Enhancing MRI brain tumor classification: A comprehensive approach integrating real-life scenario simulation and...

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Oncology and Cancer Research

Background:

• Brain cancer mortality and incidence are rising globally, necessitating advanced diagnostic tools.
• Magnetic resonance imaging (MRI) is crucial for early brain cancer detection and treatment planning.
• Limited public clinical datasets hinder the application of deep learning in brain cancer diagnosis.

Purpose of the Study:

• To address the scarcity of diverse clinical data for deep learning models in brain cancer diagnosis.
• To evaluate and enhance the generalization capabilities of deep learning models using augmented MRI data.
• To investigate the impact of data augmentation techniques on model performance in complex, real-world scenarios.

Main Methods:

• Pre-trained deep learning models were evaluated on brain cancer MRI datasets (BT-MRI and BCD-MRI).
• Model performance was tested against synthetic datasets simulating noise, blur, and patient motion.
• Data augmentation techniques, including Gaussian noise and blur, were applied during model training.

Main Results:

• Initial models achieved high performance on standard datasets but failed on synthetic, noisy data.
• Data augmentation with Gaussian noise and Gaussian blur significantly improved model robustness.
• The refined model demonstrated enhanced generalization and performance on challenging synthetic datasets.

Conclusions:

• The careful selection of data augmentation techniques is critical for improving deep learning model generalization in brain cancer diagnosis.
• Augmenting training data with noise and blur enhances model resilience to real-world imaging complexities.
• This study highlights the importance of methodological innovation for robust AI-driven medical diagnostics.