Automatic and accurate needle detection in 2D ultrasound during robot-assisted needle insertion process
Shihang Chen,Yanping Lin,Zhaojun Li,Fang Wang,Qixin Cao
DOI: https://doi.org/10.1007/s11548-021-02519-6
2021-10-22
International Journal of Computer Assisted Radiology and Surgery
Abstract:PurposeRobot-assisted needle insertion guided by 2D ultrasound (US) can effectively improve the accuracy and success rate of clinical puncture. To this end, automatic and accurate needle-tracking methods are important for monitoring puncture processes, avoiding the needle deviating from the intended path, and reducing the risk of injury to surrounding tissues. This work aims to develop a framework for automatic and accurate detection of an inserted needle in 2D US image during the insertion process.MethodsWe propose a novel convolutional neural network architecture comprising of a two-channel encoder and single-channel decoder for needle segmentation using needle motion information extracted from two adjacent US image frames. Based on the novel network, we further propose an automatic needle detection framework. According to the prediction result of the previous frame, a region of interest of the needle in the US image was extracted and fed into the proposed network to achieve finer and faster continuous needle localization.ResultsThe performance of our method was evaluated based on 1000 pairs of US images extracted from robot-assisted needle insertions on freshly excised bovine and porcine tissues. The needle segmentation network achieved 99.7% accuracy, 86.2% precision, 89.1% recall, and an F1-score of 0.87. The needle detection framework successfully localized the needle with a mean tip error of 0.45 ± 0.33 mm and a mean orientation error of 0.42° ± 0.34° and achieved a total processing time of 50 ms per image.ConclusionThe proposed framework demonstrated the capability to realize robust, accurate, and real-time needle localization during robot-assisted needle insertion processes. It has a promising application in tracking the needle and ensuring the safety of robotic-assisted automatic puncture during challenging US-guided minimally invasive procedures.
engineering, biomedical,radiology, nuclear medicine & medical imaging,surgery