Force Sensing Guided Artery-Vein Segmentation via Sequential Ultrasound Images

TL;DR

The paper addresses the difficulty of distinguishing arteries from veins in ultrasound due to morphological similarity by introducing a force sensing guided segmentation framework that leverages probe-contact force to identify two key deformation frames ($K_{ m min}$ and $K_{ m max}$) and fuses them with the current frame through a force-guided attention module and dynamic weights. The method is versatile and improves performance across multiple U-shaped networks (UNet, Swin-Unet, TransUnet) on the Mus-V dataset, the first multimodal ultrasound artery-vein dataset with synchronized force data; the authors report substantial gains in $mIoU$ and Dice, with FG-Transunet achieving $mIoU=75.63 ext{%}$ and Dice $=0.8547$. The Mus-V dataset comprises 3114 images from 105 videos with force data captured by a probe-mounted sensor, and the authors provide public release of code and data. This work advances vascular interventional robotics by enabling more accurate artery-vein segmentation through tactile force cues, and opens avenues for further optimization and dataset expansion.

Abstract

Accurate identification of arteries and veins in ultrasound images is crucial for vascular examinations and interventions in robotics-assisted surgeries. However, current methods for ultrasound vessel segmentation face challenges in distinguishing between arteries and veins due to their morphological similarities. To address this challenge, this study introduces a novel force sensing guided segmentation approach to enhance artery-vein segmentation accuracy by leveraging their distinct deformability. Our proposed method utilizes force magnitude to identify key frames with the most significant vascular deformation in a sequence of ultrasound images. These key frames are then integrated with the current frame through attention mechanisms, with weights assigned in accordance with force magnitude. Our proposed force sensing guided framework can be seamlessly integrated into various segmentation networks and achieves significant performance improvements in multiple U-shaped networks such as U-Net, Swin-unet and Transunet. Furthermore, we contribute the first multimodal ultrasound artery-vein segmentation dataset, Mus-V, which encompasses both force and image data simultaneously. The dataset comprises 3114 ultrasound images of carotid and femoral vessels extracted from 105 videos, with corresponding force data recorded by the force sensor mounted on the US probe. Our code and dataset will be publicly available.

Figures (3)

• Figure 1: (a) Hardware setup: a robotic arm, a force/torque sensor and an ultrasound probe. $\left \{ S_1\right \}$ and $\left \{ S_2\right \}$ respectively stand for coordinate systems of the sensor and the US image (b) An illustration of data acquisition. The force along the z-axis, $|F_z|$, initially increases and then decreases over time $t$. Twelve points are selected along the curve, with corresponding force magnitudes and ultrasound scans. Veins are labelled in blue, while arteries are labelled in red. Notably, veins exhibit higher deformability compared to arteries.
• Figure 2: Illustration of our proposed Force Sensing Guided Segmentation Framework. Selected key frames and the current frame are fed into a shared encoder to extract features. The force-guided attention module captures relation between embeddings of key frames and current frames, generating an enhanced embedding $\tilde{D}_V$. The value embedding of current frame $E_V$ is concatenated with $\tilde{D}_V$ to get final segmentation result through the decoder. Skip connection is applied between the encoder and the decoder.
• Figure 3: Qualitative results. The results demonstrate that integrating force sensing guidance effectively enhances the performance of three baseline networks. Specifically, our method improves the model's ability to differentiate between arteries (red) and veins (blue), and ability to segment small veins.