Accurate and real-time brain tumour detection and classification using optimized YOLOv5 architecture

Area of Science:

• Medical Imaging and Radiology
• Artificial Intelligence in Medicine
• Neuro-oncology

Background:

• Accurate diagnosis and staging of brain tumors are critical for patient management.
• Image segmentation is vital in medical imaging for surgical simulation, diagnosis, and analysis.
• Existing methods for brain tumor identification and classification require enhancement for improved accuracy.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework for brain tumor prediction and classification using MRI.
• To integrate classification and localization models for enhanced diagnostic performance.
• To improve the accuracy of brain tumor identification and categorization.

Main Methods:

• A combined framework utilizing Fully Convolutional Neural Network (FCNN) for classification and You Only Look Once version 5 (YOLOv5) for detection and segmentation was proposed.
• The FCNN model was trained to classify tumors into four categories: benign - glial, adenomas, pituitary related, and meningeal.
• The YOLOv5 architecture was employed for accurate tumor localization, followed by FCNN for segmentation mask generation.

Main Results:

• The proposed integrated model achieved an average accuracy of 98.80% in identifying and categorizing brain tumors.
• The system demonstrated superior performance compared to existing methods in terms of precision, recall, F1 score, specificity, and accuracy.
• The integration of detection and segmentation models significantly enhanced the diagnostic capabilities.

Conclusions:

• The developed deep learning approach offers a highly accurate and effective method for brain tumor diagnosis using MRI.
• Advancements in deep learning structures can substantially improve tumor diagnosis and clinical management.
• This integrated detection and segmentation technique represents a valuable contribution to the field of medical imaging.