VISION: View-specific integrated segmentation-classification framework for accurate brain tumor detection in MRI...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Brain tumors pose a significant global health challenge, necessitating accurate diagnostic tools.
• Current Magnetic Resonance Imaging (MRI)-based machine learning methods for brain tumor diagnosis have limitations, focusing on either segmentation or classification, but not both.

Purpose of the Study:

• To develop a novel framework, VISION (View-specific Integrated Segmentation-Classification), for enhanced brain tumor diagnosis.
• To integrate segmentation and classification processes for simultaneous tumor boundary identification and detection.

Main Methods:

• The VISION framework incorporates a View Classifier to determine MRI orientation (axial, coronal, sagittal).
• A view-specific integrated network combines a customized segmentation model with a classification header for unified diagnosis.

Main Results:

• The VISION framework achieved superior performance compared to existing state-of-the-art methods.
• Key performance metrics include a Dice score of 0.89, an Intersection over Union (IoU) of 0.87, and an F1 score of 0.98.
• The framework demonstrated competitive computational efficiency.

Conclusions:

• The VISION framework provides a robust and accurate solution for brain tumor diagnosis by integrating view classification, segmentation, and detection.
• Its high accuracy and efficiency indicate significant potential for clinical applications in improving tumor diagnosis and treatment planning.