Advanced Brain Tumor Classification in MR Images Using Transfer Learning and Pre-Trained Deep CNN Models

Area of Science:

• Artificial Intelligence in Medical Imaging
• Machine Learning for Diagnostics
• Deep Learning Applications in Radiology

Background:

• Accurate brain tumor classification is vital for effective patient treatment and outcomes.
• Automating the analysis of brain MRI images can significantly enhance diagnostic efficiency.
• This study addresses the need for reliable AI tools in medical diagnostics.

Purpose of the Study:

• To evaluate the efficacy of pre-trained deep learning models for classifying brain MRI images.
• To categorize images into four classes: Glioma, Meningioma, Pituitary, and No Tumor.
• To explore the potential of automated diagnostics in improving patient care.

Main Methods:

• Utilized a dataset of 7023 brain MRI images.
• Employed transfer learning to fine-tune state-of-the-art models (Xception, MobileNetV2, InceptionV3, ResNet50, VGG16, DenseNet121).
• Applied advanced preprocessing and data augmentation techniques for optimized performance.

Main Results:

• Xception achieved the highest performance with 98.73% weighted accuracy and 95.29% F1 score.
• Models demonstrated effectiveness in handling class imbalances and provided consistent results.
• Challenges remain in improving recall for Glioma and Meningioma, and enhancing model interpretability.

Conclusions:

• Deep learning holds significant potential for developing reliable and scalable diagnostic tools in medical imaging.
• Further research is needed to improve model explainability and validate performance in clinical settings.
• AI-driven systems can be integrated into healthcare workflows for enhanced diagnostics.