1.

Introduction

Over the past several years, there has been a particular interest in the development of laser surgery systems due to the advantages offered by laser-based cutting, including minimal invasiveness, noncontact interaction, precise and small cuts, functional cut geometry as well as less trauma.^1–6 Moreover, in comparison to other mechanical procedures, such as conventional surgery, bone showed faster healing after interventions with laserosteotomes.⁷ Studying the effect of different laser parameters on quality and efficiency of laser cutting in both soft and hard tissues is a topic of present interest.^8–13 Although laserosteotomy has become a generally accepted technique in various surgical applications, this technique has two main drawbacks: lack of real-time information about depth of the cut and lack of information about the properties of the ablated tissue; as a result, critical structures of the body under or near the focal spot of the laser beam are prone to iatrogenic damage. In addition, to be a practical tool, clinical lasers have to be safe and effective in removing tissue with limited collateral damage. To reduce the thermal damage to the surrounding tissue, it is vital to use a cooling system to avoid carbonization. Carbonization happens when the tissue is heated up and all the water content is evaporated. Carbonization occurs not only in hard tissues but also in soft tissues. Carbonization not only reduces the ablation efficiency but also prolongs healing.^14,15

Lack of real-time information about depth of the cut can be solved by combining the laser surgery system with a coaxial real-time optical coherence tomography setup,^16,17 and lack of information about the properties of the ablated tissue can be improved by connecting the system to an optical detection setup.¹⁸ Therefore, having a real-time feedback method to detect the possible carbonization in case of any possible error in the irrigation system is needed.

The potential optical detection methods to investigate the properties of the tissues include optoacoustic-based measurements^19,20 and also spectroscopy-based measurements, including diffuse reflectance,^21,22 laser-induced breakdown,^23–25 Raman,^26,27 and fluorescence spectroscopy.^28,29 Among the above-mentioned optical methods, laser-induced breakdown spectroscopy (LIBS) showed its potential to detect the type of tissue with high accuracy. In LIBS, the light emitted from the ablation spot, which corresponds to the recombination spectra of ionized atoms and molecules, is collected with a spectrometer to resolve the atomic composition of the ablated sample. LIBS has been applied to differentiate between different tissue pairs with a high sensitivity and specificity (normally ranging from 70% to 100%), such as cartilage and cortical bone,³⁰ nerve and gland,³¹ nerve and fat³² as well as differentiation between oral soft and hard tissues.³³ Due to the compelling performance of LIBS in tissue characterization, we assume that applying LIBS also for detecting carbonization in laser surgery will be possible while tissue characterization can be performed in parallel.

The aim of this study is to differentiate between normal and carbonized hard porcine bone samples by monitoring the laser-driven plasma generated during a nanosecond pulse ablation using a frequency-doubled nanosecond Nd:YAG laser. A high-resolving power echelle spectrometer connected to an intensified CCD (ICCD) was employed to detect carbonization. The ICCD was capable of collecting LIBS signals with high temporal resolution (ps) and also short integration time (ns) to measure spectra for the classification of the samples. The analytical approach of calculating ratios of values of specific pairs of emission lines was taken to perform the differentiation between two groups of samples: normal (noncarbonized) and carbonized porcine bone. Then the sensitivity, specificity, and accuracy of the method were calculated afterward. Based on the results of the current study, we aim at proceeding LIBS-based carbonization detection also in real time. A successful real-time autodetection of carbonization in laserosteotomy will increase the safety of laserosteotomes. Additionally, this proof of principle study is intended to pave the way for in vivo experiments with an LIBS-based autocarbonization detection system in laser surgery.

2.

Materials and Methods

2.4.

Spectroscopy Setup

An echelle spectrometer with a wavelength range of 200 to 975 nm connected to an ICCD was used to reveal the spectral distribution of the laser-generated plasma light. The spectral resolution ($\lambda /\mathrm{\Delta }\lambda$) of the spectrometer was more than 4000. The CCD was cooled down to $-30°\mathrm{C}$ to reduce the background noise level. A fiber optic with a $50\text{-}\mu \mathrm{m}$ core connected to a UV–NIR light collector with an $F$-number of 2 was used to guide the plasma light into the spectrometer. A gate delay of $1\mu \mathrm{s}$ between the laser shot and opening of the intensifier was applied to remove the continuum emission of the plasma. The gate width was set to $200\mu \mathrm{s}$. Both gate delay and width were applied using the internal delay generator of the ICCD. Figure 1 shows the schematic of the LIBS setup.

Fig. 1

Schematics of the LIBS setup: A, laser (flash-lamp-pumped Nd:YAG), B, second-harmonic generator; C, harmonic separator; D, beam blocker; E, laser beam; F, focusing lens; G, bone sample; H, generated plasma; I, light collector; J, fiber optic; K, spectrometer (echelle); and L, computer.

3.

Results

The elements detected in the bone samples through LIBS were identified as carbon (C), hydrogen (H), oxygen (O), calcium (Ca), sodium (Na), magnesium (Mg), zinc (Zn), and strontium (Sr). In addition to the atomic emission line of carbon, a molecular line of carbon-to-carbon bonding (${\mathrm{C}}_2$) was also observed. This result is in line with results described in literature.^{30–32,37–44} Note that the collected spectra may include emission lines of the elements found in the ambient air. Interestingly, the higher concentration of carbon in carbonized samples was not only observed in the average intensity of pure carbon emission line (13.77 for carbonized samples as compared with 3.15 for normal samples) but also in the carbon-related molecular emissions of the ${\mathrm{C}}_2$ (10.12 for carbonized samples as compared with 4.38 for normal samples). Moreover, a reduction in the emission intensity of the hydrogen line was observed in the carbonized sample compared to the normal one (from 9.08 to 5.05). Decreasing hydrogen emission intensity and increasing carbon emission intensity seem to indicate that the bone has dried out and will be followed by subsequent carbonization if not properly rehydrated before continuing with the laser ablation. In addition to the atomic and molecular LIBS, Raman spectroscopy results also show a change in the carbon bonding. The Raman spectra of normal and carbonized bones are shown in Fig. 2.

Fig. 2

Raman spectra of normal and carbonized bone samples.

As shown in Fig. 2, the Raman spectra of normal and carbonized bone samples, which both are normalized to the carbon-to-carbon bonding intensity at $1589{\mathrm{cm}}^{-1}$ (highlighted by light yellow in the picture),^45–47 show a significant reduction in the $\mathrm{C}─\mathrm{H}$ bonding between 2870 to $3010{\mathrm{cm}}^{-1}$^46–49 and also $\mathrm{O}─\mathrm{H}$ bonding between 3380 to $3530{\mathrm{cm}}^{-1}$^45,46,50 of the carbonized bone sample. Moreover, there is some reduction in intensity of the bonding of the carbonized bone related with phosphate (${\mathrm{PO}}_4^{3-}{\nu }_2$) around $422{\mathrm{cm}}^{-1}$, double peak of phosphate antisymmetric bending frequency (${\mathrm{PO}}_4^{3-}{\nu }_4$) around 566 and $636{\mathrm{cm}}^{-1}$, proline around $835{\mathrm{cm}}^{-1}$, phosphate (${\mathrm{PO}}_4^{3-}{\nu }_1$) around $883{\mathrm{cm}}^{-1}$, phosphate (${\mathrm{PO}}_4^{3-}{\nu }_3$) around $1024{\mathrm{cm}}^{-1}$, amide III (primarily from the in-phase combination of NH in-plane bend and CN stretch) around $1288{\mathrm{cm}}^{-1}$, pentosidine around $1495{\mathrm{cm}}^{-1}$, and amide I (primarily from the $\mathrm{C}═\mathrm{O}$ stretch and $\mathrm{C}─\mathrm{H}$ bending) around $1687{\mathrm{cm}}^{-1}$.^{48,49,51–53} Results of the Raman spectroscopy are in a good agreement with molecular LIBS. Increase in carbon concentration of the carbonized bone sample (obtained from atomic LIBS data) in combination with results of the Raman and molecular LIBS, which show bonding of carbon to carbon has not broken while that of carbon to other elements have broken, indicates that the carbonization has occurred in the carbonized samples.

Figures 3(a) and 3(b) show LIBS spectra of normal and carbonized bone samples, respectively.

Fig. 3

LIBS spectra of (a) normal and (b) carbonized bone sample showing selected peak pairs.

As shown in Figs. 3(a) and 3(b), the ratio of the intensities between the sodium (Na) peak at 321.2 nm (from $2s^{2}2p^{5}3p$ to $2s^{2}2p^{5}3s$) and the calcium peak at 612.2 nm (from $3p^{6}4s5s$ to $3p^{6}4s4p$) is different in normal and carbonized bone. These two prominent peaks with a high difference in intensity ratio in two groups of samples, which had the lowest Wilks’ lambda between the observed peak pairs, can be used for differentiation between carbonized and noncarbonized bone. Figure 4 shows the sodium-to-calcium intensity ratio for all samples. The first 50 data points were obtained from sample 1, the second 50 data points from sample 2, and so on, until the last 50 data points from sample 5.

Fig. 4

Sodium-to-calcium intensity ratio for all samples.

As shown in Fig. 4, most of the ratios related to normal bone are below the threshold line, and most of the ratios related to carbonized bone are above the threshold line. The threshold line was selected in a way to maximize the accuracy of the classifier. Statistical parameters of the classifier, including true positive rate (sensitivity), true negative rate (specificity), positive predictive value (precision), negative predictive value, and accuracy, are shown in Table 1.

Table 1

Statistical parameters of the classifier.

As written in the last column of Table 1, all statistical parameters (obtained from 500 spectra) are above 92%. The ROC curve was also plotted, and the area under curve (AUC) was calculated to confirm the performance of the classifier. Figure 5 shows the ROC curve. The AUC of the curve was over 98%.

Fig. 5

ROC curve.

4.

Discussion

In this paper, LIBS showed reliable result (accuracy of more than 95%) for carbonization detection in ex vivo condition. Ex vivo performance of LIBS for detecting carbonized bone seems reliable even with simple ratio-threshold-based methods. Thus, the ex vivo results are very likely to transfer also to in vivo experiments. However, the achieved accuracy from the ex vivo condition is likely to decrease for future in vivo experiments, where bone is not so well-prepared. Possible reasons could be the influence of the superficial contamination of the probing surfaces with blood or any rinsing solutions, such as saline or cooling water during the clinical procedures. A possible solution could be employing a double-pulse LIBS system. In double-pulse LIBS, the first pulse can remove the liquid on the focal point; then the second pulse quickly reaches the target surface before the target area is refilled with liquid. From a machine learning point of view, to further increase detection accuracy, more complex classifiers could be used that, e.g., also involve additional intensities of other elements. While the number of false positives in this experiment was very low (19 out of 500), but false positives are not a real issue in this case. From a safety point of view, it is better to assume that carbonization has occurred and increases irrigation to avoid future carbonization at the cost of a reduced cutting speed. In the current experiment, the soft tissue was carefully removed from the surface of the bone using a surgical scalpel, but it is suggested to consider the first initial shots as a cleaning shot. Although it has been reported that the type of nutrition and also age may influence the elemental composition of the tissues, this will not significantly influence the differentiation performance in this study because only the most prominent emission lines were considered for classification.^33,39 However, the differentiation performance could be improved by considering the average of multiple spectra for the analysis, but using multiple spectra in a real-time application is time-consuming both for collection and also data analysis. Moreover, it should be considered that using multiple spectra is a trade-off between the damage caused during the collection/calculation time and the increase in differentiation accuracy. Finally, it is worth noting that in this pilot study, a well-carbonized bone was used as a sample; therefore, in the further studies, the performance of the technique should also be evaluated with less carbonized samples to confirm the applicability of the proposed method.

5.

Conclusion

The preliminary results of this study demonstrate that LIBS is a powerful technique for autocarbonization detection under ex vivo conditions by monitoring the plasma plumes occurring during laserosteotomy procedures. Based on the previous reports, the elements detected during the ablation were in agreement with those expected to be found in the elemental composition of bone. The intensity ratio of sodium and calcium enabled successful differentiation of carbonized bone from noncarbonized bone with high accuracy. Sensitivity and specificity of 97% and 92% were achieved, respectively. Therefore, this suggests that carbonization monitoring during laserosteotomy could be successfully achieved using an LIBS-based detection system. However, further research will be needed to confirm the potential in vivo clinical applicability of the proposed method.