Automated Technique for Brain Tumor Detection From Magnetic Resonance Imaging Based on Local Features, Ensemble...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Delayed diagnosis of brain tumors, particularly in underdeveloped regions, leads to increased mortality and morbidity.
• Manual detection of tumors from MRI scans is time-consuming and prone to human error due to radiologist workload.
• Existing traditional methods for brain tumor detection are insufficient to address diagnostic uncertainties.

Purpose of the Study:

• To develop and evaluate novel automated techniques for detecting brain tumors from MRI scans.
• To compare the efficacy of ensemble classification and deep learning models for brain tumor identification.
• To improve the speed and accuracy of brain tumor diagnosis.

Main Methods:

• Proposed two novel automated techniques: ensemble classification (Support Vector Machine and K-Nearest Neighbors) and the YOLOv3 deep learning model.
• Utilized a dataset comprising open-source data and images collected from hospitals in Lahore, Pakistan.
• YOLOv3 was employed for both detection and outlining of tumor locations within MRI images.

Main Results:

• The ensemble classifier achieved an overall accuracy of 80.50%.
• The YOLOv3 model demonstrated superior performance with 97.80% accuracy, 97.40% precision, and 98.18% recall.
• YOLOv3 achieved a mean Intersection over Union (IoU) score of 0.65 for tumor localization.

Conclusions:

• The YOLOv3 deep learning model is a highly effective and accurate technique for automated brain tumor detection from MRI scans.
• Automated methods, particularly YOLOv3, offer a significant improvement over traditional approaches, addressing limitations in speed and accuracy.
• This research highlights the potential of AI in revolutionizing neuro-oncology diagnostics and improving patient outcomes.