RSV: Robotic Sonography for Thyroid Volumetry

TL;DR

A neural network-based segmentation with an automatic robotic ultrasound scanning for thyroid volumetry and the superiority of the motion guidance algorithms for the robot arm movement compared to a naive linear motion executed by the robot in terms of volumetric accuracy is demonstrated.

Abstract

In nuclear medicine, radioiodine therapy is prescribed to treat diseases like hyperthyroidism. The calculation of the prescribed dose depends, amongst other factors, on the thyroid volume. This is currently estimated using conventional 2D ultrasound imaging. However, this modality is inherently user-dependant, resulting in high variability in volume estimations. To increase reproducibility and consistency, we uniquely combine a neural network-based segmentation with an automatic robotic ultrasound scanning for thyroid volumetry. The robotic acquisition is achieved by using a 6 DOF robotic arm with an attached ultrasound probe. Its movement is based on an online segmentation of each thyroid lobe and the appearance of the US image. During post-processing, the US images are segmented to obtain a volume estimation. In an ablation study, we demonstrated the superiority of the motion guidance algorithms for the robot arm movement compared to a naive linear motion, executed by the robot in terms of volumetric accuracy. In a user study on a phantom, we compared conventional 2D ultrasound measurements with our robotic system. The mean volume measurement error of ultrasound expert users could be significantly decreased from 20.85+/-16.10% to only 8.23+/-3.10% compared to the ground truth. This tendency was observed even more in non-expert users where the mean error improvement with the robotic system was measured to be as high as $85\%$ which clearly shows the advantages of the robotic support.

Figures (6)

• Figure 1: An overview of the proposed robotic system. Left: the hardware setup with an US machine, a 6 DoF robotic arm, an US probe attached to the robot end-effector and a thyroid phantom, top right: the online 2D segmentation of the thyroid during a robotic scan, bottom right: the final 3D thyroid segmentation for both lobes.
• Figure 2: Our proposed workflow consists of two steps: a robotic US scan acquisition and a post-processing stage for the automatic volume estimation. The beginning of the first step involves human interaction: the user places the US probe on each lobe such that the thyroid is visible in the US image. The robot then moves the US probe while analysing the live US image. If shadows are detected or the thyroid is not centered the robot will modify the trajectory. On each step the system also evaluates if thyroid tissue is segmented. Once the end of the thyroid is reached the robot reverses the direction to acquire a sweep with the complete thyroid lobe. Finally, both the US B-mode images and the labelled US sweep are compounded (post-processing). These steps are repeated for each lobe. The merged label compounding for both thyroid lobes is used to estimate the total volume. The compounded 3D US image can be stored for potential follow-up examinations.
• Figure 3: Overview of the robotic movement. Left: the initial coordinate system is derived from the initial US probe position. The vector normal to the US plane (red, x-axis) gives the movement direction of the US probe. The robot then moves in x-direction and after detecting the first lobe end it moves in negative x-direction (red dotted arrow). Constant force is applied through an impedance control in z-axis (yellow) and the motion corrections are performed in the y-z plane (cyan dotted). Top right: path adjustment for lobe centering by translating the probe in the y-z plane (before $\rightarrow$ after), bottom right: path adjustment for shadow prevention by rotating the probe in the y-z plane (before $\rightarrow$ after).
• Figure 4: Examples of suboptimal US scans (left) with their online estimated segmentation masks (right). Top row, a) and b): the lobe is not centered, bottom row, c) and d): shadowing prevents optimal segmentation results.
• Figure 5: Qualitative results of the thyroid segmentation with shadow correction but without image centering (top), with image centering but without shadow prevention (middle) and with both motion adjustments (bottom).