Key factors behind autofluorescence changes caused by ablation of cardiac tissue

Radiofrequency ablation is a commonly used clinical procedure that destroys arrhythmogenic sources in patients suffering from atrial fibrillation and other types of cardiac arrhythmias. To improve the success of this procedure, new approaches for real-time visualization of ablation sites are being developed. One of these promising methods is hyperspectral imaging, an approach that detects lesions based on changes in the endogenous tissue autofluorescence profile. To facilitate the clinical implementation of this approach, we examined the key variables that can influence ablation-induced spectral changes, including the drop in myocardial NADH levels, the release of lipofuscin-like pigments, and the increase in diffuse reflectance of the cardiac muscle beneath the endocardial layer. Insights from these experiments suggested simpler algorithms that can be used to acquire and post-process the spectral information required to reveal the lesion sites. Our study is relevant to a growing number of multilayered clinical targets to which spectral approaches are being applied.

Our group has recently shown that the use of autofluorescence-based hyperspectral imaging (Auf-HSI) enables the circumvention of limitations imposed by the endocardial collagen layer. This approach employs illumination of the tissue surface with UV light while acquiring grayscale images across multiple wavelengths within the visible range. This creates a three-dimensional dataset, where x and y are the two spatial dimensions, while wavelength λ stands for the third dimension. Autofluorescence spectra are extracted from each x i ,y i pixel. Then, various mathematical algorithms are used to sort the pixels based on subtle differences in their normalized spectral profiles. Thereafter, the individual pixels are assigned custom pseudocolors to distinguish the ablated from the uninjured tissue 8 .
To date, we have confirmed that Auf-HSI can reveal ablation lesions made in the left atrium of large mammals, including pigs, sheep, and cows 9 , as well as in donated human heart tissue 10 . The dimensions and the shape of the lesions delineated by Auf-HSI were in perfect agreement with the conventional post-ablation staining methods such as TTC [10][11][12][13][14] . These promising bench findings have led to our ongoing efforts to incorporate Auf-HSI technology into a percutaneous imaging catheter 11,15,16 . The design of this catheter includes a saline-filled balloon to displace optically dense blood from the endocardial surface. It also includes insertable fiber optic bundles to deliver UV light and to collect the emitted visible spectra. An alternative approach is to integrate multiple singlepoint optical sensors directly into the tip of existing ablation catheters 17 . In the latter case, the autofluorescence profiles from individual points can be obtained by simply touching the tissue without the need for an inflatable balloon to create an optical window.
This paper examined the key physical factors behind spectral changes in cardiac tissue autofluorescence profiles caused by RF ablation. Equipped with such knowledge, we then identified specific wavelength ranges where ablation-induced changes are the most pronounced and consistent, allowing us to simplify both acquisition and post-processing algorithms. Insights from our studies can be applied not only to the heart, but also to other multilayered body tissues where spectral imaging offers diagnostic promise, including the skin, endovascular or epithelial surfaces.

Materials and methods
Tissue sources and ablation procedures. To fully cover the range of endocardial layer thickness reported in human subjects 18,19 , we performed experiments using freshly excised hearts from three different species. These included rat ventricles, where the endocardial thickness is negligible, and ending with market-age cows, where the endocardial layer in the left atrium can reach one millimeter in thickness. To test the effects of storage conditions on NADH and other muscle fluorophores, we used fresh rat ventricles, which enabled data acquisition immediately after the excision of the tissue. Bovine and porcine hearts were obtained from a local abattoir or after surgical training at the Washington Institute of Surgical Education and Research. The choice of the species was based on our published study in which we compared atrial tissues from market age animals to human samples 9 . Different aspects of human atrial anatomy had similarities to either porcine or bovine samples. For example, wall thickness was similar to values from market-aged pigs, while endocardial layer thickness was similar to that of a cow. The explanted hearts were transported to the laboratory on ice within a 2-3 h window after the excision, followed by dissection to expose the relevant surfaces to be ablated. RF energy was delivered with a non-irrigated ablation catheter (EP Technologies, Boston Scientific, Marlborough, MA, USA). The 4 mm ablation tip was placed perpendicular to the endocardial surface, with ablation durations varying from 5 to 30 s and tip temperatures ranging between 50 and 70 °C. These settings created lesions similar in size to those placed during clinical RF ablation therapy, as detailed previously 8 .
All animal protocols were approved by the George Washington University Institutional Animal Care and Use Committee. Experiments were performed in accordance with the United States Association for Assessment and Accreditation of Laboratory Animal Care guidelines and regulations. Imaging hardware. The LED source (Precision LED Spotlight from Mightex, Pleasanton, CA-either 365 nm UV or 5500 K cool white) was placed ~ 5 cm away from the tissue surface and positioned to reduce specular reflection. Hyperspectral datasets were acquired using a commercial HSI system (Nuance FX, PerkinElmer/ Cri, Waltham, MA, USA) fitted with Nikon AF Micro-Nikkor 60 mm f/2.8D lens. The Nuance FX system comprises a liquid crystal tunable filter (CRi LCTF) and a monochromatic charged coupled device (Sony ICX285 CCD). The Nuance FX system can capture wavelengths between 420 and 720 nm with a spatial resolution of 1,392 × 1,040 pixels. For the trans-illumination experiments, a large piece of bovine endocardium was dissected from the underlying muscle tissue and placed on top of a flat-surfaced LED lightbox. Transmission images were acquired at 700 nm using the Nuance FX optical density mode settings. The thickness values were derived from the optical density numbers and calibrated using direct caliper measurements, as described earlier 9 .
Post-processing of hyperspectral imaging datasets. The Nuance FX software package was used to perform supervised linear unmixing, with green and red pseudocolors assigned to ablated and unablated tissues, respectively. A typical signal processing protocol involved the extraction of spectra from regions of interest from the unablated and ablated tissues, followed by spectra normalization from 0 to 1 and then finding the difference between them. This difference was then corrected for the spectral sensitivity of the CCD and CRi LCTF of the Nuance FX system to facilitate compatibility of our conclusions to other acquisition systems. A detailed description of these calculations can be found in our earlier publication 8  www.nature.com/scientificreports/ coefficient was used to determine the statistical significance of the relationships between the endocardial layer thickness and the amplitude of the DBNS at individual wavelengths.

Results
Major fluorophores of cardiac tissue before and after RF ablation. Cardiac muscle is covered by epicardial and endocardial layers of connective tissue, made predominantly from interwoven collagen fibers. In this study, we present data mainly from the endocardial surface of the left atrium (LA), where most ablations take place clinically and where the endocardium is the thickest. Spectral responses from ablation sites in other cardiac locations, including ventricular sites or epicardial surfaces, were conceptually similar (data not shown). Figure 1a,b illustrate the general structure of the cardiac wall to be ablated. Two structural components of the heart surfaces are facing the ablation catheter: a layer of connective tissue and a layer of cardiac muscle beneath it. When sufficient RF energy is applied, it penetrates all the way to the underlying muscle with the goal of destroying the cells that act as arrhythmogenic culprits. The major endogenous fluorophore inside the muscle cells is mitochondrial NADH with excitation/emission peaks of around 350 and 460 nm, respectively 20,21 . Once the integrity of the cardiac muscle cells is destroyed due to RF or other means of ablation, the NADH levels irreversibly decline 13 . The secondary fluorophores found in cardiac muscle cells include flavins, flavoproteins, lipids, lipofuscin, and lipid-protein conjugates 18,22 . They fluoresce at longer wavelengths (< 500 nm), but their overall signal from the unablated muscle is weaker compared to the NADH signal. Hereafter, we will refer to them simply as 'yellow pigments' . When looking at the muscle spectra shown in Fig. 1d, one can also notice an apparent 'peak' around 630 nm. This 'peak' is caused by an increased photon absorption by heme-containing myoglobin. Muscles appear red under white light illumination due to an abundance of myoglobin which absorbs 'green' and 'yellow' photons from 520 to 600 nm range 23 . The presence of myoglobin also creates a dip in the same 'green-yellow' range of the spectrum emitted by NADH and other cardiac UV excitable fluorophores, leading to the apparent 630 nm 'peak' .
The main structural component of the endocardial layer is collagen. The latter is a highly fluorescent molecule with a broad absorption peak of around 300 nm 24 . Even when illuminated with 350-370 nm UVA light (which is closer to the absorption peak of NADH), the intensity of the endocardial layer autofluorescence is still much higher than that of the muscle layer beneath (Fig. 1c). At this excitation range, the normalized emission spectra of these two layers largely overlap (Fig. 1d). When the endocardial collagen layer is heated during the RF ablation process, the amplitude of its autofluorescence transiently declines, only to quickly return to the original levels once the tissue cools down. As a result, when the LA endocardial surface is imaged post-ablation, the changes in its autofluorescence spectrum are minimal. www.nature.com/scientificreports/ Paradoxically though, while the presence of the endocardial collagen layer obscures the drop in muscle NADH, it does, on the other hand, help to reveal an increase in the light scattering by the ablated muscle below. These two factors affect the difference between the normalized spectral profiles of native and ablated LA tissue. To understand the mechanism better, let us first examine how such difference traces were derived.
Differences between normalized spectra of native versus ablated cardiac tissues. In the absence of a collagen layer, the differences in the autofluorescent profiles of native and ablated tissues illuminated with UV light are very pronounced. To illustrate this, Fig. 2a shows the visual appearance of a crosssectioned bovine ventricular muscle illuminated with UV light. RF ablation of muscle leads to a prominent drop in the 450-480 nm range of the autofluorescence spectrum, which makes lesions appear yellowish on a blue background. The graph on the bottom left shows the raw spectral profiles taken from eight regions of interest (ROIs): four from the lesion sites and four from the proximal unablated tissue. The black traces on the right show the differences between normalized spectra from the ablated and the adjacent unablated sites. We refer to these traces as DBNS. They are obtained by normalizing the raw spectra and adjusting them for the spectral sensitivity of the Nuance FX hardware. The greater the amplitudes of the DBNS and the more similar they are to each other, the better the outcomes of the post-acquisition unmixing. When ablations are done on ventricular muscle, the DBNS are very pronounced and uniform (Fig. 2a). The use of hyperspectral imaging enhances the visual appearance of the lesion sites, although they can also be observed by an unaided eye (compare Fig. 2a top left vs. top right images).
The situation changes considerably when RF lesions are made on cardiac muscle covered by a collagen layer thicker than ~ 100 μm. This is true for large mammals, including humans, where collagen layer thickness varies across the LA endocardial surface, ranging from 0.1 to 0.8 mm 9 . Figure 2b shows an extreme case of DBNS from lesions made on bovine LA, where collagen layer thickness approaches 1 mm. The amplitudes of the raw autofluorescence profiles from the individual pixels become much more variable ( . This was possible because the initial raw traces were acquired from the lesions made on the same piece of tissue. Therefore, all other conditions that can affect spectral difference between ablated and unablated tissues (and, consequently, the shape of DBNS), such as dissimilarities in tissue composition from different animals, time since heart excision, and storage conditions, were identical.
However, much larger DBNS variability is expected between the samples from different individuals. This is because neither the subject nor the imaging conditions will be identical. Figure 3 illustrates such variability by showing the DBNS of different lesions from four porcine and bovine LAs. Such variability in DBNS traces interferes with the ability to use pre-acquired spectral libraries or other processing algorithms to reliably classify the pixels. Therefore, it is important to understand the main factors that can contribute to DBNS variability. By taking these factors into account, better protocols to acquire Auf-HSI datasets can be designed, whether for bench experiments or in vivo ablation procedures.
Main factors that impact changes in the autofluorescence spectrum of ablated cardiac surface. Figure 4A depicts our current understanding of the interplay between multiple variables impacting the overall spectral response to the illumination of cardiac surfaces with UV light. The thickness of the upper collagen layer determines the intensity of its autofluorescence. The latter acts as a secondary light source that illuminates the muscle layer below. The optical response of the muscle layer depends on the degree of its thermal damage. Specifically, lesion severity determines: (1) the amplitude of changes in the muscle autofluorescence caused by the loss of NADH and the rise in yellow pigments and (2) the degree with which the surface of ablated muscle reflects back photons coming from the endocardial collagen layer on top of it. The latter is a product of the layer thickness and collagen autofluorescence intensity. Any reflected taillight from the UV source might also be at play. The total amount of light coming back to the detector from the muscle layer is then affected by the thickness of the endocardial layer through which these photons must travel. In the next sections, we will present the results of experiments that examined how individual factors, denoted by the colored arrows in Fig. 4a,b, can affect the shape of DBNS. In these sections, we will be referring back to the Fig. 4c-e cartoons that depict the overall directions of changes that these individual factors impose on DBNS traces.
Ablation-induced drop in NADH autofluorescence. As reported by many, including our group, ischemia leads to increased levels of NADH fluorescence in both blood and saline-perfused heart preparations [25][26][27] . This is because only the reduced form of the molecule exhibits significant autofluorescence and the latter accumulates when oxygen is not available to pass electrons down the mitochondrial redox chain. After heart tissue is excised from an animal and/or taken off the perfusion system, its muscle cells gradually die, leading to a steady decline in tissue NADH levels. The rate of such decline depends on both the duration of storage and on temperature. To illustrate this effect, we ablated freshly excised saline-perfused rat ventricles, cut them into semi-equal pieces and kept them at 4, 22, and 37 °C while acquiring spectra at multiple time points. The levels of NADH autofluorescence remained stable for hours when the samples were kept at room temperature (Fig. 5) and for several days when the samples were kept on ice. By contrast, background autofluorescence rapidly declined in samples stored at 37 °C and, with it, the difference between the ablated and unablated tissues. Therefore, if tissue samples are not stored or transferred to the lab at the proper temperature, the NADH levels would start out low at the beginning of the experiment. Consequently, a much smaller drop in the 450-480 nm range will be www.nature.com/scientificreports/ recorded during the on-the-bench ablation experiment. On the other hand, if one performs ablation on a piece of freshly excised heart that is in the process of becoming ischemic, the ablation-induced NADH drop can be exaggerated. www.nature.com/scientificreports/ Therefore, the exact amplitude of the NADH drop is affected by multiple factors including: (1) amount of applied RF energy, (2) degree of tissue ischemia, and (3) storage conditions, including duration and temperature. However, regardless of its exact amplitude, NADH loss will always impact DBNS traces in the way shown in Fig. 4c.
Ablation-induced release of flavins and lipid peroxidation products. RF energy heats tissue, which enhances the formation and release of secondary fluorophores from cardiac muscle cells. To demonstrate this process directly, we performed an experiment shown in Fig. 6. RF lesions were made on the left ventricles of freshly excised saline-perfused rat hearts. The hearts were immersed in saline and Auf-HSI datasets were acquired immediately at 5 and 24 h after ablation. At the end of 24 h, the media was replaced by fresh saline and the samples were re-imaged. The autofluorescence spectra were then extracted from the pixels corresponding to the unablated (red trace) and ablated (green trace) surfaces, as well as from the media (gray trace). The representative images and spectra next to them show an immediate ablation-induced drop in NADH, followed by a slow release of heat-induced yellow pigments into the media. At the end of 24 h, the spectrum of the media becomes identical to the spectra of the ablated muscle. Multiple compounds are believed to be responsible for the heat-induced appearance of yellow autofluorescent products, including flavins 28,29 , lipofuscins 30 , lipid-protein additives, and peroxidation products 31,32 . Their formation contributes to the DBNS elevation in the 520-600 nm range, an effect observed in all tested species. The direction of wavelength-specific changes in the DBNS traces caused by the heat-induced release of yellow pigments is shown in Fig. 4d. Reflection of endocardial collagen autofluorescence by the underlying muscle layer. As we and others have previously reported 14,33 , RF ablation increases the diffuse reflectance of the muscle due to an increase in its scattering coefficient, primarily due to protein coagulation. To directly illustrate the effect of increased reflectance by the bottom layer on DBNS traces, we placed a piece of dissected bovine endocardium on top of a reflective tape, followed by Auf-HSI. The reflective tape beneath the endocardial layer led to a small increase in overall signal amplitude and a small, yet consistent red shift in the normalized spectra of the returning spectra, which enabled the successful classification of the pixels on top of the reflective tape (Fig. 7a). To make these two effects more visually evident to the reader, we conducted a similar experiment using a broad-   www.nature.com/scientificreports/ band white LED light (Mightex 5500K) instead of UV LED. In the former case, it was not the autofluorescence of endocardial collagen layer, but photons from white LED that penetrated the layer and got reflected back. Figure 7b shows an increase in the overall amplitude of the reflected light on the side of the reflective tape, with an elevated amount of returning photons at longer wavelengths visually evident (compare images from the 480 nm vs. 700 nm spectral bands extracted from a hyperspectral dataset). We argue that something similar, albeit smaller in amplitude, occurs when a layer of collagen is present on top of the ablated muscle. Specifically, the ablation-induced increase in muscle diffuse reflectance causes: (1) increased contribution of collagen layer autofluorescence to the amplitude of the overall spectrum and (2) a red shift in the normalized spectral profiles. The latter is due to an increased proportion of photons with longer wavelengths that can pass through the collagen layer on their way back to the detector. The overall direction of Figure 6. Formation and release of yellow pigments. Immediately after excision from an anesthetized rat, the heart was perfused with saline and RF ablated. Auf-HSI datasets were taken at different time points while the hearts were kept in saline at room temperature. The traces next to the RGB images are extracted Auf-HSI spectra from shown ROIs placed on native and ablated muscles and the surrounding medium. They show a gradual decline in tissue NADH levels and the release of yellow pigments into the medium. Scale bar 5 mm. Representative images and traces from three separate experiments are shown. www.nature.com/scientificreports/ wavelength-specific changes in DBNS traces caused by the increased reflection of endocardial collagen autofluorescence by the ablated muscle below is shown in Fig. 4e.

Impact of endocardial collagen layer thickness.
To quantify how the thickness of the endocardial collagen layer impacts the amplitude of the acquired spectra and, consequently, the shape of the DBNS traces, we conducted two sets of measurements. First, we quantified the relationship between the thickness of the endocardial layer and the intensity of its autofluorescence. The endocardium was carefully dissected from the underlying muscle and folded in several places. This was followed by thickness assessment and concurrent Auf-HSI. The thickness values at each pixel were obtained from the optical density measurements using a trans-illumination setup. Maximum autofluorescence intensity was then derived from the Auf-HSI dataset. The data shown in Fig. 8 confirmed that autofluorescence intensity is nearly linearly proportional to the endocardial layer thickness-up to almost 1 mm. Considering that the amplitude of muscle autofluorescence is 4-5 times lower than that of endocardial collagen (see Fig. 1d, middle), these measurements implied that the absolute values of the peak autofluorescence across the LA surface can serve as a first-order quantitative indicator of endocardial layer thickness. The second experiment involved a detailed analysis of an ablation lesion made on LA with a particularly uneven endocardium averaging about 200 μm in thickness (Fig. 9a). Spectra from multiple ROIs within the lesion were acquired (Fig. 9b). Values of DBNS at each wavelength were then plotted against the peak amplitude of the spectrum. As mentioned in the previous paragraph, the latter can serve as an indicator of endocardial thickness at each specific ROI. This analysis revealed a highly significant positive correlation between DBNS www.nature.com/scientificreports/ shift for wavelengths in the 420-480 nm range and the edocardial layer thickness (Fig. 9c). These data indicate that the thicker the endocardium, the higher is the elevation of the left shoulder of DBNS.

Optimization of acquisition and post-processing algorithms based on DBNS traces. A closer
examination of the 'up' and 'down' trends across the DBNS traces ( Fig. 4c-e) reveals that the three main variables have opposing effects on the 420-480 nm range. At the same time, all three variables, including drop in NADH, release of yellow pigments, and red shift in the reflected endocardial collagen spectra, lead to an elevation in the 500-600 nm range. This suggests that spectral changes in the 500-600 nm range can serve as a more reliable index to reveal ablated cardiac muscle in samples with variable thicknesses of collagen layer. Moreover, these observations imply that one can dramatically simplify the imaging and post-processing steps to reveal ablated tissue. This is shown in Fig. 10, which displays the outcomes of the spectral unmixing of Auf-HSI datasets containing 32 spectral bands vs. a simplified procedure that uses only two spectral bands. Specifically, the grayscale tiff image corresponding to 550 nm was divided by the tiff image corresponding to 490 nm. Since 490 nm is the wavelength where the intensity of most pixels is the highest, this procedure corresponds to spectra normalization and reveals the 550 nm elevation in the DBNS traces across all pixels of the image. The ablation lesions can then be displayed using either a grayscale, pseudocolor, or 3D mesh. To reduce the noise, one can also sum up images from adjacent spectral bands in the 520-580 range, which corresponds to the widening of the optical filter during the acquisition.

Discussion
We have previously shown that Auf-HSI can be used to reveal RF ablation lesions in heart tissue from rats, pigs, and cows and that it works in blood-perfused preparations 8,9,11 . We have also shown that this approach works in the highly collagenous left atrium of aged humans 10 and that it can reveal multiple targets, including myocardial scar, blood vessels, or adipose tissue accumulation 11 . Nonetheless, from several years of working with hundreds of samples, we have also noticed that spectral changes caused by RF ablation can vary significantly between individual animals and/or lesion sites, including cases when Auf-HSI was not able to reveal the ablated tissue. Thus, it was important for us to better understand the key factors that cause such variability. By doing so, we hoped to move closer to our ultimate goal of developing Auf-HSI into a reliable diagnostic tool that can be incorporated into the design of future ablation and/or imaging catheters. www.nature.com/scientificreports/ The cartoon in Fig. 4a summarizes what we believe are the main factors behind spectral changes caused by RF ablation of cardiac tissue. These include the autofluorescence of endocardial collagen layer and the fraction of the illuminating light reflected back by the underlying ablated muscle layer, the drop of NADH fluorescence in live muscle, and the formation of yellow pigments by the heated muscle. The Fig. 4b graph shows an averaged DBNS trace from RF-ablated porcine LA, with the colored arrows pointing to spectral ranges where these individual factors are at play. The formation of pigments and the red shift due to muscle scattering both act to elevate the DBNS right shoulder. The loss of NADH and the increased scattering of collagen layer autofluorescence by ablated muscle have the opposite effects on the DBNS left shoulder, making this range more prone to sample variability, which can be clearly seen in the experimental traces shown in Fig. 3.
The degree of ablation-induced NADH loss in experimental settings depends not only on the amount of delivered RF energy but also on the way the LA tissue is handled and stored before ablation. Storage conditions have a relatively small impact on ablation-induced changes in the muscle reflectance or autofluorescence spectrum of the endocardial layer. On the other hand, muscle NADH levels first briefly increase due to the tissue becoming ischemic, followed by temperature-and time-dependent decline (Fig. 5). These changes can impact the pre-ablation levels of muscle NADH and lead to an unanticipated drop in NADH signal intensity. This has important practical implications. In most ablation studies, including ours, freshly excised pig or cow hearts (or donated human specimens) are put in ice-cold saline for transportation from an abattoir or a surgery facility to the lab. However, these hearts are quite big and cooling them takes time. As a result, the atria might stay warm for several hours, even when placed in an ice bucket, due to them being positioned next to a bulky warm ventricle. Therefore, to preserve LA fluorophores, it is recommended to detach the atria from the ventricle immediately after excision to maximize their exposure to an ice-cold solution during transportation to the lab.
It is important to clarify the difference between the decrease in NADH levels in metabolically active cells vs cells that are irreversible damaged by ablation. The marked drop in NADH autofluorescence caused by RF ablation is due to irreversible loss of cell viability which can be readily confirmed by post-ablation staining using TTC. Acute cell necrosis occurs because heat from application of RF energy raises tissue temperature to above 60 °C-a critical point when proteins start to coagulate, and muscle cells die [33][34][35][36] . Therefore, ablation induced NADH loss represents a different scenario from reversible drop in NADH levels when tissue metabolic rate increases leading to consumption of reducing equivalents 25,37 . Similarly, the observed formation and release of yellow pigments after ablation is not an outcome of metabolic changes since the cells are already 'dead' . Multiple Notably, experiments shown in Figs. 5 and 6 were not designed to mimic in vivo clinical scenarios. Instead, they helped us to better understand the observed variability of experimental traces from our bench experiments. They are still relevant for clinical implementation of spectral imaging since mixed or negative results obtained due to improperly stored tissue can halt the development of catheters based on this technology.
One of the most interesting findings of this paper is the role of diffuse reflectance of an underlying muscle layer in changing the surface autofluorescence profile of the ablated LA. In essence, UV illumination causes the endocardial collagen layer to act as a 'lamp' that shines visible light onto the underlying muscle. When the latter is ablated, more of these photons are scattered back, altering the overall tissue autofluorescence profile. Therefore, although the presence of collagenous endocardium masks the ablation-induced loss in muscle NADH, it also helps to reveal the ablation-induced changes in muscle diffuse reflectance. With increasing thickness of endocardium, the amplitude of this beneficial effect first goes up and then declines, as thicker collagen layers start blocking ablation-induced changes in the muscle below.
In an interesting twist, changes in diffuse reflectance might also help Auf-HSI to estimate the depth of the ablation lesions. Indeed, our previous studies 10 revealed a strong correlation between the depth of the lesion (up to 6 mm) and the degree of spectral changes on the atrial surface (p < 0.001, R = 0.92). We believe that the observed correlation is a result of two, simultaneously occurring physical events. As the amount of RF energy increases, the lesion grows deeper. At the same time, the very surface of the heated muscle undergoes physical changes that causes further increase in scattering coefficient. The result is a predictive correlation: the more a tissue is thermally damaged from the surface, the deeper is the lesion. It is then possible to speculate that Auf-HSI can provide real-time information about the lesion's depth. Moreover, considering that the average thickness of the human atrial wall is about 3 mm 9 , it might be also able to predict lesion transmurality. Additional experiments are required to confirm these hypotheses.
Our analysis considered only what we thought to be the major variables. Several additional factors may also be at play. One of them is the presence of cells, including smooth muscle cells, myofibroblasts, and endothelial cells, within the endocardium (the term 'collagen layer" is an obvious simplification of a more complex structure) 38,39 . Applied RF energy destroys these cells, leading to the loss of their intracellular NADH. Relatively small amounts of these cells and a large amplitude of autofluorescence coming from collagen fibers mask this effect. Therefore, we did not observe any consistent changes in the profiles of endocardial layer taken from the top of the lesions. Yet it is possible that in vivo, a larger contribution from endocardial cells will be seen. Notably, one should not confuse ablation-induced NADH loss from viable cells present within the endocardium with a transient decrease in collagen autofluorescence upon heating 22 . In contrast to irreversible NADH decline in injured cells, collagen autofluorescence bounces back when the tissue cools to its original temperature. Although the latter phenomenon can be potentially useful for monitoring lesion formation while ablating 17,40 , our imaging studies www.nature.com/scientificreports/ are focused on permanent tissue damage, i.e. after the lesion has been formed and the tissue has returned to its pre-ablation temperature.
Another possible factor to consider is ablation-induced evaporation of water from the collagen layer, which occurs when ablations are performed on the excised specimen under open-air conditions. The main effect of this loss of moisture is an increase in the amplitude of both emitted and reflected light, which can make the lesion sites appear brighter 41 . Despite increased visibility to the human eye, such a drying effect does not improve lesion recognition based on spectral analysis. This is because HSI processing algorithms classify pixels based on differences in their normalized spectral profiles and not the amplitude of the spectrum. RF-induced moisture loss, however, can also lead to an increase in diffuse scattering within the endocardial layer, shifting spectral profiles to the right, similar to the effect caused by the reflective tape shown in Fig. 7. When RF ablations are performed in vivo, the impact of such drying-out effect is expected to be negligible, as water evaporation is minimized in the presence of surrounding blood and irrigating saline.
The ultimate goal of our studies is to determine how one might use Auf-HSI clinically. As such, concerns regarding levels of ultraviolet irradiation and its potential risks to the healthy tissue around the lesions has to be considered. In this regard, it is important to mention that the range of wavelengths best suitable for eliciting endogenous myocardial fluorescence is between 340 and 380 nm which falls within the UVA range of 315-400 nm. Compared to UVB or UVC, the UVA exposure causes the least damage due to its longer frequency which translates to lower photon energy. Secondly, our recent quantitative analysis 16 has shown that the levels of UV needed for percutaneous Auf-HSI imaging are well below the established thresholds that can damage the skin. Lastly, the layer of endocardial collagen essentially shields the underlying muscle cells from direct UV exposure further minimizing any potential damage.
The important practical implication of our studies is the dramatically reduced amount of spectral information needed to reveal the ablated tissue. In the past, we have attempted to approach this issue mathematically 12,42 . In those studies, we used combinatorial tools to screen all possible wavelength combinations from the Auf-HSI datasets for a minimal number of wavelengths that can still provide good-quality outcomes. Notably, these numerical studies pointed to similar spectral ranges, yet the mechanisms behind these findings remained unclear until the experiments detailed in this paper were performed. Moreover, the presented new findings point to specific wavelength ranges from which photons can be lumped together. This information can be useful for the implementation of this technology into the design of percutaneous catheters to be used clinically. The limited size of the catheter lumen dramatically decreases the amount of light (both delivered and collected), while imaging the inner surfaces of a beating heart requires fast acquisition to minimize contraction artifacts. Both constraints severely limit the number of photons that get back to the detector. Thus, the ability to lump together multiple spectral bands by using custom-designed bandpass filters can increase the intensity of the signal.
In summary, the presented experimental data help explain the main variables behind the spectral changes caused by RF ablation of cardiac muscle covered by endocardial collagen layer of variable thickness. Our findings suggest a way to optimize the acquisition and post-processing steps by using a lesser number of spectral bands. The conclusions from our experiments can be used for diagnostic imaging of the heart and other multilayered tissues such as the skin, endovascular or epithelial surfaces. Ultimately, we aim to incorporate this technology into the design of percutaneous ablation catheters, with the hope of improving the outcomes of surgical procedures to treat atrial fibrillation and other cardiac arrhythmias.

Scientific RepoRtS
| (2020) 10:15369 | https://doi.org/10.1038/s41598-020-72351-6 www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.