Review of machine learning for optical imaging of burn wound severity assessment.

Significance: Over the past decade, machine learning (ML) algorithms have rapidly become much more widespread for numerous biomedical applications, including the diagnosis and categorization of disease and injury. Aim: Here, we seek to characterize the recent growth of ML techniques that use imaging data to classify burn wound severity and report on the accuracies of different approaches. Approach: To this end, we present a comprehensive literature review of preclinical and clinical studies using ML techniques to classify the severity of burn wounds. Results: The majority of these reports used digital color photographs as input data to the classification algorithms, but recently there has been an increasing prevalence of the use of ML approaches using input data from more advanced optical imaging modalities (e.g., multispectral and hyperspectral imaging, optical coherence tomography), in addition to multimodal techniques. The classification accuracy of the different methods is reported; it typically ranges from ∼70% to 90% relative to the current gold standard of clinical judgment. Conclusions: The field would benefit from systematic analysis of the effects of different input data modalities, training/testing sets, and ML classifiers on the reported accuracy. Despite this current limitation, ML-based algorithms show significant promise for assisting in objectively classifying burn wound severity.

A Quantitative Assessment of Wound Healing With Oxygenated Micro/Nanobubbles in a Preclinical Burn Model.

BACKGROUND: Burns are devastating injuries, carry significant morbidity, and require long-term treatment or multiple reconstructive procedures. Wound healing and secondary insults caused by burn wound conversion are amendable to therapeutic intervention, where ischemia has been cited as one of the major factors (Dermatol Surg. 2008;34:1159-1169). Halting injury progression in the zone of stasis is crucial as conversion creates increased burn surface area and depth, leading to local and systemic consequences (J Burns Wounds. 2006;5:e2). Oxygen-carrying micro/nanobubbles, MNB(O2), offer a novel technology that can be used to effectively deliver oxygen to burn wounds and potentially counteract burn wound ischemia. METHODS: Topical irrigation with MNB(O2) of full-thickness burn wounds on a rodent model (n = 3) was compared against saline-treated controls (n = 3). Tissue structure (reduced scattering coefficient, µs'), oxyhemoglobin concentration (cHbO2), and tissue perfusion were quantified over the course of 28 days through spatial frequency domain imaging and laser speckle imaging. Histological samples taken at the end of the experiment were examined for evidence of wound healing. RESULTS: Findings in this preliminary study showed hastened healing with significant differences in spatial frequency domain imaging-measured µs' during wound healing (days 11-28) in MNB(O2) group. The healing "tipping point" seemed to occur at days 9 to 11 with increased collagen organization and increased cHbO2 occurring around that period confirming the gross healing improvements observed. In addition, histological evidence indicated that only the MNB(O2) burns had reached the remodeling phase by the end of 28-day study period. CONCLUSIONS: These preliminary findings propose the potential of MNB(O2) as a topical method for improving burn wound healing.

Spatial frequency domain imager based on a compact multiaperture camera: testing and feasibility for noninvasive burn severity assessment.

SIGNIFICANCE: Spatial frequency domain imaging (SFDI) is a wide-field imaging technique that provides quantitative maps of tissue optical properties. We describe a compact SFDI imager that employs a multispectral compound-eye camera. This design enables simultaneous image acquisition at multiple wavelengths. Such a device has potential for application for quantitative evaluation of superficial tissues by nonspecialists in low-resource settings. AIM: The aim of this work was to develop a compact SFDI imager for widefield imaging of in-vivo tissue optical properties and verify its ability to measure optical properties of tissue-simulating phantoms and in a preclinical model of burn wounds. APPROACH: This compound-eye imager was constructed using a CMOS sensor subdivided into multiple regions, each having a bandpass filter and objective lens. The ability of the instrument to image optical properties was compared with (1) a commercial SFDI imager and (2) a laboratory-based system. Initial validation of ability to accurately characterize optical properties was performed using a tissue-simulating optical phantom. It was then applied to an established murine model of thermal contact burn severity. In-vivo measurements of the optical properties of rat skin were performed before and after the application of burns. Histology was used to verify burn severity. RESULTS: Measurements of the tissue-simulating phantom optical properties made using the compound-eye imager agree with measurements made using the two comparison SFDI devices. For the murine burn model, the burns showed a decrease in the reduced scattering coefficient at all measurement wavelengths compared with preburn measurements at the same locations. This is consistent with previously reported changes in scattering that occur in full-thickness burns. CONCLUSION: We demonstrate the potential for SFDI to be translated into compact form factor using a compound-eye camera that is capable of obtaining multiple wavelengths channels simultaneously.

Impact of hemoglobin breakdown products in the spectral analysis of burn wounds using spatial frequency domain spectroscopy.

Burn wounds and wound healing invoke several biological processes that may complicate the interpretation of spectral imaging data. Through analysis of spatial frequency domain spectroscopy data (450 to 1000 nm) obtained from longitudinal investigations using a graded porcine burn wound healing model, we have identified features in the absorption spectrum that appear to suggest the presence of hemoglobin breakdown products, e.g., methemoglobin. Our results show that the calculated concentrations of methemoglobin directly correlate with burn severity, 24 h after the injury. In addition, tissue parameters such as oxygenation (StO2) and water fraction may be underestimated by 20% and 78%, respectively, if methemoglobin is not included in the spectral analysis.

Evaluating clinical observation versus Spatial Frequency Domain Imaging (SFDI), Laser Speckle Imaging (LSI) and thermal imaging for the assessment of burn depth.

While clinical examination is needed for burn severity diagnosis, several emerging technologies aim to quantify this process for added objectivity. Accurate assessments become easier after burn progression, but earlier assessments of partial thickness burn depth could lead to earlier excision and grafting and subsequent improved healing times, reduced rates of scarring/infection, and shorter hospital stays. Spatial Frequency Domain Imaging (SFDI), Laser Speckle Imaging (LSI) and thermal imaging are three non-invasive imaging modalities that have some diagnostic ability for noninvasive assessment of burn severity, but have not been compared in a controlled experiment. Here we tested the ability of these imaging techniques to assess the severity of histologically confirmed graded burns in a swine model. Controlled, graded burn wounds, 3cm in diameter were created on the dorsum of Yorkshire pigs (n=3, 45-55kg) using a custom-made burn tool that ensures consistent pressure has been employed by various burn research groups. For each pig, a total of 16 burn wounds were created on the dorsal side. Biopsies were taken for histological analysis to verify the severity of the burn. Clinical analysis, SFDI, LSI and thermal imaging were performed at 24 and 72h after burn to assess the accuracy of each imaging technique. In terms of diagnostic accuracy, using histology as a reference, SFDI (85%) and clinical analysis (83%) performed significantly better that LSI (75%) and thermography (73%) 24h after the burn. There was no statistically significant improvement from 24 to 72h across the different imaging modalities. These data indicate that these imaging modalities, and specifically SFDI, can be added to the burn clinicians' toolbox to aid in early assessment of burn severity.

Recovery of layered tissue optical properties from spatial frequency-domain spectroscopy and a deterministic radiative transport solver.

We present a method to recover absorption and reduced scattering spectra for each layer of a two-layer turbid media from spatial frequency-domain spectroscopy data. We focus on systems in which the thickness of the top layer is less than the transport mean free path ( 0.1 - 0.8l * ) . We utilize an analytic forward solver, based upon the N'th-order spherical harmonic expansion with Fourier decomposition ( SHEFN ) method in conjunction with a multistage inverse solver. We test our method with data obtained using spatial frequency-domain spectroscopy with 32 evenly spaced wavelengths within λ = 450 to 1000 nm on six-layered tissue phantoms with distinct optical properties. We demonstrate that this approach can recover absorption and reduced scattering coefficient spectra for both layers with accuracy comparable with current Monte Carlo methods but with lower computational cost and potential flexibility to easily handle variations in parameters such as the scattering phase function or material refractive index. To our knowledge, this approach utilizes the most accurate deterministic forward solver used in such problems and can successfully recover properties from a two-layer media with superficial layer thicknesses.

Method using in vivo quantitative spectroscopy to guide design and optimization of low-cost, compact clinical imaging devices: emulation and evaluation of multispectral imaging systems.

With recent proliferation in compact and/or low-cost clinical multispectral imaging approaches and commercially available components, questions remain whether they adequately capture the requisite spectral content of their applications. We present a method to emulate the spectral range and resolution of a variety of multispectral imagers, based on in-vivo data acquired from spatial frequency domain spectroscopy (SFDS). This approach simulates spectral responses over 400 to 1100 nm. Comparing emulated data with full SFDS spectra of in-vivo tissue affords the opportunity to evaluate whether the sparse spectral content of these imagers can (1) account for all sources of optical contrast present (completeness) and (2) robustly separate and quantify sources of optical contrast (crosstalk). We validate the approach over a range of tissue-simulating phantoms, comparing the SFDS-based emulated spectra against measurements from an independently characterized multispectral imager. Emulated results match the imager across all phantoms (<3 % absorption, <1 % reduced scattering). In-vivo test cases (burn wounds and photoaging) illustrate how SFDS can be used to evaluate different multispectral imagers. This approach provides an in-vivo measurement method to evaluate the performance of multispectral imagers specific to their targeted clinical applications and can assist in the design and optimization of new spectral imaging devices.

Evaluation of a pointwise microcirculation assessment method using liquid and multilayered tissue simulating phantoms.

A fiber-optic probe-based instrument, designed for assessment of parameters related to microcirculation, red blood cell tissue fraction (fRBC), oxygen saturation (SO2), and speed resolved perfusion, has been evaluated using state-of-the-art tissue phantoms. The probe integrates diffuse reflectance spectroscopy (DRS) at two source-detector separations and laser Doppler flowmetry, using an inverse Monte Carlo method for identifying the parameters of a multilayered tissue model. Here, we characterize the accuracy of the DRS aspect of the instrument using (1) liquid blood phantoms containing yeast and (2) epidermis-dermis mimicking solid-layered phantoms fabricated from polydimethylsiloxane, titanium oxide, hemoglobin, and coffee. The root-mean-square (RMS) deviations for fRBC for the two liquid phantoms were 11% and 5.3%, respectively, and 11% for the solid phantoms with highest hemoglobin signatures. The RMS deviation for SO2 was 5.2% and 2.9%, respectively, for the liquid phantoms, and 2.9% for the solid phantoms. RMS deviation for the reduced scattering coefficient (µs'), for the solid phantoms was 15% (475 to 850 nm). For the liquid phantoms, the RMS deviation in average vessel diameter (D) was 1 µm. In conclusion, the skin microcirculation parameters fRBC and SO2, as well as, µs' and D are estimated with reasonable accuracy.

Low-cost tissue simulating phantoms with adjustable wavelength-dependent scattering properties in the visible and infrared ranges.

We present a method for low-cost fabrication of polydimethylsiloxane (PDMS) tissue simulating phantoms with tunable scattering spectra, spanning visible, and near-infrared regimes. These phantoms use optical polishing agents (aluminum oxide powders) at various grit sizes to approximate in vivo tissue scattering particles across multiple size distributions (range: 17 to 3 µm). This class of tunable scattering phantoms is used to mimic distinct changes in wavelength-dependent scattering properties observed in tissue pathologies such as partial thickness burns. Described by a power-law dependence on wavelength, the scattering magnitude of these phantoms scale linearly with particle concentration over a physiologic range [µs'=(0.5 to 2.0 mm−1)] whereas the scattering spectra, specific to each particle size distribution, correlate to distinct exponential coefficients (range: 0.007 to 0.32). Aluminum oxide powders used in this investigation did not detectably contribute to the absorption properties of these phantoms. The optical properties of these phantoms are verified through inverse adding-doubling methods and the tolerances of this fabrication method are discussed.

Quantitative assessment of graded burn wounds in a porcine model using spatial frequency domain imaging (SFDI) and laser speckle imaging (LSI).

Accurate and timely assessment of burn wound severity is a critical component of wound management and has implications related to course of treatment. While most superficial burns and full thickness burns are easily diagnosed through visual inspection, burns that fall between these extremes are challenging to classify based on clinical appearance. Because of this, appropriate burn management may be delayed, increasing the risk of scarring and infection. Here we present an investigation that employs spatial frequency domain imaging (SFDI) and laser speckle imaging (LSI) as non-invasive technologies to characterize in-vivo burn severity. We used SFDI and LSI to investigate controlled burn wounds of graded severity in a Yorkshire pig model. Burn wounds were imaged starting at one hour after the initial injury and daily at approximately 24, 48 and 72 hours post burn. Biopsies were taken on each day in order to correlate the imaging data to the extent of burn damage as indicated via histological analysis. Changes in reduced scattering coefficient and blood flow could be used to categorize burn severity as soon as one hour after the burn injury. The results of this study suggest that SFDI and LSI information have the potential to provide useful metrics for quantifying the extent and severity of burn injuries.

Visible spatial frequency domain imaging with a digital light microprojector.

There is a need for cost effective, quantitative tissue spectroscopy and imaging systems in clinical diagnostics and pre-clinical biomedical research. A platform that utilizes a commercially available light-emitting diode (LED) based projector, cameras, and scaled Monte Carlo model for calculating tissue optical properties is presented. These components are put together to perform spatial frequency domain imaging (SFDI), a model-based reflectance technique that measures and maps absorption coefficients (µa) and reduced scattering coefficients (µs') in thick tissue such as skin or brain. We validate the performance of the flexible LED and modulation element (FLaME) system at 460, 530, and 632 nm across a range of physiologically relevant µa values (0.07 to 1.5 mm-1) in tissue-simulating intralipid phantoms, showing an overall accuracy within 11% of spectrophotometer values for µa and 3% for µs'. Comparison of oxy- and total hemoglobin fits between the FLaME system and a spectrophotometer (450 to 1000 nm) is differed by 3%. Finally, we acquire optical property maps of a mouse brain in vivo with and without an overlying saline well. These results demonstrate the potential of FLaME to perform tissue optical property mapping in visible spectral regions and highlight how the optical clearing effect of saline is correlated to a decrease in µs' of the skull.

Quantitative assessment of partial vascular occlusions in a swine pedicle flap model using spatial frequency domain imaging.

The use of tissue transfer flaps has become a common and effective technique for reconstructing or replacing damaged tissue. While the overall failure rate associated with these procedures is relatively low (5-10%), the failure rate of tissue flaps that require additional surgery is significantly higher (40-60%). The reason for this is largely due to the absence of a technique for objectively assessing tissue health after surgery. Here we have investigated spatial frequency domain imaging (SFDI) as a potential tool to do this. By projecting wide-field patterned illumination at multiple wavelengths onto a tissue surface, SFDI is able to quantify absolute concentrations of oxygenated and deoxygenated hemoglobin over a large field of view. We have assessed the sensitivity of SFDI in a swine pedicle flap model by using a controlled vascular occlusion system that reduced blood flow by 25%, 50%, 75%, or 100% of the baseline values in either the vein or artery. SFDI was able to detect significant changes for oxygenated hemoglobin, deoxygenated hemoglobin, or tissue oxygen saturation in partial arterial occlusions of at least 50% and partial venous occlusions of at least 25%. This shows SFDI is sensitive enough to quantify changes in the tissue hemoglobin state during partial occlusions and thus has the potential to be a powerful tool for the early prediction of tissue flap failure.

Quantitative fluorescence imaging of protoporphyrin IX through determination of tissue optical properties in the spatial frequency domain.

The ability to quantitatively determine tissue fluorescence is of interest for the purpose of better understanding the details of photodynamic therapy of skin cancer. In particular, we are interested in quantifying protoporphyrin IX (PpIX) in vivo. We present a method of correcting fluorescence for effects of native tissue absorption and scattering properties in a spatially resolved manner that preserves the resolution of the fluorescence imaging system, based off a homogeneous representation of tissue. Validation was performed using a series of liquid turbid phantoms having varying concentrations of absorber, scatterer, and fluorophore (PpIX). Through the quantification of tissue optical properties via spatial frequency domain imaging, an empirical model based on Monte Carlo simulations was deployed to successfully decouple the effects of absorption and scattering from fluorescence. From this we were able to deduce the concentration of the PpIX to within 0.2 µg/ml of the known concentration. This method was subsequently applied to the determination of PpIX concentration from in vivo normal skin where the model-based correction determined a concentration of 1.6 µg/ml, which is in agreement with literature.

Quantitative determination of dynamical properties using coherent spatial frequency domain imaging.

Laser speckle imaging (LSI) is a fast, noninvasive method to obtain relative particle dynamics in highly light scattering media, such as biological tissue. To make quantitative measurements, we combine LSI with spatial frequency domain imaging, a technique where samples are illuminated with sinusoidal intensity patterns of light that control the characteristic path lengths of photons in the sample. We use both diffusion and radiative transport to predict the speckle contrast of coherent light remitted from turbid media. We validate our technique by measuring known Brownian diffusion coefficients (D(b)) of scattering liquid phantoms. Monte Carlo (MC) simulations of radiative transport were found to provide the most accurate contrast predictions. For polystyrene microspheres of radius 800 nm in water, the expected and fit D(b) using radiative transport were 6.10E-07 and 7.10E-07 mm²/s, respectively. For polystyrene microspheres of radius 1026 nm in water, the expected and fit D(b) were 4.7E-07 and 5.35 mm²/s, respectively. For scattering particles in water-glycerin solutions, the fit fractional changes in D(b) with changes in viscosity were all found to be within 3% of the expected value.

Method for depth-resolved quantitation of optical properties in layered media using spatially modulated quantitative spectroscopy.

We have demonstrated that spatially modulated quantitative spectroscopy (SMoQS) is capable of extracting absolute optical properties from homogeneous tissue simulating phantoms that span both the visible and near-infrared wavelength regimes. However, biological tissue, such as skin, is highly structured, presenting challenges to quantitative spectroscopic techniques based on homogeneous models. In order to more accurately address the challenges associated with skin, we present a method for depth-resolved optical property quantitation based on a two layer model. Layered Monte Carlo simulations and layered tissue simulating phantoms are used to determine the efficacy and accuracy of SMoQS to quantify layer specific optical properties of layered media. Initial results from both the simulation and experiment show that this empirical method is capable of determining top layer thickness within tens of microns across a physiological range for skin. Layer specific chromophore concentration can be determined to <±10% the actual values, on average, whereas bulk quantitation in either visible or near infrared spectroscopic regimes significantly underestimates the layer specific chromophore concentration and can be confounded by top layer thickness.

Hybrid diffusion and two-flux approximation for multilayered tissue light propagation modeling.

Accurate and rapid estimation of fluence, reflectance, and absorbance in multilayered biological media has been essential in many biophotonics applications that aim to diagnose, cure, or model in vivo tissue. The radiative transfer equation (RTE) rigorously models light transfer in absorbing and scattering media. However, analytical solutions to the RTE are limited even in simple homogeneous or plane media. Monte Carlo simulation has been used extensively to solve the RTE. However, Monte Carlo simulation is computationally intensive and may not be practical for applications that demand real-time results. Instead, the diffusion approximation has been shown to provide accurate estimates of light transport in strongly scattering tissue. The diffusion approximation is a greatly simplified model and produces analytical solutions for the reflectance and absorbance in tissue. However, the diffusion approximation breaks down if tissue is strongly absorbing, which is common in the visible part of the spectrum or in applications that involve darkly pigmented skin and/or high local volumes of blood such as port-wine stain therapy or reconstructive flap monitoring. In these cases, a model of light transfer that can accommodate both strongly and weakly absorbing regimes is required. Here we present a model of light transfer through layered biological media that represents skin with two strongly scattering and one strongly absorbing layer.

Postoperative quantitative assessment of reconstructive tissue status in a cutaneous flap model using spatial frequency domain imaging.

BACKGROUND: The purpose of this study was to investigate the capabilities of a novel optical wide-field imaging technology known as spatial frequency domain imaging to quantitatively assess reconstructive tissue status. METHODS: Twenty-two cutaneous pedicle flaps were created on 11 rats based on the inferior epigastric vessels. After baseline measurement, all flaps underwent vascular ischemia, induced by clamping the supporting vessels for 2 hours (either arteriovenous or selective venous occlusions); normal saline was injected into the control flap and hypertonic-hyperoncotic saline solution was injected into the experimental flap. Flaps were monitored for 2 hours after reperfusion. The spatial frequency domain imaging system was used for quantitative assessment of flap status over the duration of the experiment. RESULTS: All flaps demonstrated a significant decline in oxyhemoglobin and tissue oxygen saturation in response to occlusion. Total hemoglobin and deoxyhemoglobin were increased markedly in the selective venous occlusion group. After reperfusion and the administration of solutions, oxyhemoglobin and tissue oxygen saturation in those flaps that survived gradually returned to baseline levels. However, flaps for which oxyhemoglobin and tissue oxygen saturation did not show any signs of recovery appeared to be compromised and eventually became necrotic within 24 to 48 hours in both occlusion groups. CONCLUSIONS: Spatial frequency domain imaging technology provides a quantitative, objective method of assessing tissue status. This study demonstrates the potential of this optical technology to assess tissue perfusion in a very precise and quantitative way, enabling wide-field visualization of physiologic parameters. The results of this study suggest that spatial frequency domain imaging may provide a means for prospectively identifying dysfunctional flaps well in advance of failure.

Noninvasive assessment of burn wound severity using optical technology: a review of current and future modalities.

Clinical examination alone is not always sufficient to determine which burn wounds will heal spontaneously and which will require surgical intervention for optimal outcome. We present a review of optical modalities currently in clinical use and under development to assist burn surgeons in assessing burn wound severity, including conventional histology/light microscopy, laser Doppler imaging, indocyanine green videoangiography, near-infrared spectroscopy and spectral imaging, in vivo capillary microscopy, orthogonal polarization spectral imaging, reflectance-mode confocal microscopy, laser speckle imaging, spatial frequency domain imaging, photoacoustic microscopy, and polarization-sensitive optical coherence tomography.

Investigation of a probe design for facilitating the uses of the standard photon diffusion equation at short source-detector separations: Monte Carlo simulations.

We design a special diffusing probe to investigate the optical properties of human skin in vivo. The special geometry of the probe enables a modified two-layer (MTL) diffusion model to precisely describe the photon transport even when the source-detector separation is shorter than 3 mean free paths. We provide a frequency domain comparison between the Monte Carlo model and the diffusion model in both the MTL geometry and conventional semiinfinite geometry. We show that using the Monte Carlo model as a benchmark method, the MTL diffusion theory performs better than the diffusion theory in the semiinfinite geometry. In addition, we carry out Monte Carlo simulations with the goal of investigating the dependence of the interrogation depth of this probe on several parameters including source-detector separation, sample optical properties, and properties of the diffusing high-scattering layer. From the simulations, we find that the optical properties of samples modulate the interrogation volume greatly, and the source-detector separation and the thickness of the diffusing layer are the two dominant probe parameters that impact the interrogation volume. Our simulation results provide design guidelines for a MTL geometry probe.

Wide-field spatial mapping of in vivo tattoo skin optical properties using modulated imaging.

BACKGROUND AND OBJECTIVES: Modulated imaging is a new modality capable of wide-field, spatially resolved measurement of in vivo optical properties. Based on spatial light modulation, the method is inexpensive, non-contact, and allows spatial mapping of tissue absorption and reduced scattering coefficients at any wavelength between 450 and 1,100 nm. Currently, clinicians rely on qualitative visual inspection to guide parameter selection for laser-based tattoo removal. MI provides quantitative measurements of multi-colored tattooed skin which may help guide treatment and objectively assess response. STUDY DESIGN/MATERIALS AND METHODS: We have measured the spatially varying optical properties of multi-colored tattooed skin over a 50 mm x 50 mm field of view at wavelengths ranging from 650 to 970 nm using MI. These measurements were compared to a similar field of view of non-tattooed skin from an adjacent area. RESULTS: We have determined the differentiated optical properties in vivo of multi-colored tattooed skin versus non-tattooed skin. CONCLUSIONS: MI provides spatially resolved quantitative information with potential for quantitative assessment of response to treatment and may provide guidance for laser tattoo removal in the future.