Brain tumor segmentation based on the U-NET+⁣+ network with efficientnet encoder.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Neuro-oncology

Background:

• Brain tumors are aggressive and fatal, causing significant neurological deficits and cognitive impairment.
• Symptoms include epilepsy, headaches, and sensory loss, impacting patient quality of life.
• Accurate detection and segmentation are crucial for effective treatment planning.

Purpose of the Study:

• To develop a highly accurate method for brain tumor detection and segmentation.
• To improve upon existing segmentation models for enhanced clinical utility.

Main Methods:

• A novel U-Net++ network architecture was proposed, incorporating EfficientNet as the encoder.
• The model was optimized by removing dense skip connections to reduce computational complexity.
• Feature map connections at the same resolution were retained to preserve semantic spatial information.

Main Results:

• The proposed model achieved a Dice coefficient of 0.9180 on the Kaggle LGG brain tumor dataset.
• The adjusted U-Net++ model demonstrated efficient computation while maintaining rich feature representation.
• Comparative analysis of loss functions validated their effectiveness in segmentation.

Conclusions:

• The developed U-Net++ with EfficientNet provides an effective solution for brain tumor segmentation.
• The model's performance indicates its potential for improving diagnostic accuracy in neuro-oncology.
• Further research can explore additional optimizations and clinical validation.