Automated Brain Tumour Detection and Classification using Deep Features and Bayesian Optimised Classifiers.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Accurate brain tumor detection and classification are crucial for effective patient treatment.
• Manual diagnosis by radiologists can be time-consuming and requires specialized expertise.
• Developing automated diagnostic tools can enhance efficiency and accuracy in neuro-oncology.

Purpose of the Study:

• To develop a Computer-Aided Diagnosis (CAD) tool for automated brain tumor detection and classification.
• To leverage Machine Learning (ML) and Deep Learning (DL) techniques for improved diagnostic performance.
• To investigate the efficacy of feature fusion and hyperparameter optimization for enhancing CAD systems.

Main Methods:

• Utilized Magnetic Resonance Images (MRI) from a publicly available Kaggle dataset.
• Employed a pretrained Resnet18 network for deep feature extraction, combined with shallow layer features.
• Applied Bayesian Algorithm (BA) optimized ML classifiers (SVM, KNN, DT) and evaluated performance using standard metrics.

Main Results:

• The fusion of shallow and deep features, classified by a BA-optimized SVM, achieved a maximum accuracy of 99.11%.
• This approach also yielded high performance across other metrics including sensitivity (98.99%), specificity (99.22%), and F1 score (99.09%).
• Feature fusion demonstrated superior classification performance compared to using deep features alone.

Conclusions:

• The proposed framework, integrating deep feature extraction, feature fusion, and optimized ML classifiers, significantly enhances brain tumor detection and classification.
• This automated system shows potential as a valuable assistive tool for radiologists in clinical practice.
• The study highlights the effectiveness of combining advanced DL architectures with ML for medical image analysis.