Intelligent brain tumor detection using hybrid finetuned deep transfer features and ensemble machine learning...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Brain tumors (BTs) are severe neurological disorders with a high global mortality rate.
• Early diagnosis of BTs is crucial for improving patient survival rates.
• Detecting BTs is challenging due to their varied nature.

Purpose of the Study:

• To develop and evaluate a new system for detecting brain tumors using a combination of deep learning (DL) and machine learning (ML) techniques.
• To assess the performance of various DL feature extraction and ML ensemble classification models for BT detection from MRI scans.

Main Methods:

• Utilized DL models (Inception-V3, ResNet-50, VGG-16) for feature extraction and PCA for dimensionality reduction.
• Employed ensemble ML methods (Stacking, k-NN, Gradient Boosting, AdaBoost, MLP, SVM) for classification.
• Preprocessed MRI scans (224x224 pixels, normalized intensities, Gaussian filter) and used data augmentation (Keras Image Data Generator).

Main Results:

• The stacking ensemble model, using ResNet-50 features reduced by PCA, achieved the highest accuracy (0.957) and AUC (0.996).
• This performance significantly outperformed baseline models (p < 0.01).
• Neural networks and gradient-boosting models also demonstrated strong performance.

Conclusions:

• The developed DL and ensemble ML system is robust and reliable for brain tumor detection.
• This approach significantly improves early detection of brain tumors, aiding medical applications.
• Future research will explore multi-modal imaging to further enhance diagnostic accuracy.