Brain tumor classification using deep CNN features via transfer learning.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Oncology Diagnostics

Background:

• Accurate brain tumor classification is crucial for effective computer-aided diagnosis (CAD).
• Glioma, meningioma, and pituitary tumors represent common and distinct brain pathologies requiring precise differentiation.
• Existing CAD systems face challenges with limited labeled medical imaging datasets.

Purpose of the Study:

• To develop and evaluate a deep transfer learning system for classifying three primary brain tumor types: glioma, meningioma, and pituitary tumors.
• To assess the performance of the proposed system against state-of-the-art methods using a publicly available MRI dataset.
• To investigate the efficacy of transfer learning in scenarios with restricted training data.

Main Methods:

• Utilized a pre-trained GoogLeNet model for feature extraction from brain MRI scans.
• Integrated established classifier models to process the extracted deep features.
• Employed a patient-level five-fold cross-validation strategy for robust performance evaluation.
• Assessed performance using metrics including accuracy, AUC, precision, recall, F-score, and specificity.

Main Results:

• Achieved a mean classification accuracy of 98%, surpassing existing state-of-the-art techniques.
• Demonstrated high performance across all evaluated metrics (AUC, precision, recall, F-score, specificity).
• Showcased the system's robustness and effectiveness when trained with a reduced number of samples.

Conclusions:

• Deep transfer learning, utilizing pre-trained models like GoogLeNet, is a highly effective approach for brain tumor classification.
• The proposed system offers a significant advancement in CAD for neuro-oncology, providing high accuracy and reliability.
• Transfer learning presents a viable solution for overcoming data scarcity challenges in medical image analysis.