Abstract: International audience
Introduction
Optical Coherence Tomography (OCT) has emerged as a transformative technology for gastrointestinal (GI) diagnostics, often referred to as an "optical biopsy." By providing high-resolution, cross-sectional imaging of tissue layers, it allows clinicians to identify early-stage malignancies without the need for invasive physical tissue removal. However, the clinical utility of OCT in the GI tract is limited by its narrow field of view and shallow depth of focus.
The human gut is a dynamic and deformable environment, characterized by constant motion from respiration and peristalsis. Maintaining the precise distance and contact force required for high-quality imaging is a significant challenge for human operators. This paper presents an autonomous robotic control system designed to stabilize OCT scanning by treating the imaging probe itself as a tactile sensor.
Robotic Control Strategy
The researchers utilized the STRAS robotic system, a tendon-driven continuum endoscopic robot, integrated with a steerable and compliant OCT instrument arm. The system employs a hybrid actuation strategy that coordinates the fine, high-speed adjustments of the OCT arm with the larger displacements of the main endoscope body.
To manage the complex geometry of the colon, the system implements a "multi-pullback" scanning strategy. Instead of traditional sweeping motions, the robot executes multiple parallel translational sweeps, which provides a more robust coverage of the cylindrical lumen while maintaining stable contact.
Image-Based Tactile Sensing
A core innovation of this work is the "image-as-sensor" approach. Rather than relying on bulky distal force sensors, the system derives tactile feedback directly from the OCT images:
- Preprocessing: Real-time correction of "Non-Uniform Rotational Distortion" (NURD) is performed using a Convolutional Neural Network (CNN).
- Segmentation: A dedicated network (ACE-Net) identifies the probe's sheath and the tissue surface at approximately 8 Hz.
- Tactile Proxies: The system calculates a "contact ratio" (the proportion of the probe circumference touching tissue) and the "minimum distance" to the surface. These metrics serve as proxies for interaction force.
This feedback is fed into a Proportional-Derivative (PD) controller with an adaptive leaky integrator, allowing the robot to maintain a target contact area across tissues of varying stiffness.
Experimental Validation and Results
The system was validated using custom-engineered mechanical and optical phantoms designed to mimic the stiffness (26–560 kPa) and layered structure of human colonic mucosa. Preliminary in-vivo tests in a swine model further verified the visual realism of these models.
Key performance results include:
- Force Regulation: The system consistently maintained interaction forces at a safe level of approximately 50 mN. In comparison, traditional fixed-angle approaches often resulted in force spikes exceeding 200 mN, which could pose a risk of tissue damage.
- Visibility and Robustness: The controller achieved a visibility rate of >93% during complex scanning tasks. It successfully tracked tissue moving at speeds up to 14 mm/s over a 30 mm range, effectively compensating for typical human physiological motion.
- Subsurface Quality: Qualitative analysis demonstrated that the system could clearly resolve the boundary between the mucosa and submucosa layers even during active phantom motion, a task where non-autonomous robotic approaches failed.
Conclusion
By closing the loop between deep-learning-based image analysis and robotic actuation, this system enables stable, high-resolution OCT scanning in dynamic environments. While future work is needed to address more complex anatomical geometries like haustra and polyps, this research represents a significant milestone toward fully autonomous optical biopsies. Integrating this system with standard White Light Endoscopy (WLE) could eventually provide a comprehensive "global map" to guide targeted, autonomous microscopic inspections.
Summary updated