Surgical instrument segmentation based on multi-scale and multi-level feature network.

Area of Science:

• Computer-aided surgery
• Medical image analysis
• Deep learning in healthcare

Background:

• Accurate surgical instrument segmentation is vital for computer-aided surgery systems.
• Existing deep learning methods often struggle with semantic ambiguity due to limited feature extraction (either multi-scale or multi-level).

Purpose of the Study:

• To develop a novel neural network architecture for enhanced surgical instrument segmentation.
• To address the limitations of current methods by extracting both multi-scale and multi-level features simultaneously.

Main Methods:

• A U-net based neural network incorporating a cascaded and double convolutional feature pyramid.
• Introduction of a Dilation Feature-Pyramid (DFP) module in the decoder to capture multi-scale and multi-level information.
• Evaluation on two public datasets using five standard metrics.

Main Results:

• The proposed algorithm demonstrated superior performance across five evaluation metrics compared to existing methods.
• The DFP module effectively extracts comprehensive multi-scale and multi-level features, reducing semantic ambiguity.
• Experimental results validate the efficacy of the new network for surgical instrument segmentation.

Conclusions:

• The novel neural network architecture significantly advances surgical instrument segmentation accuracy.
• The integration of multi-scale and multi-level feature extraction provides a more robust solution for computer-aided surgery.
• This work offers a promising tool for improving precision and safety in surgical procedures.