Related Experiment Video
Updated: Aug 9, 2026
06:24
A Spine Robotic-Assisted Navigation System for Pedicle Screw Placement
Published on: May 11, 2020
A lightweight alignment-aware DBNet for surgical instrument code detection
Ke Yang1,2, Yun Xue1,2, Zhe Du1,2
1School of Medical Technology and Engineering, Henan University of Science and Technology, Luoyang, China.
Plos One
|August 7, 2026
Summary
This study introduces LA-DBNet, a lightweight deep learning model for accurately detecting engraved surface codes on surgical instruments. The new framework enhances traceability by improving detection accuracy and efficiency, even with challenging visual conditions.
Area of Science:
- Computer Vision
- Medical Device Technology
- Machine Learning
Background:
- Accurate detection of engraved surface codes on surgical instruments is crucial for traceability.
- Challenges include metallic reflections, motion blur, scale variation, and weak textures, hindering stable localization.
Purpose of the Study:
- To develop a lightweight and efficient deep learning framework (LA-DBNet) for detecting engraved surface codes on surgical instruments.
- To improve detection performance and robustness against common imaging challenges.
Main Methods:
- LA-DBNet utilizes MobileNetV4 with LiteFPN for reduced complexity and multi-scale feature representation.
- Incorporates a Directional Edge Collaborative Alignment (DECA) module for improved feature alignment and an Efficient Channel Attention (ECA) mechanism to enhance target features.
- Employs region-weighted consistency learning for improved robustness to degraded samples.
Main Results:
- LA-DBNet achieved an F1 score of 95.8% on a surgical instrument code dataset, outperforming DBNet by 3.6%.
- The model has 3.35 million parameters and achieves an inference speed of 33.6 FPS.
- Attained an F1 score of 86.1% on the ICDAR2015 dataset.
Conclusions:
- LA-DBNet offers improved detection performance for surgical instrument codes.
- The framework significantly reduces model size and maintains efficient inference, addressing practical detection challenges.
