From Synchronized Testbed to MaskForceNet: End-to-End Force Estimation for Neurovascular Guidewires

초록

In robot-assisted neurovascular interventions, providing haptic feedback is crucial for preventing vessel perforation but remains challenging due to the physical constraints of integrating sensors into micro-guidewires. While vision-based force estimation offers a promising sensor-less alternative, existing studies predominantly target cardiac ablation catheters, overlooking the micro-force regime and complex deformations inherent to fragile neurovascular tools. To address this gap, we present an end-to-end framework for estimating 3-axis contact forces using catheter deformation. We developed a custom robotic testbed capable of acquiring synchronized multimodal data, pairing bi-plane video with high-precision ground-truth force measurements using a 0.035-inch guidewire. Leveraging this dataset, we propose MaskForceNet, a lightweight multi-modal architecture that fuses Siamese visual features from guidewire-centric segmentation masks with robot kinematic priors. Experimental results demonstrate that the proposed segmentation-based approach significantly outperforms raw image inputs in terms of robustness against environmental noise, as confirmed by Grad-CAM analysis. The model achieved a high coefficient of determination for insertion forces, establishing a solid foundation for realizing real-time, sensor-less haptic feedback in telesurgery. © 2026 IEEE.

키워드