Isolated Convolutional-Neural-Network-Based Deep-Feature Extraction for Brain Tumor Classification Using Shallow...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Brain tumors pose a significant health risk, with advanced stages leading to poor prognoses.
• Accurate brain tumor classification is vital for effective treatment strategies and improved patient outcomes.
• Machine learning offers potential for enhancing diagnostic accuracy in neuro-oncology.

Purpose of the Study:

• To develop and validate a deep-feature-trained model for brain tumor detection and differentiation.
• To investigate the efficacy of classical/linear machine learning classifiers (MLCs) trained on deep features extracted from convolutional neural networks (CNNs).
• To compare the performance of the proposed model against established pre-trained deep-feature models.

Main Methods:

• Constructed and trained multiple CNN models with varying layers (19, 22, 25) to extract deep features from brain MRI scans.
• Employed transfer learning to utilize these deep features for training various MLCs.
• Validated the approach using available brain MRI datasets and compared performance against pre-trained models.

Main Results:

• The proposed CNN deep-feature-trained support vector machine model demonstrated superior accuracy compared to other tested models.
• The model achieved 98% accuracy in detecting and distinguishing brain tumors.
• A classification rate of 97.2% was obtained on an independent, unseen dataset.

Conclusions:

• The developed deep-feature-trained model shows high potential for accurate brain tumor diagnosis.
• This approach can serve as a valuable tool to assist clinicians in the diagnostic process.
• The findings suggest a promising application of AI in improving neuro-oncological diagnostics.