BTdiagAI: A Web-Deployed Hybrid Framework for Brain Tumor Classification Using Optimized MRI Preprocessing and Deep...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Computational Neuroscience

Background:

• Brain tumor diagnosis using MRI is complex due to artifacts, heterogeneity, and manual evaluation variability.
• Limited comparative studies exist on MRI preprocessing techniques, hindering optimal accuracy.
• Advancements in deep learning offer potential for automated and accurate brain tumor classification.

Purpose of the Study:

• To systematically evaluate five MRI preprocessing methods for brain tumor diagnosis.
• To develop and validate a deep learning framework for automated classification of brain tumors.
• To assess the clinical applicability of the developed diagnostic tool.

Main Methods:

• Comparative analysis of five MRI preprocessing techniques: CLAHE, Nyul normalization, N4 bias correction, template registration, and White Stripe normalization.
• Development of a deep learning framework using fine-tuned InceptionV3, DenseNet121, and Xception networks, with feature concatenation and selection.
• Application of data augmentation to address class imbalance and ensure robust model training.

Main Results:

• All evaluated preprocessing methods achieved over 98% accuracy in brain tumor classification.
• CLAHE preprocessing demonstrated superior performance (99.8% on Dataset 1, 99.61% on Dataset 2) with minimal computational cost.
• The deep learning framework achieved state-of-the-art accuracy, validated on a public web platform (BTdiagAI).

Conclusions:

• Preprocessing is a critical factor for accurate MRI-based brain tumor diagnosis.
• CLAHE is recommended for its efficiency and high performance in MRI preprocessing for tumor detection.
• The developed AI framework provides a rapid, scalable, and clinically applicable solution for automated brain tumor classification.