Anatomical Region Recognition and Real-time Bone Tracking Methods by Dynamically Decoding A-Mode Ultrasound Signals

TL;DR

The paper addresses non-invasive knee bone tracking using A-mode ultrasound and tackles peak-detection accuracy with a deep learning framework. It introduces cascaded U-Nets with a Sampling-based Proposal and an anatomical region classifier to jointly detect bone peaks and identify transducer location. The method achieves sub-millimeter peak localization ($0.5 \pm 1$ mm) and ~97% region-classification accuracy in dynamic knee tracking, outperforming traditional peak-detection approaches. The approach enables real-time performance (~15 ms per signal) and holds potential for calibration in robotic-assisted surgeries and prosthetic devices.

Abstract

Accurate bone tracking is crucial for kinematic analysis in orthopedic surgery and prosthetic robotics. Traditional methods (e.g., skin markers) are subject to soft tissue artifacts, and the bone pins used in surgery introduce the risk of additional trauma and infection. For electromyography (EMG), its inability to directly measure joint angles requires complex algorithms for kinematic estimation. To address these issues, A-mode ultrasound-based tracking has been proposed as a non-invasive and safe alternative. However, this approach suffers from limited accuracy in peak detection when processing received ultrasound signals. To build a precise and real-time bone tracking approach, this paper introduces a deep learning-based method for anatomical region recognition and bone tracking using A-mode ultrasound signals, specifically focused on the knee joint. The algorithm is capable of simultaneously performing bone tracking and identifying the anatomical region where the A-mode ultrasound transducer is placed. It contains the fully connection between all encoding and decoding layers of the cascaded U-Nets to focus only on the signal region that is most likely to have the bone peak, thus pinpointing the exact location of the peak and classifying the anatomical region of the signal. The experiment showed a 97% accuracy in the classification of the anatomical regions and a precision of around 0.5$\pm$1mm under dynamic tracking conditions for various anatomical areas surrounding the knee joint. In general, this approach shows great potential beyond the traditional method, in terms of the accuracy achieved and the recognition of the anatomical region where the ultrasound has been attached as an additional functionality.

Figures (3)

• Figure 1: Steps to build our method: A cadaver experiment to get the dataset and annotate the bone peak location. Our network is trained by the dataset and infers the actual bone depth and anatomical region.
• Figure 2: The proposed network had two U-Nets with different scales of perception fields (for bone peak detection) and input signal classification. The input was a 1D ultrasound signal. The outputs were the signal classification results and the prediction of the peak location. The peak location prediction was a segment after thresholding the predicted probability sequence. The two U-Nets were connected by Sampling-based Proposal.
• Figure 3: Closer look at the large and small bias. The distribution of bias along the time (across the entire 1017 samples) is plotted on the left. The 3D position of knee joint in the middle was the same moment of the specified bias. In the right waveform figure, the bias was visually showed.