TL;DR
This paper introduces a novel method for identifying critical bone layers during pedicle screw surgery using acoustic emission (AE) signals. By analyzing the frequency distribution of AE signals and fitting a specific function to these distributions, the researchers achieve an 84.2% recognition rate for bone layers, offering a more precise and less invasive approach than traditional methods.
What Problem Does This Paper Address?
Pedicle screw fixation is a crucial surgical technique for spinal stabilization, but accurately controlling the depth of the drilling bit is paramount to avoid damaging surrounding nerves and tissues. The transition zone between different bone layers (cortical, cancellous, and inner cortical) is particularly challenging to identify due to the lack of direct visual feedback and the subtle changes in mechanical properties. Traditional methods rely heavily on surgeon experience, which can lead to errors. While other sensing modalities like force/torque signals exist, they can be noisy and introduce delays. This paper aims to address this by leveraging acoustic emission (AE) signals, which are generated by the drilling process itself, to provide real-time, accurate identification of bone layers, especially the critical transition zones.
Methods
The core of this research lies in a novel Frequency Distribution-Based (FDB) algorithm. The process begins with acquiring AE signals during bone drilling. These signals are then pre-processed using a recursive Fast Fourier Transform (FFT) to shift them into the frequency domain. The key innovation is the FDB algorithm, which:
- Selects a specific frequency band (10-15 kHz): This band is chosen to minimize noise interference and capture relevant drilling characteristics.
- Normalizes and sorts the frequency distribution: This step ensures data is on a comparable scale and helps reveal underlying patterns.
- Applies a coordinate system conversion: This transforms the data to better suit fitting techniques.
- Fits a specialized function: A sine-based function of the form $y = a \cdot e^{bx} \cdot \sin[(2\pi + c \cdot x(1-x)) \cdot x^d]$ is used to model the transformed AE signal characteristics. The coefficients (a, b, c, d) of this fitting function are found to uniquely represent different bone layers.
- Utilizes a Neural Network: A Backpropagation (BP) neural network with three hidden neurons is trained using the extracted coefficients as input features to classify the bone layers.
Datasets
The paper describes experiments conducted using fresh bone samples with typical cortical and cancellous bone structures, including trabecular bone as a transition region. The data collected from these experiments was divided into a training set and a testing set. The training set consisted of ten marked data sets, and the testing set comprised twenty sets of data. The paper does not mention the use of any pre-existing public datasets for this specific application.
Key Insights and Ideas
- AE signals as a sensitive indicator: The study demonstrates that AE signals, often overlooked for this specific application, contain rich information about the drilling process and bone layer transitions.
- Frequency distribution reveals bone structure: The way AE signal energy is distributed across specific frequencies (10-15 kHz) changes significantly between cortical, cancellous, and transitional bone layers.
- Fitting function coefficients are unique identifiers: The parameters of the chosen fitting function (a, b, c, d) act as distinct "fingerprints" for each bone layer, capturing subtle differences in the AE signal's frequency characteristics.
- Transition zones are characterized by uniform energy distribution: Unlike cortical and cancellous bone, transition zones exhibit a more uniform energy distribution across the analyzed frequency band, which is effectively captured by the FDB algorithm.
- Neural networks excel at classifying complex features: The BP neural network effectively learns the relationship between the fitting function coefficients and the bone layer identity, achieving a high recognition rate.
Key Results
The proposed FDB algorithm, coupled with a BP neural network, achieved a maximum bone layer recognition rate of 84.2%. The study found that:
- Cortical bone layers show amplitude concentration in high-frequency bands.
- Cancellous bone layers are concentrated in lower-frequency bands.
- Transition regions exhibit a more uniform distribution, with coefficients a and d being larger and b and c being smaller compared to cortical and cancellous bone.
- The algorithm's performance was compared to force signal analysis, with the FDB method providing more accurate and detailed identification, especially for transition regions.
Limitations and Future Work
- Limitations: The paper acknowledges that the internal structure of bone layers is not a single substance, and sound signal acquisition can be influenced by the surrounding bone media as depth increases. This can lead to variations in characteristic coefficients and potentially incorrect identifications. The recognition rates, while good, are not perfect (up to 84.2%).
- Future Work: The authors plan to focus on applying the proposed algorithm to real-time control of drilling robots. This would involve integrating the algorithm into robotic systems to enable immediate adjustments during surgery, further verifying its effectiveness and accuracy through more extensive experiments.
Summary updated