Deep Learning based acoustic measurement approach for robotic applications on orthopedics

TL;DR

The paper tackles the need for accurate, noninvasive bone-position tracking in robotic orthopedic surgery by using A-mode ultrasound and a novel deep learning architecture. It introduces CasAtt-UNet, a cascade of coarse and refined attention UNets connected via a sampling-based proposal to locate sparse bone peaks in 1D US signals. On cadaver data across four lower-limb regions, the method delivers sub-millimeter accuracy (average <0.48 mm) with robust peak detection, outperforming traditional, expert-driven approaches in precision. The findings suggest real-time, safe bone localization feasible for robotic TKA and potentially other surgeries, though generalization beyond a single cadaver requires further study.

Abstract

In Total Knee Replacement Arthroplasty (TKA), surgical robotics can provide image-guided navigation to fit implants with high precision. Its tracking approach highly relies on inserting bone pins into the bones tracked by the optical tracking system. This is normally done by invasive, radiative manners (implantable markers and CT scans), which introduce unnecessary trauma and prolong the preparation time for patients. To tackle this issue, ultrasound-based bone tracking could offer an alternative. In this study, we proposed a novel deep learning structure to improve the accuracy of bone tracking by an A-mode ultrasound (US). We first obtained a set of ultrasound dataset from the cadaver experiment, where the ground truth locations of bones were calculated using bone pins. These data were used to train the proposed CasAtt-UNet to predict bone location automatically and robustly. The ground truth bone locations and those locations of US were recorded simultaneously. Therefore, we could label bone peaks in the raw US signals. As a result, our method achieved sub millimeter precision across all eight bone areas with the only exception of one channel in the ankle. This method enables the robust measurement of lower extremity bone positions from 1D raw ultrasound signals. It shows great potential to apply A-mode ultrasound in orthopedic surgery from safe, convenient, and efficient perspectives.

Figures (7)

• Figure 1: Pipeline overview: Our method had three steps (from left to right): performed cadaver experiment to collect ultrasound signals (network input) and calculate bone positions (network output), Use the dataset to train CasAtt-UNet, recover the bone position and evaluate.
• Figure 2: Location of the US holdersniu2018situ: Our ultrasound holders were installed on the six locations of the left leg: Trochanter, Mid Tibia, Femur Epicondyle, Tibia Epicondyle, Mid Tibia and Ankle. Each holder was tied using bandages. Notice that the distribution of optical markers and transducers were different with the ones in the image.
• Figure 3: Steps to determine ground truth labels: The optical markers in the predefined and the experiment case determined the transformation, which was used to transform US transducers and waves directions to the experiment coordinate frame. The label was calculated using Euclidean distance and the speed of US.
• Figure 4: Intersection between bones and US waves: This showed one moment in the experiment that the US waves intersected with the bony surface. The intersection positions produced ground truth locations. Red dots were transducer probes positions. Black dots were intersection positions. Green lines were waves directions. The ground truth distance (later used for labeling) was the line segments between black dots and red dots. For some waves there was no black dot as there was no intersection.
• Figure 5: Pipeline of Algorithm: Given a 130mm US raw signal, the first coarse U-Net captured the approximate range (1mm) of bone peak. The output was used to segment a continuous length (3mm) of signal region, which is the input of the refined U-Net to determine the exact position of the bone peak.