Transparent brain tumor detection using DenseNet169 and LIME.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Deep Learning for Medical Diagnosis

Background:

• Brain tumor classification is vital for diagnosis and treatment planning.
• Existing deep learning models require improvement in performance and interpretability.
• Medical imaging analysis demands accurate and reliable automated tools.

Purpose of the Study:

• To propose DenseNet169-LIME-TumorNet, a deep learning model for enhanced brain tumor classification.
• To improve classification performance and model interpretability using DenseNet169 and LIME.
• To evaluate the model's efficiency and practical utility in clinical settings.

Main Methods:

• Development of DenseNet169-LIME-TumorNet, integrating DenseNet169 with LIME (Local Interpretable Model-agnostic Explanations).
• Training and evaluation on the Brain Tumor MRI Dataset (2,870 images, three tumor types).
• Comparative analysis against established deep learning architectures (Inception V3, ResNet50, etc.).

Main Results:

• Achieved a classification accuracy of 98.78% on brain tumor classification.
• Outperformed several widely used deep learning architectures in accuracy.
• Demonstrated minimal computational overhead for efficient deployment.

Conclusions:

• DenseNet169-LIME-TumorNet offers superior performance and interpretability for brain tumor classification.
• The model's efficiency and transparency support clinical decision-making and real-time diagnostics.
• Future work should focus on multi-modal learning and real-time AI-assisted diagnostic applications.