Explainable Machine Learning Framework for Image Classification Problems: Case Study on Glioma Cancer Prediction.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Deep convolutional neural networks (CNNs) excel in image classification accuracy but function as "black box" models.
• Lack of interpretability in CNNs hinders trust and ethical application in critical domains like medical diagnosis.
• Existing methods for validating explanation quality are subjective and unclear.

Purpose of the Study:

• To propose an accurate and interpretable machine learning framework for image classification.
• To develop feature and explanation extraction methods that provide high-quality, human-understandable explanations.
• To establish general conditions for validating the quality of prediction explanations.

Main Methods:

• Developed a feature extraction framework to create transparent, high-level image features.
• Developed an explanation extraction framework utilizing extracted features and model internals.
• Proposed three general conditions to validate the quality of model explanations.

Main Results:

• The proposed framework achieves sufficient prediction accuracy for image classification tasks.
• The framework generates explanations that are interpretable in simple human terms.
• Demonstrated efficiency in a case study using brain tumor MRI for glioma prediction.

Conclusions:

• The developed framework offers a solution for creating accurate and interpretable machine learning models.
• Explainable AI is crucial for critical applications, particularly in healthcare.
• The proposed validation conditions offer a standardized approach to assessing explanation quality.