A New Deep Hybrid Boosted and Ensemble Learning-Based Brain Tumor Analysis Using MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Brain tumor analysis is crucial for timely diagnosis and effective treatment.
• Tumor morphology in medical images presents significant analytical challenges.
• Deep learning offers potential for improved brain tumor detection and classification.

Purpose of the Study:

• To propose a novel two-phase deep learning framework for brain tumor detection and categorization in MRIs.
• To develop a deep-boosted features space and ensemble classifiers (DBFS-EC) for tumor detection.
• To create a hybrid features fusion approach for classifying different brain tumor types.

Main Methods:

• Phase 1: Deep-boosted features space and ensemble classifiers (DBFS-EC) using deep convolutional neural networks (CNNs) and machine learning (ML) classifiers.
• Phase 2: Hybrid features fusion combining static (Histogram of Gradients - HOG) and dynamic features from a novel Brain Region-Edge Net (BRAIN-RENet) CNN.
• Validation on benchmark datasets (Kaggle, Figshare) including glioma, meningioma, pituitary, and normal MRIs.

Main Results:

• The DBFS-EC detection scheme achieved high performance: 99.56% accuracy, 0.9991 precision, 0.9899 recall, 0.9945 F1-Score.
• The classification scheme using BRAIN-RENet and HOG fusion significantly outperformed state-of-the-art methods on the CE-MRI dataset.
• Classification metrics included 99.20% accuracy, 0.9906 precision, 0.9913 recall, and 0.9909 F1-Score.

Conclusions:

• The proposed two-phase deep learning framework demonstrates high efficacy in detecting and classifying brain tumors from MRIs.
• The DBFS-EC and hybrid features fusion methods offer significant improvements over existing techniques.
• This framework holds promise for enhancing the accuracy and efficiency of clinical brain tumor diagnosis.