All-ﬁber-optic LIBS system for tissue diﬀerentiation: A prospect for endoscopic smart laser osteotomy

Minimally-invasive laser surgeries could beneﬁt from a ﬁber-optic laser-induced breakdown spectroscopy (FO- LIBS) setup for real-time tissue characterization. In FO-LIBS, the sample receives limited light irradiance due to the ﬁber’s low damage threshold and diminished laser beam quality. Therefore, the plasma created with FO- LIBS is less luminant than that of free-space LIBS. Furthermore, only a small portion of plasma emission can be collected, as the lens’s size at the ﬁber tip is restricted to ﬁt inside the narrow channel of an endoscope. A high optical throughput Echelle spectrometer was developed to compensate for low-intensity light collection with FO-LIBS. The Echelle spectrometer was tested for tissue diﬀerentiation when combined with a ﬂexible ﬁber bundle delivery setup and a small lens at the bundle’s tip. The customized FO-LIBS setup, coupled with multivariate data analysis, successfully diﬀerentiated bone from surrounding soft tissue (muscle, fat, and bone marrow) with 100% cross-validated (CV) sensitivity and speciﬁcity. The CV sensitivity and speciﬁcity for diﬀerentiation between all tissues were 90.2% and 96.7%, respectively. The results demonstrate, to the best of our knowledge, the ﬁrst ﬂexible FO-LIBS system, which may provide a further step towards the development of a smart endoscopic laser scalpel.


Introduction
Traditional mechanical tools (e.g., saws, drills, chisels) used for bone cutting -the gold standard in osteotomy for thousands of years -require a certain degree of mechanical force (like grinding or hammering) to function [ 1 , 2 ]. The mechanical pressure exerted by saws and drills (contact mode cutting) presents some disadvantages for use as an osteotomy tool, including poor surface evenness, high risk of contamination, substantial material loss, metal abrasion, limited cutting geometry, bone fragmentation, and thermal damage. Moreover, saws and drills can create an amorphous, mineral-rich carbon layer on the bone surface, changing the mineralized matrix, which results in a prolonged healing process [ 3 , 4 ]. Due to the many side effects of using conventional osteotomy tools, alternative solutions are being sought. Some of the lesserknown techniques for osteotomy proposed and examined over the last decades include machining based on microwaves, ion beams, ultrasonic energy, water jets, and lasers [ 5 , 6 ]. Laser beams do not deflect while traveling (unlike water jets) and do not carry momentum. Lasers interact with materials without mechanical interaction forces and thus contactless. Without such mechanical forces, the body movements during surgery are minimized, and consequently, intraoperative image guid-ance systems can perform with higher accuracy. Several studies have been carried out to compare the performance of laserosteotomy with the conventional mechanical saws and drills as well as piezoelectric cutting tools. Using a laser beam offers a high axial and lateral resolution, allowing for a high degree of freedom when cutting. This freedom provides significant advantages during surgery, particularly oral and maxillofacial surgery, where implants are used to replace parts of the bone. Microstructural analysis, namely micro-computed tomography (CT) and histological examinations, have shown that more new mineralized bone is formed in cuts being performed with a laser rather than a piezo-osteotome (PZE) [7] . Accelerated healing of laserosteotomy is also reported here [8] . Histological and radiological examinations revealed primary gap healing in sheep two months post-laserosteotomy, and almost no visible osteotomy gap after three months [9] . While preoperative planning can navigate the laser to the required area, a lack of real-time monitoring risks iatrogenic damage due to any possible number of unpredicted errors or body movement. Therefore, real-time feedback is vital to preventing damage to the surrounding soft tissues during laserosteotomy. To avoid such damage, several optical approaches have been developed, including laser-induced breakdown spectroscopy (LIBS), laser-induced breakdown thresholding, diffuse reflectance spec- troscopy (DRS), Raman spectroscopy, autofluorescence spectroscopy, random lasing, optical coherence tomography (OCT), speckle analysis, shockwave measurement, and combustion/pyrolysis light analysis. A short overview of these feedback methods is provided elsewhere [10] . Of the approaches mentioned above, LIBS is an up-and-coming technique for analyzing biological tissues, especially calcified tissues like bone. It allows for analysis without sample preparation and, with a single shot, reveals the material's chemical composition under the cut in a few milliseconds [11][12][13][14][15][16][17] . Multiple studies have shown the potential of using LIBS to differentiate bone from other soft tissues, such as muscle, fat, skin, nerve, mucosa, liver, tendon, as well as cartilage [18][19][20][21][22][23][24][25][26] . The studies differentiated between the soft tissues surrounding bone, except for bone marrow. Moreover, none of the studies mentioned could simultaneously fulfill the complete set of requirements for a minimallyinvasive laserosteotomy feedback system, including the possibility of insertion inside the narrow channel of a flexible endoscope. A delivery system capable of transmitting tens of megawatts of nanosecond pulses of frequency-doubled Nd:YAG (532 nm) was recently developed [27] . The fiber delivery system developed showed high flexibility and bendinsensitive behavior. Flexibility is of high importance for compatibility with flexible endoscopes. This study aims to develop a method capable of providing optical feedback on the type of tissue being cut, namely bone, bone marrow, muscle, and fat (adipose), using the flexable fiber delivery system and a custom-made sensitive Echelle spectrometer dedicated to this application.

Materials and methods
While a large body of research on using LIBS for biological tissue analysis exists, most studies utilized free-space beam delivery and fiberoptic light collection [28][29][30][31][32][33][34][35] . In free-space LIBS, sufficient peak power can be delivered to the sample surface by simply using a few mirrors and/or lenses. Moreover, the laser beam maintains focusability, as it does not pass through an optical fiber. Therefore, the beam can be focused onto a relatively small focal spot, allowing for higher irradiance of the sample. Without space limitations, free-space LIBS can use bulky light collectors/collimators to collect more light from the laserinduced plasma. The light collectors/collimators employ wide diameter lenses/mirrors to maximize optical throughput light collection (low F-number). Although this is the most efficient way of performing LIBS analysis, the bulky setup can not be miniaturized for endoscopic applications. Minimally-invasive surgery requires a miniaturized beam delivery setup. In other words, the laser pulse should be delivered to the sample via optical fibers; this limits the amount of deliverable peak power to the sample (due to the limited laser-induced damage threshold (LIDT) of optical fibers) and prevent the use of bulky light collectors/collimators. In such a setup, the plasma emissions cannot be collected with high optical throughput (i.e., the smaller the lens/mirror diameter, the lower the optical throughput). Commercially available Echelle spectrometers typically have a high F-number of 10 or more [36][37][38][39][40] , which results in low optical throughput. For this reason, a custom-made Echelle spectrometer was designed and developed with higher optical throughput than those available on the market. For use in minimally-invasive surgery, fiberoptic LIBS (FO-LIBS) must also collect the plasma light using the same fiber employed for laser beam delivery, as space inside the endoscope is limited. For flexible endoscopic applications, the fiber system should be flexible down to a bending radius of approximately 15 mm [41] and operate without bending loss. While a few researchers have used either a fiber laser [21] or a laser coupled to a fiber optic [ 42 , 43 ] for beam delivery during biological tissue analysis with LIBS, none of them achieved the conditions for flexible endoscopic application -namely, a common fiber for both beam delivery and light collection with flexible and bendinsensitive behavior, and small focusing/collection optics at the tip of the fiber.

Fiber-based laser beam delivery
Due to the limited LIDT of glass, large solid-core fiber or hollow-core fiber are typically used for delivering the high peak powers required for LIBS, typically in the gigawatt (GW) range. Using large-core fibers limits the bending radius, which depends on the fiber's core and cladding thickness. Therefore, a high-power delivery setup was recently developed to achieve bend-insensitive flexible beam delivery. The setup uses a beam shaper and a fused-end fiber bundle of 800 individual fibers (Ce-ramOptec GmbH, Bonn, Germany), with low numerical aperture (NA) launch conditions [27] . A single small half-ball lens focuses the laser beam on the sample and collects the plasma emission. Fig. 1 shows the dimensions of the fiber system.
As shown in Fig. 1 , the length of the solid parts inside of the endoscope is 13.5 mm; other parts are flexible with a minimum bending radius of 15 mm (tested) [27] . The diameter of the parts inside the endoscope are small enough to fit inside an endoscope with an internal diameter of 10 mm [44] . At the bundle input, the jackets were removed and individual fibers were fused together, providing a solid structure without gaps between the individual fibers. The length and diameter of the fused part of the bundle were 10 mm and 3 mm, respectively. For other parts of the bundle, the jackets of individual fibers were not removed and the individual fibers remained unattached to maintain flexibility. The diameter of the total bundle was 3.5 mm. At the bundle output, individual fibers were glued to each other. The core, cladding, and jacket diameter of individual fibers were 100 μm, 106 μm, and 112 μm, respectively. Fig. 2 shows the input and output bundle facets.
This study used a pulsed Q-switched Nd:YAG laser (Q-smart 450, Quantel, France) working at its fundamental harmonic (1064 nm), with a pulse duration of 6 ns. A green, low-power continuous wave (CW) diode laser, aligned with the pulsed laser beam, was used to visualize the focal spot. A top-hat beam shaper (GTH-5-250-4-532, TOPAG Lasertechnik GmbH, Darmstadt, Germany) was employed to illuminate all individual fibers within the fiber bundle with equal energy. An air nozzle helped to prevent dust from reaching the surface of the fiber bundle. The bundle exhibited bend-insensitive behavior due to using loworder modes to launch the bundle (tested with different bending radiuses down to 1.5 cm). More information about the fiber delivery process can be found elsewhere [27] . Fig. 3 (a) and (b) shows the experimental setup for laser beam delivery and plasma emission collection.
The same optical fiber and half-ball lens used to deliver the laser pulses were used to collect the plasma emission. The surface of the halfball lens was wiped off with a lens cleaning tissue when needed (to clean the surface from the deposited particle due to tissue ablation). An off-axis parabolic (OAP) mirror with a through-hole (the pierced mirror in Fig. 3 ) allowed the laser beam to pass through while simultaneously directing the plasma emission to the spectrometer's optical fiber.

Echelle spectrometer
Bandwidth, resolution, and optical throughput are three spectrometer parameters that cannot be maximized simultaneously. Attempts to maximize all parameters at the same time result in either a very bulky setup (not portable and very expensive, like spectrometers designed for observatories), or a spectrometer incapable of snapshot collection (spectrometers with mechanical movable parts), which cannot be used for fast measurements [45] . Two-dimension spectroscopic configurations such as the Echelle designs have helped achieve wider bandwidth without losing resolution. Still, it is not possible to accomplish high-resolution, high-throughput, and wide bandwidth at the same time. Therefore, parameters were optimized according to the application.
The sensitivity of a spectrometer is a function of its optical throughput. The optical throughput, i.e. how much light can pass through the spectrometer, is indicated by its F-number; the lower the F-number, the higher the throughput. High F-numbers can reduce the sensitivity of the spectrometer. The plasma plume created with FO-LIBS is less luminant Fig. 1. Dimensions of the developed fiber system (input from the left side, the output from the right side) in relation to endoscopic placement. The half-ball lens shown on the right was used to focus the laser beam and collect the plasma emission simultaneously. (due to limited deliverable energy and reduced focusability) than that created with free-space LIBS. This fact, along with the small portion of plasma light collected by the small lens at the fiber tip, results in the spectrometer receiving low-intensity light. Therefore, a custom-made Echelle spectrometer was developed to compensate for low throughput light collection. The spectrometer was developed by simulating the mechanical and optical parts in Solidworks, Matlab, and Zemax (OpticStudio) [46] , to find the most suitable components and to provide better performance than the previously developed Echelle spectrometer by the authors [45] . Fig. 4 shows the experimental setup of the customized Echelle spectrometer.
Light fed by an optical fiber patch cord (NA = 0.12, core diameter = 50 μm, acting as input slit) inside the spectrometer travels from the fiber output to the first OPA mirror (with an effective focal length (EFL) of 152.4 mm). After reflecting off of the mirror's surface, the collimated light reaches the Echelle grating in a quasi-Littrow orientation. A Littrow configuration would cause the center of each diffracted order to reflect back exactly on its path, thereby returning the light to the fiber. By tilting the Echelle grating from its initial position (quasi-Littrow configuration), the fiber no longer poses an obstacle to the beam. There are several ways of inducing this kind of off-axis angle in the system. The Echelle grating can be tilted either around its grooves (perpendicular to the optical table) or around the diffraction orientation (perpendicular to grooves). The former displaces the beam laterally, while the latter shifts the beam vertically. Unlike lateral displacement, vertical displacement is the same for all diffracted orders. If displacement is the same for all wavelengths in the interval of the Echellogram, then any defocusing can be corrected by simply adjusting the imaging lens. However, different displacements result in a curved imaging plane that cannot be adjusted easily unless additional optics are used. Another way to induce vertical displacement is to tilt the Echelle grating around its surface normal. Downside vertical displacement is not helpful as it directs the light to the fiber's holder (optical post and post holder). Therefore, the Echelle grating was tilted around its surface normal in a way that yielded an upside displacement. The tilt angle can be determined by gradually tilting the grating until the beam is no longer blocked by the fiber. The beam is collimated by a second OPA mirror with the same EFL as the first OPA mirror, after passing above the fiber. The second OPA mirror was installed in an antiparallel configuration to cancel out the aberration caused by the first one [47] . Both OPA mirrors have an off-axis angle of 90 o ; this angle was chosen based on its availability on the market. However, a mirror with a lower off-axis angle (e.g., 30 o ), which was not commercially available through optics catalogs at the time of development, might slightly diminish the aberration [47] . The collimated beam reflected from the second OPA mirror's surface is directed toward a cross-disperser to separate the overlapping Echelle orders. A blazed ruled grating was used here as the cross-disperser. The grooves in the cross-disperser grating were perpendicular to those of the Echelle grating. After the cross-disperser, similar to the beam's situation after being diffracted from the Echelle grating, an off-axis angle is needed to allow the beam to reach the imaging lens without obstruction from the second OPA mirror. The degree of tilt required here is higher than that of the Echelle grating to overcome the larger obstruction of the OPA mirror (compared to the smaller fiber obstruction). A high optical density (OD) longpass filter was used to filter half of the beam's cross-section diffracted from the cross-disperser (also known as ordered sorting). The separated orders are directed to the imaging lens to form the Echellogram on the camera's pixels. The optical base plate is 90 × 60 cm 2 . The size of the setup without the camera is (l)44 cm × (w)36 cm × (h)13 cm.  The ICCD camera (PI-MAX 4, Princeton Instruments, USA) was synchronized with the laser Q-switched using TTL trigger pulses to have the proper time window for collecting plasma emission. The customized spectrometer was installed on top of a three-tier utility cart for portability. The middle tier was left empty for this study but could be used for a laser coupling system to achieve a portable LIBS setup. The bottom tier was dedicated to calibration instruments (a NIST traceable balanced Deuterium-Halogen light source for intensity calibration and some gas/vapor spectral lamps, including Mercury-Argon, Argon, Neon, and Krypton for wavelength calibration).

Specimen
Five bisected femora from porcine specimens (purchased from a local slaughterhouse) were used in this study. Each bisected tissue consisted of bone, bone marrow, fat, and muscle. From each side of each bisected specimen, 50 LIBS data points were collected from each tissue. In total, 2000 data points were collected (5 specimens × 4 tissue types × 2 sides × 50 shots).

Data analysis
The data collected from the five specimens were divided into five folds (each 400) to train a classifier in cross-validation (CV) mode. Afterward, each spectrum was normalized to its mean. Principal component analysis (PCA) was used to reduce data dimensionality while preserving as much of the data's variation as possible. The output of the PCA was used as input for a quadratic support vector machine (Q-SVM). An SVM with quadratic kernel was selected. It showed slightly better classification accuracy over other kernels. It was subsequently used to train a classifier on the training dataset and test the classifier on each fold's testing dataset. The data were analyzed using Matlab R2019b. . 5 shows the Echellogram of the Deuterium-Halogen (DH) lamp. Fig. 6 shows the measured spectrum of four calibration lamps used for calibrating the spectrometer's wavelength.

Fig
The spectrometer's resolution was in the Angstrom range, sufficient for resolving peaks in the LIBS spectra (representing biological tissues) throughout the entire bandwidth. The spectrum of tissues is shown in Fig. 7 .    After reducing the dimensionality of the data through PCA, the first six principal components (PCs) were used to feed the Q-SVM, which could classify bone from other soft tissues with 100% sensitivity and specificity (CV). The muscle was classified from other tissues with very high CV sensitivity (98%) and specificity (96.7%), as well. However, fat and bone marrow were classified with slightly lower CV sensitivity (84.6% and 78.2%) and specificity (94.1% and 96.1%).
Moreover, a receiver operating characteristic (ROC) curve was depicted and the area under the curve (AUC) was calculated ( Fig. 8 ). Table 1 shows the result of the classification.

Discussion
A few FO-LIBS systems had previously been proposed for biological tissue differentiation. However, endoscopic smart laser surgery systems require miniaturization and flexibility, as well. Therefore, a custommade, bend-insensitive bundle delivery setup with sufficient flexibility was tested, in combination with a tiny half-ball lens at the bundle's tip (please see Section 2.1 for details of fiber delivery setup). The setup's dimensions and flexibility allowed for endoscopic application. Use of a bundle prevents self-focusing, as light does not penetrate between the individual fibers within the bundle. Compared to that of a single fiber, the larger cross-section of the bundle makes it less sensitive to mechanical movements (more robust setup), offering yet another benefit.
Due to the need to enhance sensitivity in an FO-LIBS setup, a custommade Echelle spectrometer was developed and optimized for our application (please see Section 2.2 for details of the spectrometer design). We aimed to achieve the following specifications: 500 nm spectral bandwidth from 330 to 830 nm, at least 0.6 nm resolution, high optical throughput (low F-number), and compact size. Astigmatism aberrations in the spectrometer were minimized via off-axis parabolic mirrors. Coma aberrations were quelled by two collimators with an antiparallel configuration. Employing two collimators (first collimator in dual-pass mode) instead of one enabled us to maintain the Echelle grating in a quasi-Littrow configuration, with minimal deviation from the optical axis. Camera tilt compensated for longitudinal chromatic aberrations. Therefore, an appropriate resolving power was achieved within a sufficiently wide spectral bandwidth. Using off-axis mirrors eliminated the need for high focal-length designs. Therefore, we were able to keep the F-number of our design very low (i.e., 3). In the other designs, using a short focal-length necessitated tilting the mirror (producing extensive off-axis aberrations into the system). Mirrors with a 90 o degree off-axis were used in this study because of their availability. However, a lower off-axis angle (e.g., 30 o ) would decrease aberrations in the system [47] . The compact size of the spectrometer made it portable. A portable LIBS spectrometer is remarkably valuable for testing and characterizing different biological samples on-site. This is an excellent capability, especially if the target sample is potentially contagious. The customized LIBS setup was tested to differentiate bone from surrounding soft tissues, namely muscle, fat, and bone marrow. It is worth mentioning that since ablation is threshold-based, and plasma can initiate with lower energy in bone than in soft tissues [10] , finding an optimum laser and ICCD parameter was challenging. For this study, the ICCD gate was open for 1 ms, after a 1 μs delay of the laser pulse, while the laser energy applied to specimens was 130 mJ. A higher intensity spectrum could be obtained with higher energies, but it leads to saturation of the ICCD's pixels while collecting the spectrum from the bone specimen. That is why in Fig. 7 , the bone specimen yields a more intense signal than the soft tissues. The setup was able to differentiate bone from surrounding soft tissues (the requirement for endoscopic laserosteotomy) with 100% sensitivity and specificity (please see Section 2.3 and 2.4 for the details of the specimen and data analysis, respectively). The overall sensitivity and specificity for differentiating all tissues (the requirement for general endoscopic laser surgery) were 90.2% and 96.7%, respectively. The differentiation between fat and bone marrow accounts for most of the errors, as tissues have a similar chemical composition. An average classification error of 57.7% for differentiating bone marrow was reported after optoacoustic monitoring of Nd:YAG-laser-produced plasma using a Mach-Zehnder Interferometer [48] . However, averaging two subsequent data points might improve accuracy [49] . It should be noted that as bone marrow and fat are not adjacent tissues in the body, differentiating between them is less important than other tissues. The all-fiber-optic LIBS offers a stand-alone setup that does not need an additional laser for characterization, a common shortcoming of other miniaturized sensors developed for laser surgery monitoring [50] .

Conclusion
The fiber delivery system proposed herein, including the bundle and the lens, was suitable for endoscopic application thanks to its small size and bend-insensitive flexibility. However, for successful endoscopic LIBS, a proper Echelle spectrometer was required as well. The main feature of most commercially available Echelle spectrometers is the ability to achieve the highest possible resolution and bandwidth at the same time. This meets the demand for a general device that is suitable for most applications. However, since optimizing all of the spectrometer parameters simultaneously is not possible, optical throughput is typically sacrificed to achieve simultaneous high-resolution and bandwidth. Low optical throughput reduces the system's sensitivity, so it does not work correctly in low light conditions (like those of miniaturized systems). The spectrometer was designed based on the spectral interval with most significant number of LIBS peaks for tissue differentiation (330 to 830 nm), and the resolution better than distance between the two closest peaks (0.6 nm) [24] . In our design, we reduced the bandwidth and resolution compared to the levels associated with commercially available Echelle spectrometers. Therefore, we were able to increase the optical throughput of the spectrometer. Our design leads to an F-number of 3, which is a significant improvement over other designs (typically ten or more). At the same time, we achieved the spectral bandwidth required to cover the essential biological LIBS peaks, and the resolution of the system exceeded the initial requirement. The results show that the customized system was both sensitive and specific (selective) when differentiating tissues, label-free and in a single shot without sample preparation. The proposed setup is thus suitable for minimally-invasive smart laser surgery.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.