Abstract
Maintaining precise needle alignment within the narrow 2D imaging plane of an ultrasound probe is a critical challenge in guided interventions. When a needle shifts out-of-plane, it becomes invisible to standard computer vision algorithms, leading to procedural failure. This paper presents a novel approach using a vibration-based energy metric to autonomously restore alignment. By inducing low-frequency mechanical vibrations in the needle and analyzing the resulting energy heatmap, a robotic system can detect and re-align with the needle even when it is completely absent from standard B-mode images. Experimental results on ex-vivo porcine tissue demonstrate sub-millimeter translational accuracy and sub-degree rotational precision.
1. Introduction
In ultrasound-guided procedures such as biopsies, regional anesthesia, and vascular access, the clinician must keep the needle tip and shaft within the thin 2D ultrasound slice. This "in-plane" alignment is notoriously difficult to maintain manually. While Robotic Ultrasound Systems (RUSS) have been developed to automate this process, most rely on visual servoing. These methods often fail in clinical settings due to speckle noise, artifacts, and the fundamental problem of "out-of-plane" motion: once the needle leaves the imaging plane, there is no visual data left to track.
This research proposes a paradigm shift from "sight" to "signal." By leveraging the physics of motion through tissue, the system identifies the needle’s location based on the energy signature of its vibrations rather than its visual representation in a static image.
2. Methodology
2.1 Hardware Configuration
The experimental setup utilizes a dual-arm robotic configuration. One UR5e robotic arm manipulates a linear ultrasound probe (ACUSON Juniper), while a second arm manages the surgical needle. A custom motorized actuator is attached to the needle to provide controlled mechanical excitation.
2.2 Vibration-Based Energy Heatmap
The needle is vibrated at a specific low frequency (approximately 2 Hz). This frequency is chosen to be distinguishable from physiological noise and standard ultrasound artifacts. The system processes a sequence of ultrasound frames using a bandpass filter centered at 2 Hz for every pixel. This isolates the motion induced by the needle from the static tissue background.
An "Energy Heatmap" is then generated by calculating the intensity of these filtered signals. The core metric, the Average Energy ($E_{Avg}$), is derived from this heatmap. The researchers observed that $E_{Avg}$ follows a monotonic decay as the probe moves away from the needle, creating a predictable "energy hill" that can be used for navigation.
2.3 Autonomous Control Strategy
The robot employs a two-phase control strategy to restore alignment:
- Directional Search: The robot performs a small exploratory movement. If $E_{Avg}$ increases, the robot continues in that direction; if it decreases, the robot reverses its trajectory.
- Proportional Alignment: Once the direction is established, a proportional controller moves the probe to maximize the energy metric, effectively "homing in" on the needle's position until the peak of the energy hill is reached.
3. Experimental Validation
The system was evaluated using ex-vivo porcine tissue samples to simulate realistic acoustic properties. Eight separate insertions were performed across three different tissue samples. The system was tested against initial misalignments of up to 3 mm in translation and 12.5 degrees in rotation—conditions under which the needle is typically invisible in standard B-mode ultrasound.
4. Results and Discussion
4.1 Accuracy and Robustness
The vibration-based system demonstrated high precision in restoring alignment:
- Translational Error: 0.41 ± 0.27 mm
- Rotational Error: 0.51 ± 0.19 degrees
Crucially, the system remained functional at a 3.6 mm offset. At this distance, the needle is entirely invisible to traditional computer vision methods, yet the energy heatmap provided a sufficient signal to guide the robot back to the needle.
4.2 Key Insights
The primary advantage of this method is its reliance on a physical signature rather than visual segmentation. By creating a "signature" for the needle through controlled vibration, the software can bypass the complexities of ultrasound speckle noise and the need for deep learning-based needle detection, which often requires massive labeled datasets and high computational power.
5. Limitations and Future Work
While the current results are promising, several areas for improvement remain. The current control algorithm handles translation and rotation as separate degrees of freedom, whereas clinical reality often involves simultaneous 3D misalignment. Furthermore, the processing time required for image sequences makes the current navigation relatively slow. Future research will focus on integrating this energy metric with deep learning to handle complex 3D motions and adapting the system for varying tissue types, such as fat and muscle, which may affect vibration propagation.
6. Conclusion
This work introduces a robust vibration-based energy metric for autonomous needle tracking in robotic ultrasound. By moving beyond visual-only tracking, the system provides a reliable method for restoring alignment even when the needle is out-of-plane. The sub-millimeter accuracy achieved in ex-vivo trials suggests significant potential for improving the safety and success rates of ultrasound-guided interventions.
Summary updated