Explainable Deep Learning for Interpretable Brain Tumor Diagnosis from MRI Images

This paper presents a new method for analyzing medical images using neural networks to improve the interpretation and explainability of diagnostic decisions. The method combines the advantages of deep learning for accurate image analysis and boolean logic for interpretability. The obtained results demonstrate the possibility of obtaining transparent, interpretable neural network decisions. The proposed method includes the following steps: building a complex convolutional neural network (VGG-16) for image analysis; applying an attention mechanism to highlight key areas; and developing a simplified interpreted model (DRN) to understand the decisions. The results showed that VGG-16 achieved a classification accuracy of over 0.95, while DRN extracted logical rules based on the features identified by VGG-16 with an accuracy of 0.76. The DRN rule analysis allowed us to understand the impact of individual areas on the decision. The combination of accurate VGG-16 and interpreted decisions ensures high quality and transparency. The combination of convolutional network and DRN allowed to combine high accuracy with interpretability, providing medical professionals with an understanding of key features for making diagnostic decisions. The paper demonstrates that the proposed approach achieves a balance between the accuracy of an intelligent system and its interpretability, which is key to the trust and application of artificial intelligence in medicine. The results obtained pave the way for the safe and responsible implementation of AI technologies in medical practice in accordance with ethical principles and regulatory requirements.