Related Experiment Video
Updated: Dec 23, 2025
10:35
Multimodal Imaging and Spectroscopy Fiber-bundle Microendoscopy Platform for Non-invasive, In Vivo Tissue Analysis
Published on: October 17, 2016
8.2K
Multi-level feature aggregation network for instrument identification of endoscopic images
Yakui Chu1,2, Xilin Yang1,2, Heng Li1
1Beijing Engineering Research Center of Mixed Reality and Advanced Display, School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081 People's Republic of China.
Physics in Medicine and Biology
|April 29, 2020
Summary
This study introduces a novel deep learning network (MLFA-Net) for identifying surgical instruments in endoscopic images. The MLFA-Net enhances surgical scenario understanding and assistive surgery through improved instrument detection, segmentation, and pose estimation.
Area of Science:
- Computer Vision
- Medical Imaging
- Surgical Technology
Background:
- Accurate identification of surgical instruments is vital for endoscopic image-guided surgery.
- Existing methods may struggle with the complexity and variability of endoscopic surgical scenes.
Purpose of the Study:
- To propose a novel multilevel feature-aggregated deep convolutional neural network (MLFA-Net) for accurate surgical instrument identification in endoscopic images.
- To enhance localization, feature propagation, and information diversity for improved surgical instrument recognition.
Main Methods:
- Developed a multilevel feature-aggregated deep convolutional neural network (MLFA-Net).
- Incorporated a global feature augmentation layer, modified cross-channel feature interaction, and a multiview fusion branch.
- Enabled multitask instrument identification including object detection, mask segmentation, and pose estimation within a single network.
Main Results:
- Achieved high performance on laparoscopic images from the MICCAI 2017 Endoscopic Vision Challenge.
- Reported mean average precision (AP) and average recall (AR) of 79.1% and 63.2% for bounding box regression.
- Obtained AP and AR of 78.1% and 62.1% for mask segmentation, and 67.1% and 55.7% for pose estimation.
Conclusions:
- The proposed MLFA-Net effectively improves surgical instrument recognition accuracy in endoscopic images.
- The method demonstrates superior performance compared to other state-of-the-art approaches.
- MLFA-Net offers a robust solution for enhancing surgical scenarios and assistive processes in image-guided surgery.
