Brain Tumor Classification and Segmentation Using Dual-Outputs for U-Net Architecture: O2U-Net

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Accurate brain tumor segmentation and classification are crucial for effective treatment planning.
• Existing deep learning models, like U-Net, show promise but can be further optimized for combined tasks.

Purpose of the Study:

• To propose a modified U-Net architecture for simultaneous segmentation and classification of brain tumors.
• To enhance diagnostic performance by integrating a classification module within the U-Net framework.

Main Methods:

• A modified U-Net architecture with an additional classification output was developed.
• Fully connected layers were utilized to classify images using features extracted during down-sampling.
• The segmentation output is generated via U-Net's standard up-sampling process.

Main Results:

• The modified U-Net achieved competitive results: 80.83% Dice coefficient, 99.34% accuracy, and 77.39% sensitivity.
• Performance was evaluated on a large dataset of 3064 MRI brain tumor images.
• Results were comparable to existing models for brain tumor analysis.

Conclusions:

• The proposed modified U-Net architecture demonstrates efficacy in combined brain tumor segmentation and classification.
• Integrating classification early in the U-Net process shows potential for improved diagnostic outcomes.
• This approach offers a promising advancement in automated brain tumor analysis using medical imaging.