Histogram-Based Features Selection and Volume of Interest Ranking for Brain PET Image Classification

Area of Science:

• Medical Imaging
• Neuroscience
• Artificial Intelligence

Background:

• Positron emission tomography (PET) is crucial for diagnosing neurodegenerative diseases like Alzheimer's disease (AD).
• Computer-aided diagnosis using medical image analysis aids in the quantitative evaluation of brain diseases.
• Accurate identification of affected brain regions is vital for early AD diagnosis.

Purpose of the Study:

• To present a novel method for ranking the effectiveness of brain volumes of interest (VOIs) in differentiating Alzheimer's disease (AD) patients from healthy controls using PET images.
• To develop a quantitative approach for selecting the most informative brain regions for AD diagnosis.
• To enhance the diagnostic performance of computer-aided diagnosis systems for neurodegenerative diseases.

Main Methods:

• Brain PET images were mapped to anatomical volumes of interest (VOIs) using an atlas.
• Histogram-based features were extracted from VOIs.
• The area under the curve (AUC) was employed to select and rank VOIs based on their ability to discriminate between healthy and AD subjects.
• Top-ranked VOIs were utilized as input for a support vector machine classifier.

Main Results:

• The developed method successfully ranked brain VOIs based on their diagnostic effectiveness for Alzheimer's disease.
• The area under the curve (AUC) parameter proved effective in creating a hierarchy of VOI discriminatory power.
• Utilizing AUC for VOI selection outperformed traditional feature selection methods in classification accuracy for two-group separation.

Conclusions:

• The proposed AUC-based VOI ranking method enhances the classification accuracy for distinguishing Alzheimer's disease from healthy controls using PET imaging.
• This approach offers a more effective strategy for identifying critical brain regions in AD diagnosis.
• The findings support the integration of advanced image analysis techniques for improved neurodegenerative disease diagnosis.