AI framework boosts ultrasound robot tracheostomy accuracy with YOLOv8+SAM2

Tracheostomy requires precise localization of the incision site, but traditional manual palpation is subjective and ultrasound remains operator-dependent. To address this, Hiu Ching Cheung and colleagues propose a learning-based hierarchical tracheal anatomy understanding framework tailored for ultrasound robots. The system uses a two-stage perception pipeline: a lightweight YOLOv8n backbone for coarse localization, followed by a sparse, prompt-optimized SAM2 decoder for high-fidelity segmentation. A hybrid training strategy bridges curated lab data with unconstrained clinical sequences, ensuring robustness. Experimental benchmarks show the decoupled architecture achieves a consistent Mean Dice Similarity Coefficient (DSC) of 0.777 across both controlled and generalized domains, a major improvement over U-Net baselines that suffer from anatomical fragmentation (generalization DSC ≤ 0.494).

By constraining mask decoding to targeted, sparse regions of interest, the model achieves a throughput of 6.92 FPS, which is vital for closed-loop robotic teleoperation during surgery. This performance validates that robust hierarchical tracheal anatomy understanding can be derived by coupling lightweight localization with foundation-level vision models. The framework establishes a scalable foundation for standardized, autonomous surgical assistance, effectively navigating the variability of real-world ultrasound to enhance safety and precision in robotic-assisted tracheostomy. The work has been accepted at the 2026 International Conference on Cyborg and Bionic Systems (2026 ICCBS).