Train a segmentation model on synchronized ultrasound + contact force, then condition it on real-time force at inference (e.g., FiLM-style modulation) so bone boundaries stay stable even as probe pressure, angle, and tissue vary. From each B-mode frame plus a scalar/vector force, the system outputs a clean bone mask (with optional boundary confidence) ready for 3D reconstruction and scan planning. Why force? Pressure changes appearance—compression, sliding, shadowing—so image-only models drift; force-conditioning anchors predictions, cutting false edges and reducing cross-subject/site variance. The result is real-time, in-the-loop performance that holds up under different probe poses, enabling reliable 3D surfaces, force-aware control, and quality monitoring—no markers, no extra hardware beyond your robot/sensor.
