 2.4% higher F1-score on CT scans.
• Exceeded comparison methods by over 7.3% in accuracy and 4.9% in F1-score on MRI scans.
• The method proved robust across different brain medical image intensity ranges.

Conclusions:

• A robust corner-based brain medical image classifier has been developed.
• The novel methods for corner detection and matching enhance diagnostic information utilization.
• Experimental results confirm the proposed classifier's superiority over state-of-the-art techniques.