Multiwavelength Fluorescence and Diffuse Reflectance Spectroscopy for an In Situ Analysis of Kidney Stones

Abstract

This study explores the use of diffuse reflectance spectroscopy (DRS) and multiwavelength fluorescence spectroscopy for real-time kidney stone identification during laser lithotripsy. Traditional methods are not suitable for in situ analysis, so the research focuses on optical techniques that can be integrated with lithotripsy fibers. Experiments were conducted ex vivo, using DRS and multiwavelength fluorescence spectroscopy (emission–excitation matrix (EEM)) to distinguish between 48 urinary stones of three types: urate, oxalate and hydroxyapatite, with infrared spectroscopy as a reference. A classification model was developed based on EEM and DRS data. Initial classification relying solely on EEM data achieved an f1-score of 87%, which increased to 92% when DRS data were included. The findings suggest that optical spectroscopy can effectively determine stone composition during laser lithotripsy, potentially enhancing surgical outcomes via the real-time automatic optimization of laser radiation parameters.

1. Introduction

Urolithiasis is one of the most prevalent health issues worldwide. In America, it affects 13% of men and 7% of women [1]. Additionally, it is reported that in Europe, about 2000 individuals per million are diagnosed with this disease annually, which is about 0.1–0.4% of the total population [2]. Laser lithotripsy is one of the most effective therapeutic approaches for treating urolithiasis, involving the fragmentation of stones using laser infrared radiation [3,4]. Stones are known to vary in chemical composition and morphological characteristics [5,6]. In this respect, surgeons must adjust laser settings such as the frequency and energy of laser pulses manually to optimize stone fragmentation. The ability to determine the composition and structural properties of stones intraoperatively would enhance the efficiency of fragmentation and reduce the duration of surgery by enabling automatic adjustment of the optimum laser lithotripsy settings.

Common methods for analyzing stone composition include X-ray diffraction analysis, scanning electron microscopy (SEM), chemical analysis and Fourier Transform Infrared Spectroscopy (FTIR) [7,8,9,10,11,12]. However, these techniques are applicable only in laboratory settings after the stone has been surgically removed from the patient. There are also methods to determine the type of calculus, such as wet chemical analysis, thermogravimetry or laser-induced breakdown spectroscopy (LIBS) [13,14,15,16,17], which require the destruction of the sample for analysis. Thus, none of these methods can be used intraoperatively. Consequently, there is a need for the development of new methods that can determine the composition of urinary stones during laser lithotripsy procedures.

One promising approach is the use of optical spectroscopy methods, which can be implemented through a single-fiber scheme. Laser lithotripsy is typically performed using multimodal optical fibers that deliver energy to fragment the stones. Numerous studies focused on determining the composition of stones based on their Raman spectra [18,19,20,21]. It has also been shown that Raman spectroscopy can potentially be used in conjunction with a surgical optical fiber probe for diagnostics during laser lithotripsy [22]. Fluorescence spectroscopy is another technique used for the multicomponent analysis of substances, including biological tissues [23,24]. Notably, uric acid stones have been found to exhibit high fluorescence intensity when exposed to green light compared to other types of stones [25]. Recent research has also demonstrated the ability of fluorescence spectroscopy to distinguish between calcium-containing stones and those without calcium with high accuracy [26]. Furthermore, the spectral analysis of diffusely scattered light by a stone can provide information about its scattering properties which relate to optical heterogeneities within the stone and absorbing properties associated with the concentration of various chromophores. This fact suggests that the color characteristics and compositional differences in stones may be determined according to their diffuse reflectance spectra.

This study explores the feasibility of using diffuse reflectance spectroscopy (DRS) and fluorescence spectroscopy to determine the composition of kidney stones under measurement conditions that closely resemble those encountered intraoperatively. Experimental setups were built to simulate such conditions and implement probing methods for diffuse reflectance and fluorescence spectroscopy, including the measurement of three-dimensional excitation–emission matrices (EEM). Using these setups, the optical response of calculi with different compositions was measured ex vivo. Subsequently, the data obtained were used to build a classification model capable of differentiating between three types of stones—urate, oxalate and hydroxyapatite—and an assessment of the classification quality was conducted.

2. Materials and Methods

2.1. Samples

The stone samples for this study were collected from the MSU Medical Research and Education Center. All patients gave informed consent for the study before measuring the calculi. The study was approved by the Local Ethical Committee of the MSU Medical Research and Educational Center. The calculi were surgically extracted from 48 patients (patients’ ages ranged from 20 to 70 years). All stones were rinsed with distilled water and dried postoperatively. The sample sizes varied from 0.5 to 1 mm, most of them being of pure composition (i.e., no mixed chemical structures were observed within the stones). There were three types of stones in the dataset obtained: hydroxyapatite, urate and oxalate calculi.

2.2. Experimental Setups

The samples were measured using three spectroscopy methods: DRS, multiwavelength fluorescence spectroscopy (EEM) and infrared spectroscopy.

2.2.1. EEM Measurements

A custom-built experimental setup was used for EEM measurements (Figure 1A). Radiation from a plasma light source (XWS-65, 400–1000 nm, 20 W power, Troitsk Engineering Center, Moscow, Russia) passed through a 550 μm diameter optical fiber (NA 0.22, Ocean Optics, Dunedin, FL, USA), which was connected to the input port of a monochromator (tuning range 300–700 nm, OceanOptics MonoScan2000, Dunedin, FL, USA). It was used to select a 10 nm wide emission band (tuning range 280–480 nm with a step of 20 nm). The emitted light from the monochromator was delivered via an optical fiber (550 μm core diameter, NA 0.22, IPG Photonics, Oxford, MS, USA) to the sample for fluorescence excitation. A second fiber (550 μm core diameter, NA 0.22, IPG Photonics, Oxford, MS, USA) was used to detect the fluorescence of the calculi. The fibers were mounted in a special holder, ensuring they were symmetrically aligned with respect to the vertical axis. This arrangement resulted in an angle of 30° between the fibers. The detected signal was recorded by a spectrometer (USB 2000, Ocean Optics, Dunedin, FL, USA; spectral resolution 0.5 nm) in the range of 400–1000 nm.

2.2.2. DRS Measurements

To record diffuse reflectance spectra, an experimental setup was built to simulate the lighting conditions in a surgery room (Figure 1B). A halogen lamp (30 W, 12 V) was used as the light source. The diffuse reflection signal from the calculus was recorded using a surgical optical fiber (IPG Photonics, core diameter 150 μm, NA 0.22) by a spectrometer (YIXIST, Zhejiang, China; spectral resolution 12 nm) in the range of 300 to 1100 nm and acquisition time of 80 ms. The distance between the sample and the fiber was 1 mm. The reference spectrum was measured using a white standard with a reflection coefficient of 99% (LabSphere, North Sutton, NH, USA) at a distance of 1 mm. The dark spectrum was recorded in the absence of illumination.

2.2.3. FTIR Spectroscopy

For infrared spectroscopy measurements, the stones were ground into a powder. Infrared spectra were recorded in the diamond attenuated total reflectance geometry (ATR) using a spectrometer (The Bruker INVENIO R FT-IR Spectrometer, Bruker, Mannheim, Germany) in the range of 200–4000 cm⁻¹ with a spectral resolution of 4 cm⁻¹ and acquisition time of 10 ms.

2.3. Data Processing

2.3.1. DRS

To estimate the optical density (OD) spectra from the diffuse reflectance spectra, the following formula was used:

$$\mathrm{OD}\left(\mathsf{\lambda }\right)=-{\log }_{10}\frac{\mathrm{I}\left(\mathsf{\lambda }\right)-{\mathrm{I}}_{\mathrm{dark}}\left(\mathsf{\lambda }\right)}{{\mathrm{I}}_{\mathrm{ref}}\left(\mathsf{\lambda }\right)-{\mathrm{I}}_{\mathrm{dark}}\left(\mathsf{\lambda }\right)},$$

(1)

where OD(λ) is the OD spectrum, I(λ) is the recorded diffuse reflectance spectrum of the calculus, I_ref(λ) is the lamp reference spectrum and I_dark(λ) is the background spectrum. Next, the OD spectra were smoothed with a median filter with a window of 8 nm, and cut off in the range of 475–500 nm.

To compare the spectra, the slope parameter was calculated using the following formula:

$$\mathrm{Slope}={\mathrm{OD}}_{450\mathrm{nm}}-{\mathrm{OD}}_{600\mathrm{nm}}$$

(2)

2.3.2. Multiwavelength Fluorescence Spectroscopy

For each fluorescence excitation wavelength, the corresponding fluorescence emission spectra were smoothed by a median filter with a 5 nm window, and the spectral band corresponding to the Rayleigh scattering was cut off. Thus, the spectra were cut off in the range of 500–850 nm.

2.3.3. FTIR Spectroscopy

The IR spectra were processed in the following way: the background was subtracted for each spectrum in order to improve the peak differentiation. The background was calculated using the adaptive iteratively reweighted penalized least-squares method. After that, the spectra were smoothed with the Savitzky–Golay filter (5 nm window, polynomial order 5) and were cut off in the range of 500–2500 cm⁻¹. Next, the types of calculi were determined from the characteristic IR peaks according to known reference spectra [27,28,29]. The typical FTIR spectra of stones characterized by different chemical composition are presented in Figure 2. The calculus was classified as oxalate, hydroxyapatite or urate if all mentioned peaks were present in the FTIR spectrum. The coordinated oxalate group’s asymmetric and symmetric COO- stretching vibrations correspond to the strong IR bands at 1610 and 1315 cm⁻¹. The bands at 777 and 514 cm⁻¹ in the fingerprint region can be identified as the O−C=O bending in-plane and CO₂ wagging mode, respectively. The phosphate group’s stretching and bending vibrations, corresponding to hydroxyapatite, can be assigned to the bands at 1020 and 561 cm⁻¹. The C−C stretching frequencies can be attributed to the sharp bands at 1121 and 990 cm⁻¹, while the relatively broad and intense band at 1670 cm⁻¹ can be assigned to the asymmetric C=O vibration of the urate calculus [30,31,32].

Some calculi had a mixed composition, i.e., the IR spectra contained peaks characteristic of several types of calculi. In this case, the calculus was assigned to the type with the dominating intensity of the characteristic peaks.

2.4. Classification

Two mathematical models were built to classify three types of calculi: based on EEM alone, and on both EEM combined with diffuse reflectance spectra.

The pipeline for the classification included the following steps. Before classification, the spectra preprocessing described above was performed. For the fluorescence spectra, dimensionality reduction was performed for each excitation wavelength using principal component analysis (PCA). The reduced dimensionality matrices for different excitation wavelengths were then combined into a single matrix and standardization was performed. The principal component analysis was reapplied to the resulting matrix. The final projections were used for classification.

The pipeline for processing the EEM together with the diffuse reflectance spectra was constructed in a similar way with the following modification. The data were pre-standardized. The dimensionality of the matrix corresponding to the DRS spectra of the samples was downsampled using the PCA method and then combined with the corresponding matrix obtained by combining the downsampled matrices for the fluorescence spectra. The dimensionality of the merged matrix was also downsampled using the PCA method.

Next, multiclass classification was performed (three classes: hydroxyapatite, oxalate and urate). The BalancedRandomForest method from the imbalanced library was chosen as a classification model to account for class imbalance (the minor class being hydroxyapatite) due to considerations of model stability when training on a small amount of data. FTIR spectroscopy data were used as reference to evaluate classification quality. For training, 60% of the data for each class was selected, and then the model was tested on the remaining data (train set).

Leave-One-Out cross-validation was performed to select the optimal hyperparameters of the described pipeline (number of PCA components, max_depth for the classifier), and the f1-score metric was optimized. The f1-score can be interpreted as a harmonic mean of the precision and recall, and the arithmetic mean of the f1-score for each class was calculated for the final result.

3. Results

Stone types defined via FTIR spectroscopy were used as a reference standard for classification based on EEM and DRS data. The samples included 6 hydroxyapatite, 13 urate and 29 oxalate stones.

Characteristic fluorescence emission spectra for excitation wavelengths that exhibited the largest variance are presented in Figure 3A–C. The fluorescence emission spectra of urate stones differ significantly from those of other types due to the red-shifted maximum of the fluorescence band. The EEMs of oxalate stones exhibited a minor secondary peak at an emission wavelength of around 620 nm and excitation wavelength of 420 nm, which was absent in stones of other types.

Each calculus was measured at five points at least. Figure 4A illustrates representative normalized average fluorescence spectrum from a single hydroxyapatite stone. The dispersion of the fluorescence spectrum, due to the heterogeneity of the concretion, was significantly less than the discrepancies in the fluorescence spectra resulting from varying stone compositions (Figure 4B).

The averaged OD spectra of the three types of stones are shown in Figure 5A. Since optical density is the minus logarithm of reflectance, Figure 5A suggests that the difference in reflectance in the red (600 nm) and the blue (450 nm) region is more pronounced for urate stones in comparison with oxalate and hydroxyapatite stones, which is connected to the color of urate stones, which are mainly yellow, while most oxalate and hydroxyapatite stones tend to be black or gray. As can be seen, the average slope of the OD spectra of urate stones within the interquartile range is significantly larger than that of the spectra from other types of stones (Figure 5B).

During cross-validation, optimal hyperparameters were selected for stones classification based on the EEM data. The optimum number of PCA components for decomposing fluorescence spectra for each excitation wavelength was set to five, and for decomposing reduced-dimensionality matrix to three components. For classifying EEMs in conjunction with the OD spectra, the optimal number of PCA components for decomposing both fluorescence and diffuse reflectance spectra was also found to be equal to five; and for decomposing the reduced-dimensionality matrix, the optimum number of components was five as well. To assess classification quality, confusion matrices were calculated for both the training and test datasets, and the f1-score was calculated on the test data (Figure 6).

The value of the f1-score metric for EEM classification was 87%. When diffuse reflectance spectra were added, the classification quality increased and the f1-score reached 92%.

4. Discussion

Previous studies, such as the one by Xing et al., have utilized EEMs for two types of classification: distinguishing calcium stones (calcium oxalate monohydrate and dihydrate) from other types of stones (urate and struvite) [26]. The presented study demonstrated classification accuracies exceeding 90% for the multi-class task. The method proposed here differs from that in [26], where binary classification based solely on the EEMs of stones was performed. In contrast, our model is multi-class and achieves high accuracy by combining two methods of optical spectroscopy. Moreover, to the best of our knowledge, the combination of DRS and EEM for stone classification has not been assessed previously. The accuracy of our classification model using both EEMs and diffuse reflectance spectra reached 92%, with the addition of DRS data enhancing the accuracy metric relative to measurements using only EEMs.

The Implemented scheme requires only minor modifications for measurements using a single fiber, meaning it can be adapted for equipment used by surgeons during endoscopic operations. Since the excitation wavelength range (260–480 nm) and the fluorescence spectra registration range (above 500 nm) do not overlap, a simple setup using a dichroic mirror with a cutoff wavelength around 500 nm is sufficient along the lines of Xing et al.’s study [26]. Diffuse reflectance spectra in the 450–600 nm range can also be measured using a single fiber with endoscope illumination. The most time-consuming part, EEM measurement, took less than 5 s in our case. A similar measurement time can be expected when measuring using an endoscope, which is an acceptable value since stone removal can take tens of minutes, so type assessment can be carried out much faster. The study in [33] presented an example of the implementation of DRS and fluorescence spectroscopy experimental setup using endoscope, and evaluated the possibility of the application of these methods during lithotripsy. Thus, such measurements could be feasible in a surgery room using an endoscope and a single multimode fiber, which could then be used for stone fragmentation.

Currently, dual-energy computed tomography (DECT) is used to determine the chemical composition of stones and optimize lithotripsy regimes. This method distinguishes only urate stones from other types [34,35]. The sensitivity and specificity of DECT for identifying urate stones are 98% and 89%, respectively, corresponding to an f1-score of 92%. Our proposed method maintains the accuracy of identifying urate stones and also allows for the determination of oxalate and hydroxyapatite stone types. Thus, it could potentially optimize lithotripsy procedures further while avoiding the radiation exposure associated with DECT. Additionally, optical spectroscopy measurements can be conducted during the stone removal operation, which could allow for the real-time refinement of stone type classification, particularly if the stone is significantly heterogeneous. This could enable the adaptation of laser parameters during the surgery, reducing the time needed for the procedure. Therefore, further development of this method with a focus on adapting the scheme for in situ measurements during surgeries presents significant interest.

Finally, we would like to address an interesting yet enigmatic question about the origin of fluorophores responsible for the observed fluorescence in kidney stones. The authors of [18] studied the optical response of urinary stones, focusing on struvite and calcium oxalate monohydrate (COM) ones. An important fact was that the fluorescence signal was observed in the region of COM samples ex vivo, while no fluorescence signal was recorded from the synthetic COM powder. This fact suggests that fluorescence is likely predominantly generated by pigments and organic macromolecules that are integrated into the COM lattice during stone growth. Moreover, it was shown that different layers of the COM provided different fluorescence intensities that can be associated with changes in pigment/macromolecule density among the layers.

Hence, the general concept is that stone fluorescence is due to the emission of organic compounds with heterogeneous distribution and diverse optical properties within a sample. We note that the concept of oxidation-induced autofluorescence yielding broadband fluorescence emission with properties similar to that of stones was previously proposed by our group [36]. From this point of view, differences in organic compounds and related pigments incorporated into stones yield differences in the observed spectra. This is the origin of the differences observed for uric stones. Namely, oxidation and an accumulation of uric acid and other specific molecules may result in higher fluorescence and shifted emission. Similar explanations are presented in a recent paper on stones autofluorescence [26], where it is stated that “a plausible explanation is that stones with different inorganic substances contain distinct classes of matrix proteins, the latter of which, likely played a major role in shaping the spectra as we already show that pure inorganic substances exhibit very weak autofluorescence”. Some additional considerations regarding the fluorescence of organic components within stones can be found in [25], where discrimination between organic and inorganic fractions was performed based on non-linear microscopy.

5. Conclusions

This study demonstrated the potential of diffuse reflectance spectroscopy (DRS) and multiwavelength fluorescence spectroscopy as viable methods for determining the composition of urinary stones during laser lithotripsy procedures. The classification model developed from the ex vivo experimental data showed a high accuracy for differentiating between urate, oxalate and hydroxyapatite stone types. The integration of EEM and DRS data into the classification process resulted in an f1-score of 92%, indicating an improvement over using EEM data alone. These findings suggest that the implementation of optical spectroscopy methods could enhance intraoperative decision-making and potentially improve patient outcomes by allowing for the real-time analysis of urinary stone composition.