A Two-Step Approach for Classification in Alzheimer's Disease.

Area of Science:

• Medical Imaging
• Machine Learning
• Artificial Intelligence

Background:

• Deep learning models excel in medical image classification accuracy but lack transparency, hindering clinical adoption.
• The
• black box
• nature of deep learning raises concerns among medical practitioners due to the absence of explanations for their decisions.

Purpose of the Study:

• To develop and evaluate an interpretable machine learning method for medical image classification.
• To address the transparency limitations of deep learning in medical applications.
• To utilize an evolutionary algorithm for classification and explicit knowledge extraction.

Main Methods:

• A two-step approach involving image filtering to generate numerical datasets.
• Classification using an evolutionary algorithm that simultaneously classifies images and extracts IF-THEN rules.
• Application to Alzheimer's disease datasets using Magnetic Resonance Imaging (MRI) brain scans.

Main Results:

• Achieved 100% accuracy and F-score on a two-class (non-demented vs. moderate demented) MRI dataset.
• Attained 91.49% accuracy and 0.9149 F-score on a three-class (non-demented, mild, moderate demented) MRI dataset.
• Outperformed several well-known classifiers in accuracy and F-score for both classification tasks.

Conclusions:

• The proposed interpretable machine learning method demonstrates high performance in medical image classification.
• The use of evolutionary algorithms provides explicit, rule-based knowledge, enhancing model transparency.
• This approach offers a promising alternative to black box models in clinical settings, particularly for Alzheimer's disease detection.