Machine learning for direct oxygen saturation and hemoglobin concentration assessment using diffuse reflectance spectroscopy

Designing a use-error robust machine learning model for quantitative analysis of diffuse reflectance spectra

Significance

Machine learning (ML)-enabled diffuse reflectance spectroscopy (DRS) is increasingly used as an alternative to the computation-intensive inverse Monte Carlo (MCI) simulation to predict tissue's optical properties, including the absorption coefficient, μ a and reduced scattering coefficient, μ s ' .

Aim

We aim to develop a use-error-robust ML algorithm for optical property prediction from DRS spectra.

Approach

We developed a wavelength-independent regressor (WIR) to predict optical properties from DRS data. For validation, we generated 1520 simulated DRS spectra with the forward Monte Carlo model, where μ a = 0.44 to 2.45    cm - 1 , and μ s ' = 6.53 to 9.58    cm - 1 . We introduced common use-errors, such as wavelength miscalibrations and intensity fluctuations. Finally, we collected 882 experimental DRS images from 170 tissue-mimicking phantoms and compared performances of the WIR model, a dense neural network, and the MCI model.

Results

When compounding all use-errors on simulated data, the WIR model best balanced accuracy and speed, yielding errors of 1.75% for μ a and 1.53% for μ s ' , compared to the MCI's 50.9% for μ a and 24.6% for μ s ' . Regarding experimental data, WIR model had mean errors of 13.2% and 6.1% for μ a and μ s ' , respectively. The errors for MCI were about eight times higher.

Conclusions

The WIR model presents reliable use-error-robust optical property predictions from DRS data.

Neural network-based inverse model for diffuse reflectance spectroscopy

In diffuse reflectance spectroscopy, the retrieval of the optical properties of a target requires the inversion of a measured reflectance spectrum. This is typically achieved through the use of forward models such as diffusion theory or Monte Carlo simulations, which are iteratively applied to optimize the solution for the optical parameters. In this paper, we propose a novel neural network-based approach for solving this inverse problem, and validate its performance using experimentally measured diffuse reflectance data from a previously reported phantom study. Our inverse model was developed from a neural network forward model that was pre-trained with data from Monte Carlo simulations. The neural network forward model then creates a lookup table to invert the diffuse reflectance to the optical coefficients. We describe the construction of the neural network-based inverse model and test its ability to accurately retrieve optical properties from experimentally acquired diffuse reflectance data in liquid optical phantoms. Our results indicate that the developed neural network-based model achieves comparable accuracy to traditional Monte Carlo-based inverse model while offering improved speed and flexibility, potentially providing an alternative for developing faster clinical diagnosis tools. This study highlights the potential of neural networks in solving inverse problems in diffuse reflectance spectroscopy.

Workload and sex effects in comprehensive assessment of cutaneous microcirculation

INTRODUCTION

Workload and sex-related differences have been proposed as factors of importance when evaluating the microcirculation. Simultaneous assessments with diffuse reflectance spectroscopy (DRS) and laser Doppler flowmetry (LDF) enable a comprehensive evaluation of the microcirculation. The aim of the study was to compare the response between sexes in the microcirculatory parameters red blood cell (RBC) tissue fraction, RBC oxygen saturation, average vessel diameter, and speed-resolved perfusion during baseline, cycling, and recovery, respectively.

METHODS

In 24 healthy participants (aged 20 to 30 years, 12 females), cutaneous microcirculation was assessed by LDF and DRS at baseline, during a workload generated by cycling at 75 to 80 % of maximal age-predicted heart rate, and recovery, respectively.

RESULTS

Females had significantly lower RBC tissue fraction and total perfusion in forearm skin microcirculation at all phases (baseline, workload, and recovery). All microvascular parameters increased significantly during cycling, most evident in RBC oxygen saturation (34 % increase on average) and perfusion (9-fold increase in total perfusion). For perfusion, the highest speeds (>10 mm/s) increased by a factor of 31, whereas the lowest speeds (<1 mm/s) increased by a factor of 2.

CONCLUSION

Compared to a resting state, all studied microcirculation measures increased during cycling. For perfusion, this was mainly due to increased speed, and only to a minor extent due to increased RBC tissue fraction. Skin microcirculatory differences between sexes were seen in RBC concentration and total perfusion.

Machine Learning Diffuse Optical Tomography Using Extreme Gradient Boosting and Genetic Programming

Diffuse optical tomography (DOT) is a non-invasive method for detecting breast cancer; however, it struggles to produce high-quality images due to the complexity of scattered light and the limitations of traditional image reconstruction algorithms. These algorithms can be affected by boundary conditions and have a low imaging accuracy, a shallow imaging depth, a long computation time, and a high signal-to-noise ratio. However, machine learning can potentially improve the performance of DOT by being better equipped to solve inverse problems, perform regression, classify medical images, and reconstruct biomedical images. In this study, we utilized a machine learning model called "XGBoost" to detect tumors in inhomogeneous breasts and applied a post-processing technique based on genetic programming to improve accuracy. The proposed algorithm was tested using simulated DOT measurements from complex inhomogeneous breasts and evaluated using the cosine similarity metrics and root mean square error loss. The results showed that the use of XGBoost and genetic programming in DOT could lead to more accurate and non-invasive detection of tumors in inhomogeneous breasts compared to traditional methods, with the reconstructed breasts having an average cosine similarity of more than 0.97 ± 0.07 and average root mean square error of around 0.1270 ± 0.0031 compared to the ground truth.

Speed-resolved perfusion imaging using multi-exposure laser speckle contrast imaging and machine learning

SIGNIFICANCE

Laser speckle contrast imaging (LSCI) gives a relative measure of microcirculatory perfusion. However, due to the limited information in single-exposure LSCI, models are inaccurate for skin tissue due to complex effects from e.g. static and dynamic scatterers, multiple Doppler shifts, and the speed-distribution of blood. It has been demonstrated how to account for these effects in laser Doppler flowmetry (LDF) using inverse Monte Carlo (MC) algorithms. This allows for a speed-resolved perfusion measure in absolute units %RBC × mm/s, improving the physiological interpretation of the data. Until now, this has been limited to a single-point LDF technique but recent advances in multi-exposure LSCI (MELSCI) enable the analysis in an imaging modality.

AIM

To present a method for speed-resolved perfusion imaging in absolute units %RBC × mm/s, computed from multi-exposure speckle contrast images.

APPROACH

An artificial neural network (ANN) was trained on a large simulated dataset of multi-exposure contrast values and corresponding speed-resolved perfusion. The dataset was generated using MC simulations of photon transport in randomized skin models covering a wide range of physiologically relevant geometrical and optical tissue properties. The ANN was evaluated on in vivo data sets captured during an occlusion provocation.

RESULTS

Speed-resolved perfusion was estimated in the three speed intervals 0 to 1    mm / s , 1 to 10    mm / s , and > 10    mm / s , with relative errors 9.8%, 12%, and 19%, respectively. The perfusion had a linear response to changes in both blood tissue fraction and blood flow speed and was less affected by tissue properties compared with single-exposure LSCI. The image quality was subjectively higher compared with LSCI, revealing previously unseen macro- and microvascular structures.

CONCLUSIONS

The ANN, trained on modeled data, calculates speed-resolved perfusion in absolute units from multi-exposure speckle contrast. This method facilitates the physiological interpretation of measurements using MELSCI and may increase the clinical impact of the technique.

VIS-NIR Diffuse Reflectance Spectroscopy System with Self-Calibrating Fiber-Optic Probe: Study of Perturbation Resistance

We report on the comparative analysis of self-calibrating and single-slope diffuse reflectance spectroscopy in resistance to different measurement perturbations. We developed an experimental setup for diffuse reflectance spectroscopy (DRS) in a wide VIS-NIR range with a fiber-optic probe equipped with two source and two detection fibers capable of providing measurements employing both single- and dual-slope (self-calibrating) approaches. In order to fit the dynamic range of a spectrometer in the wavelength range of 460-1030 nm, different exposure times have been applied for short (2 mm) and long (4 mm) source-detector distances. The stability of the self-calibrating and traditional single-slope approaches to instrumental perturbations were compared in phantom and in vivo studies on human palm, including attenuations in individual channels, fiber curving, and introducing optical inhomogeneities in the probe-tissue interface. The self-calibrating approach demonstrated high resistance to instrumental perturbations introduced in the source and detection channels, while the single-slope approach showed resistance only to perturbations introduced into the source channels.

Performance estimation of optical skin probe in short wavelength infrared spectroscopy based on Monte-Carlo simulation

Optical throughput and optical path length are key parameters to obtain high signal to noise ratio and sensor sensitivity for the detection of skin tissue components based on short wavelength infrared (SWIR) spectroscopy. These parameters should be taken into account at the stage of optical system design. We aim to develop a method to estimate the optical efficiency and the effective water path length of a newly designed SWIR spectroscopy skin measurement system using Monte-Carlo photon migration simulation. To estimate the optical efficiency and the effective water path length, we investigated the characteristics of Monte-Carlo photon migration simulation utilizing one layered simple skin model. Simulation of photon transport in skin was conducted for transmission, transflection, and reflection optical configurations in both first overtone (1540 ~ 1820 nm) and combination (2040 ~ 2380 nm) wavelength ranges. Experimental measurement of skin spectrum was done using Fourier transform infrared spectroscopy based system to validate the estimation performance. Overall, the simulated results for optical efficiency and effective water path length are in good agreements with the experimental measurements, which shows the suggested method can be used as a means for the performance estimation and the design optimization of various in-vivo SWIR spectroscopic system.

Applications of Neural Networks in Biomedical Data Analysis

Neural networks for deep-learning applications, also called artificial neural networks, are important tools in science and industry. While their widespread use was limited because of inadequate hardware in the past, their popularity increased dramatically starting in the early 2000s when it became possible to train increasingly large and complex networks. Today, deep learning is widely used in biomedicine from image analysis to diagnostics. This also includes special topics, such as forensics. In this review, we discuss the latest networks and how they work, with a focus on the analysis of biomedical data, particularly biomarkers in bioimage data. We provide a summary on numerous technical aspects, such as activation functions and frameworks. We also present a data analysis of publications about neural networks to provide a quantitative insight into the use of network types and the number of journals per year to determine the usage in different scientific fields.

Laser Doppler Flowmetry Combined with Spectroscopy to Determine Peripheral Tissue Perfusion and Oxygen Saturation: A Pilot Study in Healthy Volunteers and Patients with Peripheral Arterial Disease

Background: In this study, we assessed the ability of the EPOS system (Perimed AB, Järfälla, Stockholm, Sweden) to detect differences in tissue perfusion between healthy volunteers and patients with peripheral arterial disease (PAD) with different severity of disease. Methods: This single-center prospective pilot study included 10 healthy volunteers and 20 patients with PAD scheduled for endovascular therapy (EVT). EPOS measurements were performed at rest at 32 °C and 44 °C, followed by transcutaneous oxygen pressure (TcPo₂) measurements. The measurements were performed on the dorsal and medial side of the foot, as well as the lateral side of the calf. EPOS parameters included hemoglobin oxygen saturation (HbSo₂) and speed-resolved red blood cell (RBC) perfusion. Results: HbSo₂ at 44 °C was significantly different between the three groups for all measurement locations. The overall speed-resolved RBC perfusion at 44 °C was statistically significant between the groups on the dorsal and medial side of the foot but not on the calf. TcPo₂ values were not significantly different between the three groups. Conclusions: This study demonstrates that the EPOS system can depict differences in tissue perfusion between healthy volunteers, patients with Fontaine class IIb PAD, and those with Fontaine class III or IV PAD but only after heating to 44 °C.

Regression-based neural network for improving image reconstruction in diffuse optical tomography

Diffuse optical tomography (DOT) is a non-invasive imaging technique utilizing multi-scattered light at visible and infrared wavelengths to detect anomalies in tissues. However, the DOT image reconstruction is based on solving the inverse problem, which requires massive calculations and time. In this article, for the first time, to the best of our knowledge, a simple, regression-based cascaded feed-forward deep learning neural network is derived to solve the inverse problem of DOT in compressed breast geometry. The predicted data is subsequently utilized to visualize the breast tissues and their anomalies. The dataset in this study is created using a Monte-Carlo algorithm, which simulates the light propagation in the compressed breast placed inside a parallel plate source-detector geometry (forward process). The simulated DL-DOT system's performance is evaluated using the Pearson correlation coefficient (R) and the Mean squared error (MSE) metrics. Although a comparatively smaller dataset (50 nos.) is used, our simulation results show that the developed feed-forward network algorithm to solve the inverse problem delivers an increment of ∼30% over the analytical solution approach, in terms of R. Furthermore, the proposed network's MSE outperforms that of the analytical solution's MSE by a large margin revealing the robustness of the network and the adaptability of the system for potential applications in medical settings.

Multispectral snapshot imaging of skin microcirculatory hemoglobin oxygen saturation using artificial neural networks trained on in vivo data

SIGNIFICANCE

Developing algorithms for estimating blood oxygenation from snapshot multispectral imaging (MSI) data is challenging due to the complexity of sensor characteristics and photon transport modeling in tissue. We circumvent this using a method where artificial neural networks (ANNs) are trained on in vivo MSI data with target values from a point-measuring reference method.

AIM

To develop and evaluate a methodology where a snapshot filter mosaic camera is utilized for imaging skin hemoglobin oxygen saturation (SO2), using ANNs.

APPROACH

MSI data were acquired during occlusion provocations. ANNs were trained to estimate SO2 with MSI data as input, targeting data from a validated probe-based reference system. Performance of ANNs with different properties and training data sets was compared.

RESULTS

The method enables spatially resolved estimation of skin tissue SO2. Results are comparable to those acquired using a Monte-Carlo-based approach when relevant training data are used.

CONCLUSIONS

Training an ANN on in vivo MSI data covering a wide range of target values acquired during an occlusion protocol enable real-time estimation of SO2 maps. Data from the probe-based reference system can be used as target despite differences in sampling depth and measurement position.

Comprehensive imaging of microcirculatory changes in the foot during endovascular intervention - A technical feasibility study

Chronic limb-threatening ischemia (CLTI) has a major impact on patient's lives and is associated with a heavy health care burden with high morbidity and mortality. Treatment by endovascular intervention is mostly based on macrocirculatory information from angiography and does not consider the microcirculation. Despite successful endovascular intervention according to angiographic criteria, a proportion of patients fail to heal ischemic lesions. This might be due to impaired microvascular perfusion and variations in the supply to different angiosomes. Non-invasive optical techniques for microcirculatory perfusion and oxygen saturation imaging have the potential to provide the interventionist with additional information in real-time, supporting clinical decisions during the intervention. This study presents a novel multimodal imaging system, based on multi-exposure laser speckle contrast imaging and multi-spectral imaging, for continuous use during endovascular intervention. The results during intervention display spatiotemporal changes in the microcirculation compatible with expected physiological reactions during balloon dilation, with initially induced ischemia followed by a restored perfusion, and local administration of a vasodilator inducing hyperemia. We also present perioperative and postoperative follow-up measurements with a pulsatile microcirculation perfusion. Finally, cases of spatial heterogeneity in the observed oxygen saturation and perfusion are discussed. In conclusion, this technical feasibility study shows the potential of the methodology to characterize changes in microcirculation before, during, and after endovascular intervention.

Optical property recovery with spatially-resolved diffuse reflectance at short source-detector separations using a compact fiber-optic probe

We describe a compact fiber-optic probe (2 mm outside diameter) that utilizes spatially-resolved diffuse reflectance for tissue optical property recovery. Validation was performed in phantoms containing Intralipid 20% as scatterer, and methylene blue (MB), MnTPPS, and/or India ink as absorbers. Over a range of conditions, the reduced scattering coefficient was recovered with a root mean square error (RMSE) of 0.86-2.7 cm^-1 (average error = 3.8%). MB concentration was recovered with RMSE = 0.26-0.52 µM (average error = 15.0%), which did not vary with inclusion of MnTPPS (p=0.65). This system will be utilized to determine optical properties in human abscesses, in order to generate treatment plans for photodynamic therapy.

Spectral characterization of liquid hemoglobin phantoms with varying oxygenation states

SIGNIFICANCE

For optical methods to accurately assess hemoglobin oxygen saturation in vivo, an independently verifiable tissue-like standard is required for validation. For this purpose, we propose three hemoglobin preparations and evaluate methods to characterize them.

AIM

To spectrally characterize three different hemoglobin preparations using multiple spectroscopic methods and to compare their absorption spectra to commonly used reference spectra.

APPROACH

Absorption spectra of three hemoglobin preparations in solution were characterized using spectroscopic collimated transmission: whole blood, lysed blood, and ferrous-stabilized hemoglobin. Tissue-mimicking phantoms composed of Intralipid, and the hemoglobin solutions were characterized using spatial frequency-domain spectroscopy (SFDS) and enhanced perfusion and oxygen saturation (EPOS) techniques while using yeast to deplete oxygen.

RESULTS

All hemoglobin preparations exhibited similar absorption spectra when accounting for methemoglobin and scattering in their oxyhemoglobin and deoxyhemoglobin forms, respectively. However, systematic differences were observed in the fitting depending on the reference spectra used. For the tissue-mimicking phantoms, SFDS measurements at the surface of the phantom were affected by oxygen diffusion at the interface with air, associated with higher values than for the EPOS system.

CONCLUSIONS

We show the validity of different blood phantoms and what considerations need to be addressed in each case to utilize them equivalently.

Vasomotion analysis of speed resolved perfusion, oxygen saturation, red blood cell tissue fraction, and vessel diameter: Novel microvascular perspectives

BACKGROUND

Vasomotion is the spontaneous oscillation in vascular tone in the microcirculation and is believed to be a physiological mechanism facilitating the transport of blood gases and nutrients to and from tissues. So far, Laser Doppler flowmetry has constituted the gold standard for in vivo vasomotion analysis.

MATERIALS AND METHODS

We applied vasomotion analysis to speed-resolved perfusion, oxygen saturation, red blood cell tissue (RBC) tissue fraction, and average vessel diameter from five healthy individuals at rest measured by the newly developed Periflux 6000 EPOS system over 10 minutes. Magnitude scalogram and the time-averaged wavelet spectra were divided into frequency intervals reflecting endothelial, neurogenic, myogenic, respiratory, and cardiac function.

RESULTS

Recurrent high-intensity periods of the myogenic, neurogenic, and endothelial frequency intervals were found. The neurogenic activity was considerably more pronounced for the oxygen saturation, RBC tissue fraction, and vessel diameter signals, than for the perfusion signals. In a correlation analysis we found that changes in perfusion in the myogenic, neurogenic, and endothelial frequency intervals precede changes in the other signals. Furthermore, changes in average vessel diameter were in general negatively correlated to the other signals in the same frequency intervals, indicating the importance of capillary recruitment.

CONCLUSION

We conclude that vasomotion can be observed in signals reflecting speed resolved perfusion, oxygen saturation, RBC tissue fraction, and vessel diameter. The new parameters enable new aspects of the microcirculation to be observed.

Effect of curvature correction on parameters extracted from hyperspectral images

SIGNIFICANCE

Hyperspectral imaging (HSI) has emerged as a promising optical technique. Besides optical properties of a sample, other sample physical properties also affect the recorded images. They are significantly affected by the sample curvature and sample surface to camera distance. A correction method to reduce the artifacts is necessary to reliably extract sample properties.

AIM

Our aim is to correct hyperspectral images using the three-dimensional (3D) surface data and assess how the correction affects the extracted sample properties.

APPROACH

We propose the combination of HSI and 3D profilometry to correct the images using the Lambert cosine law. The feasibility of the correction method is presented first on hemispherical tissue phantoms and next on human hands before, during, and after the vascular occlusion test (VOT).

RESULTS

Seven different phantoms with known optical properties were created and imaged with a hyperspectral system. The correction method worked up to 60 deg inclination angle, whereas for uncorrected images the maximum angles were 20 deg. Imaging hands before, during, and after VOT shows good agreement between the expected and extracted skin physiological parameters.

CONCLUSIONS

The correction method was successfully applied on the images of tissue phantoms of known optical properties and geometry and VOT. The proposed method could be applied to any reflectance optical imaging technique and should be used whenever the sample parameters need to be extracted from a curved surface sample.

Laser Doppler Spectrum Analysis Based on Calculation of Cumulative Sums Detects Changes in Skin Capillary Blood Flow in Type 2 Diabetes Melitus

In this article, we introduce a new method of signal processing and data analysis for the digital laser Doppler flowmetry. Our approach is based on the calculation of cumulative sums over the registered Doppler power spectra. The introduced new parameter represents an integral estimation for the redistribution of moving red blood cells over the range of speed. The prototype of the device implementing the technique is developed and tested in preliminary clinical trials. The methodology was verified with the involvement of two age groups of healthy volunteers and in a group of patients with type 2 diabetes mellitus. The main practical result of the study is the development of a set of binary linear classifiers that allow the method to identify typical patterns of the microcirculation for the healthy volunteers and diabetic patients based on the presented diagnostic algorithm.