Stable distance regression via spatial-frequency state space model for robot-assisted endomicroscopy
Mengyi Zhou1, Chi Xu2, Stamatia Giannarou1
1Hamlyn Centre for Robotic Surgery, Department of Surgery and Cancer, Imperial College London, London, SW7 2AZ, UK.
Summary
This study introduces a new deep learning model for probe-based confocal laser endomicroscopy (pCLE) to accurately measure probe-tissue distance. The SF-BiS4D model enhances robotic scanning precision for real-time surgical imaging.
Area of Science:
- Medical Imaging
- Robotics
- Computer Vision
Background:
- Probe-based confocal laser endomicroscopy (pCLE) offers real-time microscopic tissue visualization.
- Maintaining precise probe-tissue distance is critical for pCLE accuracy.
- Automated distance regression is needed for robotic tissue scanning with pCLE.
Purpose of the Study:
- To develop an automated method for regressing probe-tissue distance in pCLE.
- To enable precise robotic scanning for real-time intraoperative imaging.
Main Methods:
- Proposed the spatial frequency bidirectional structured state space model (SF-BiS4D) for distance regression.
- Utilized bidirectional image sequence processing and combined spatial/frequency domain analysis.
- Introduced guided trajectory planning with pseudo-distance labels for stable robotic control.
- Implemented hierarchical guided fine-tuning (GF) for efficient model inference.
Main Results:
- The SF-BiS4D model demonstrated superior accuracy and stability compared to state-of-the-art methods.
- Performance was validated both qualitatively and quantitatively on the pCLE regression dataset (PRD).
Conclusions:
- The deep learning framework significantly improves distance regression for microscopic visual servoing.
- The approach shows promise for integration into surgical procedures requiring precise real-time imaging.
