Robotic breast ultrasound with knowledge-guided path planning and vision feedback.

aimsStandardizing robotic breast ultrasound remains challenging in practice, particularly during near-lesion approach, where small nonlinearities can degrade imaging consistency and needle guidance.

methodsThis study uses publicly available datasets to derive skill-informed motion primitives for an ontology-guided path-planning framework and to define a representative breast ultrasound workspace with lesion-oriented linear trajectories. BUSI-derived lesion information is used to generate image-informed scan paths in the imaging plane, while JIGSAWS kinematic sequences are used to initialize motion primitives and neuro-fuzzy inverse-kinematics rules. The proposed robotic ultrasound scan controller maps Cartesian targets to joint commands, and its parameters are subsequently refined using an ant colony optimization for continuous domains (ACOR) strategy. Performance is evaluated in terms of joint-angle error, forward-kinematics path tracking, motion smoothness measured by mean absolute jerk, and needle-heading alignment relative to the planned trajectory.

resultsAmong the untuned variants, the controller initialized by data-driven clustering showed the strongest generalization across the evaluation workspace. ACOR-based refinement further reduced path-tracking error and markedly improved motion smoothness. Orientation alignment also improved relative to the baseline, although performance remained variable across test segments and the predefined success threshold was not achieved in all cases.

conclusionThe proposed knowledge-driven robotic ultrasound framework integrates ontology-guided linear path planning, neuro-fuzzy inverse kinematics, and metaheuristic parameter refinement to achieve accurate and smoother tracking in representative breast-imaging tasks. In the present proof-of-concept study, the image-based component is used primarily for lesion-informed trajectory generation and evaluation rather than full real-time visual servoing. These findings support the feasibility of standardized robotic ultrasound workflows and motivate future integration of explicit orientation costs, force regulation, and real-time image feedback.