VibNet: Vibration-Boosted Needle Detection in Ultrasound Images

TL;DR

VibNet tackles robust needle detection in ultrasound images by exploiting periodic external vibration of the needle to reveal a distinct frequency signature in sequential frames. The framework encodes time-varying motion with a neural STFT-like pipeline, aggregates frequency features, and localizes shaft and tip via a Deep Hough Transform, supervised with Gaussian-blurred ground truth and focal loss. Across ex vivo porcine and bovine tissues, VibNet achieves substantial accuracy gains, particularly for nearly invisible needles (e.g., tip error down to $1.61\pm1.56$ mm and angle error $1.64\pm1.86^{\circ}$ in challenging cases), and demonstrates strong generalization and angle-robustness compared with UNet and WNet. The method runs at approximately 12 Hz, offering a practical, vibration-assisted approach to improve safety and efficiency in ultrasound-guided interventions, with potential extension to other imaging modalities and in-plane applications.

Abstract

Precise percutaneous needle detection is crucial for ultrasound (US)-guided interventions. However, inherent limitations such as speckles, needle-like artifacts, and low resolution make it challenging to robustly detect needles, especially when their visibility is reduced or imperceptible. To address this challenge, we propose VibNet, a learning-based framework designed to enhance the robustness and accuracy of needle detection in US images by leveraging periodic vibration applied externally to the needle shafts. VibNet integrates neural Short-Time Fourier Transform and Hough Transform modules to achieve successive sub-goals, including motion feature extraction in the spatiotemporal space, frequency feature aggregation, and needle detection in the Hough space. Due to the periodic subtle vibration, the features are more robust in the frequency domain than in the image intensity domain, making VibNet more effective than traditional intensity-based methods. To demonstrate the effectiveness of VibNet, we conducted experiments on distinct \textit{ex vivo} porcine and bovine tissue samples. The results obtained on porcine samples demonstrate that VibNet effectively detects needles even when their visibility is severely reduced, with a tip error of $1.61\pm1.56~mm$ compared to $8.15\pm9.98~mm$ for UNet and $6.63\pm7.58~mm$ for WNet, and a needle direction error of $1.64\pm1.86^{\circ}$ compared to $9.29\pm15.30^{\circ}$ for UNet and $8.54\pm17.92^{\circ}$ for WNet. Code: https://github.com/marslicy/VibNet.

Figures (8)

• Figure 1: A representative of the nearly invisible needle in US images. (a) and (b) demonstrate data acquisition setup on an ex-vivo animal tissue sample. (c) is a typical B-mode image with a nearly invisible needle, and (d) is the B-mode image acquired at the same location with an optimized beam angle, adjusting manually. The dashed orange rectangles indicate the needle's location.
• Figure 2: An intuitive explanation of the philosophy of the proposed method. (a). presents three image patches selected for analysis. (b). The vibrated needle induces different motion patterns in the US image sequence, making the needle itself perceptible to the network. It is noteworthy that the sinusoidal patterns were manually created to represent the ideal variations in vibration frequency and amplitude at each pixel. (c)-(e) are computed corresponding spectrograms of the three patches, demonstrating the distinct motion patterns.
• Figure 3: Overview of needle detection pipeline. VibNet comprises modules for (a). Temporal feature extraction in the spatio-temporal space, (b). Frequency feature extraction and aggregation in Frequency space, and (c). Needle shaft and tip prediction in Hough space. The needle's location is finally determined by (d). Post-processing the output images from VibNet. For further details of module (d), one can refer to Section\ref{['sec:method']}-C 2).
• Figure 4: Hardware setup and connections for data acquisition.
• Figure 5: An illustration of (a) a normal image with good needle visibility, (b) and (c) show a challenging case where US images are obtained at the same imaging plane with and without an optimal beam steering angle adjustment, respectively. The insertion angle for the normal case and challenging cases are $45^{\circ}$ and $30^{\circ}$, respectively.