Revolutionizing Brain Tumor Detection Using Explainable AI in MRI Images.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Brain tumor identification via MRI is challenging due to complex anatomy and tissue similarities.
• Manual tumor detection is error-prone and time-consuming, necessitating automated solutions.
• Existing AI models require further enhancement for superior accuracy and interpretability.

Purpose of the Study:

• To develop an advanced convolutional neural network (CNN) framework for improved brain tumor detection accuracy.
• To enhance the interpretability and diagnostic reliability of AI-driven neuro-oncology tools.
• To outperform existing state-of-the-art models in brain tumor identification.

Main Methods:

• An improved CNN framework based on DenseNet121 was developed.
• The model was rigorously evaluated against 12 baseline CNNs and 5 advanced architectures (ViT, ConvNeXt, MobileNetV3, FastViT, InternImage).
• Explainable AI (XAI) techniques, specifically Grad-CAM++, were integrated for enhanced interpretability and localization.

Main Results:

• The proposed model achieved superior accuracy rates of 98.4% and 99.3% on two distinct datasets.
• It outperformed all 17 evaluated models, including Vision Transformer and ConvNeXt.
• XAI successfully highlighted critical tumor areas, improving diagnosis of challenging cases like small metastatic lesions.

Conclusions:

• The developed AI model offers a significant advancement in automated brain tumor detection and localization.
• Integration with XAI enhances diagnostic transparency and reliability for healthcare practitioners.
• This approach promises improved patient outcomes through more precise and interpretable AI-driven neuro-oncology diagnostics.