BFNet: a full-encoder skip connect way for medical image segmentation.

Area of Science:

• Deep Learning
• Medical Image Analysis
• Computer Vision

Background:

• Convolutional Neural Networks (CNNs) and U-shaped architectures are pivotal in medical image segmentation.
• Existing models like U-Net and U-Net++ have limitations in fully utilizing encoder information and can increase model complexity.

Purpose of the Study:

• To introduce BFNet, a novel deep learning model for medical image segmentation.
• To enhance the utilization of encoder feature maps within the decoder layers.
• To improve segmentation accuracy and reduce model parameters compared to existing architectures.

Main Methods:

• Proposed BFNet architecture that utilizes all encoder feature maps at each decoder layer.
• Reconnecting encoder and decoder layers to improve positional and boundary information learning.
• Evaluation of BFNet's performance on a custom dataset.

Main Results:

• BFNet achieved a significant 1.4 percent improvement in segmentation accuracy.
• The proposed model demonstrated a reduction in network parameters.
• BFNet effectively enhances learning of positional information, boundary details, and abstract semantics.

Conclusions:

• BFNet offers an effective solution for medical image segmentation by optimizing feature map utilization.
• The novel architecture improves accuracy and parameter efficiency.
• Further research can explore different loss functions and potential model enhancements.