In this talk, I will introduce autonomous robotic imaging, which aims to reduce operator dependency and produce higher-quality, more consistent ultrasound scans. A key challenge in conventional freehand lung ultrasound (LUS) is navigating the probe to the standardized imaging plane (SIP), which is highly operator-dependent. To address this, we developed a robotic LUS system that autonomously identifies and performs probe placement using an RGB-D camera and a learning-based human pose estimation algorithm to determine the vertical attachment point. We also developed a SIP navigation framework that extracts generalizable anatomical features from real-time LUS images and uses template matching with image-based visual servoing to guide the probe. Integrated with a customized active-sensing end-effector (A-SEE), the system maintains optimal probe alignment using external body geometry to preserve signal quality. A human study showed robot-acquired images had superior image quality (Contrast-to-Noise metric) and improved consistency through constant probe pressure. Notably, left-sided scan regions closer to the robot base achieved statistically significant improvements over human operators, highlighting the influence of robot configuration on outcomes. This approach promotes consistency, interpretability, and adaptability across patients.
Zhang et al. (Wed,) studied this question.