Abstract
Robotic-Assisted Minimally Invasive Surgery (RAMIS) provides surgeons with enhanced precision but lacks the critical haptic feedback necessary to feel tissue resistance. Vision-based force estimation has emerged as a hardware-free alternative to physical sensors, yet it suffers from "visual uncertainty" when surgical tools occlude the camera's view of the contact point. This paper introduces V-ANFIS (Voting-Adaptive Neuro-Fuzzy Inference System), a novel framework that leverages spatial redundancy through a combinatorial voting mechanism to estimate tissue displacement under occlusion. By integrating fuzzy logic with neural networks, the system improves force estimation accuracy from 50% to 77%, providing a robust path toward haptic-enabled robotic surgery.
Introduction
In modern RAMIS, surgeons operate through tiny incisions using robotic manipulators. While these systems offer superior visualization and dexterity, the absence of force feedback—the ability to "feel" the stiffness or resistance of tissue—remains a significant limitation. This "sensory gap" increases the risk of accidental tissue trauma.
Integrating physical force sensors into surgical instruments is technically challenging due to requirements for biocompatibility, miniaturization, and sterilization. Consequently, researchers have turned to vision-based methods that infer force from observed tissue deformation. However, these methods frequently fail when the surgical tool itself blocks the camera's view of the deformation zone, creating a state of visual uncertainty.
The V-ANFIS Framework
The proposed solution, V-ANFIS, addresses visual uncertainty by exploiting the physical relationship between a contact point and its surrounding tissue. The system operates through a multi-stage pipeline:
- 3D Surface Reconstruction: A stereo-camera setup generates a 3D lattice of control points on the tissue surface, tracking their movement in real-time.
- The Voting Process: This is the core innovation. When the primary contact point is occluded by a tool, the system analyzes the displacements of immediate neighboring points. Assuming that neighboring tissue moves in a correlated fashion, a combinatorial "voting" strategy is used to reach a consensus on the true displacement of the hidden contact point.
- Neuro-Fuzzy Inference: The corrected displacement data is fed into an Adaptive Neuro-Fuzzy Inference System (ANFIS). This hybrid model combines the interpretability of fuzzy logic—ideal for handling "vague" visual data—with the learning capabilities of neural networks to map displacements to precise force values.
Experimental Validation
To validate the system, the researchers developed a high-fidelity experimental platform consisting of a Stäubli RX60B industrial robot arm, IDS stereo cameras, and artificial tissue mimicking human soft tissue. An ATI Gamma force sensor was utilized to provide ground truth measurements. This setup allowed the team to compare the V-ANFIS predictions against both standard visual tracking and actual physical force readings.
Results and Discussion
The V-ANFIS approach demonstrated a substantial performance leap over existing vision-only models. Key metrics include:
- Accuracy: Force estimation accuracy rose from approximately 50% to 77%.
- Precision: The system achieved a Root Mean Square (RMS) error of 0.0315 N, significantly outperforming previous benchmarks (e.g., 0.13 N and 0.07 N from other prominent studies).
- Displacement Tracking: Absolute error in measuring tissue displacement was reduced from 2.0 mm to less than 0.2 mm, even during periods of tool occlusion.
Conclusion
V-ANFIS successfully demonstrates that visual uncertainty in robotic surgery can be mitigated by treating neighboring tissue points as "voters" in a consensus-based estimation model. While the current study was conducted in a controlled laboratory environment, the results suggest that neuro-fuzzy systems are well-suited for the inherent vagueness of surgical video. Future work will focus on adapting the system to the more complex environments of real surgery, including challenges like blood, smoke, and varying lighting conditions.
Summary updated