Deep-Learning Approach for Tissue Classification using Acoustic Waves during Ablation with an Er:YAG Laser (Updated)

TL;DR

This work proposes a tissue classification method analyzing acoustic waves generated during laser ablation, demonstrating its applicability in an ex-vivo experiment and results indicated that combining time-dependent data with its frequency spectrum achieved the highest classification accuracy.

Abstract

Today's mechanical tools for bone cutting (osteotomy) cause mechanical trauma that prolongs the healing process. Medical device manufacturers aim to minimize this trauma, with minimally invasive surgery using laser cutting as one innovation. This method ablates tissue using laser light instead of mechanical tools, reducing post-surgery healing time. A reliable feedback system is crucial during laser surgery to prevent damage to surrounding tissues. We propose a tissue classification method analyzing acoustic waves generated during laser ablation, demonstrating its applicability in an ex-vivo experiment. The ablation process with a microsecond pulsed Er:YAG laser produces acoustic waves, acquired with an air-coupled transducer. These waves were used to classify five porcine tissue types: hard bone, soft bone, muscle, fat, and skin. For automated tissue classification, we compared five Neural Network (NN) approaches: a one-dimensional Convolutional Neural Network (CNN) with time-dependent input, a Fully-connected Neural Network (FcNN) with either the frequency spectrum or principal components of the frequency spectrum as input, and a combination of a CNN and an FcNN with time-dependent data and its frequency spectrum as input. Consecutive acoustic waves were used to improve classification accuracy. Grad-Cam identified the activation map of the frequencies, showing low frequencies as the most important for this task. Our results indicated that combining time-dependent data with its frequency spectrum achieved the highest classification accuracy (65.5%-75.5%). We also found that using the frequency spectrum alone was sufficient, with no additional benefit from applying Principal Components Analysis (PCA).

Figures (5)

• Figure 1: Setup contained an Er:YAG laser with a wavelength of $2940nm$, where the pulses had a repetition rate of $2Hz$, and a duty cycle of 260µs The tissue was placed at the focal point of the laser, at a distance of $30mm$ from the lens. A transducer measured the acoustic wave generated during the ablation at a distance of $5cm$, with an angle of $45°$. We used the measured signal as the input to our NNs to classify different tissue types.
• Figure 2: Five exemplary tissues used for the experiments: A total of ten samples of each tissue were used. Five 100-pulse ablations were performed on each tissue.
• Figure 3: Visualization of the preprocessing steps: Top left, a $1ms$ window shows the measured acoustic wave from a soft bone ablation. The top right is the absolute median filtered acoustic wave (kernel size 11). To remove the ToF, we take five times the maximum value of the first $96µs$ of the filtered wave (green dotted line). The intersection is marked with a red dot, where the area of interest (marked in magenta) starts with a window size of $384µs$. The bottom left shows the resulting acoustic wave in the area of interest, which we used as an input for our NNs. The bottom right is the frequency spectrum of $0-1M\Hz$ of the acoustic wave shown on the bottom left. These frequency spectra are used as input for our NNs.
• Figure 4: Overview of the different NNs we used. $N$ is the number of consecutive acoustic waves used as an input. $M$ is the input size of the FcNNs when using frequency-dependent data: Namely for the frequency range $0-1M\Hz$ we have $M=385$, for the low-frequency ($0-0.333M\Hz$) and mid-frequency ($0.333-0.666M\Hz$) is $M=128$, and for high-frequency ($0.666-1M\Hz$) is $M=129$.
• Figure 5: The figure visualizes the activation map resulting from CNN$_{\text{fft}}^\dagger$. The higher the value, the more important the frequency is for the classification task. The green area under the blue curve is the largest (low-frequency), then the high-frequency area market as yellow, and the smallest area is the mid-frequency market with cyan.