Physics inspired neural network for optical property retrieval from diffuse reflectance

In clinical diagnostic procedures, gingival inflammation is considered as the initial stage of periodontal breakdown. This is often detected clinically by bleeding on probing as it is an objective measure of inflammation. Since conventional diagnostic procedures have several inherent drawbacks, development of novel non-invasive diagnostic techniques assumes significance. This clinical study was carried out in 15 healthy volunteers and 25 patients to demonstrate the applicability of diffuse reflectance (DR) spectroscopy for quantification and discrimination of various stages of inflammatory conditions in periodontal disease. The DR spectra of diseased lesions recorded using a point monitoring system consisting of a tungsten halogen lamp and a fiber-optic spectrometer showed oxygenated hemoglobin absorption dips at 545 and 575 nm. Mean DR spectra on normalization shows marked differences between healthy and different stages of gingival inflammation. Among the various DR intensity ratios investigated, involving oxy Hb absorption peaks, the R620/R575 ratio was found to be a good parameter of gingival inflammation. In order to screen the entire diseased area and its surroundings instantaneously, DR images were recorded with an EMCCD camera at 620 and 575 nm. We have observed that using the DR image intensity ratio R620/R575 mild inflammatory tissues could be discriminated from healthy with a sensitivity of 92% and specificity of 93%, and from moderate with a sensitivity of 83% and specificity of 96%. The sensitivity and specificity obtained between moderate and severe inflammation are 82% and 76% respectively.

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.

This dissertation presents a series of studies in the development of contact and non-contact optical measurement techniques, and Monte Carlo (MC) based data analysis methods in visible diffuse reflectance and auto-fluorescence spectroscopy/imaging for tissue characterization. Their applications in early epithelial cancer diagnosis and tissue viability prediction are demonstrated. Firstly, the background in epithelial cancer and flap tissue viability as well as the principles/state of art of optical spectroscopy and imaging in tissue optics was presented in Chapter 1. Then a general survey on the capability of MC modeling of light transport in tissues was provided in Chapter 2 due to the importance of MC modeling in the field of tissue optics. The recent progress in the development of methods for speeding up MC simulations and the potential directions of future development were discussed. Based on the literature review on MC methods given in the previous Chapter, we developed a MC method to simulate diffuse reflectance and fluorescence from a layered tissue with embedded objects to mimic an early epithelial caner model in Chapter 3. With the help of this MC method, a series of numerical studies were performed to provide the guidelines for the selection of a proper epithelial cancer model in MC simulations. However, it is very time consuming to use the standard MC method for the simulation of light transport in a layered tissue with embedded object. To overcome this problem, a hybrid method, in which the scaling method and the perturbation MC method were integrated coherently, was proposed and validated to accelerate the simulations of diffuse reflectance from a layered tissue with embedded objects in Chapter 4. This method is suitable for simulating diffuse reflectance spectra or creating a MC database to extract optical properties of an early epithelial cancer model. All the above studies were designed for contact optical measurements using fiber-optics. However, inconsistent probe-sample contact could induce significant errors in diagnosis of early epithelial cancer. To address this problem, lens based setup was investigated for non-contact optical measurements in Chapter 5 and Chapter 6. We firstly developed a MC method to simulate diffuse reflectance and fluorescence measurements by convex lenses based non-contact setup. Then a series of numerical studies were performed to achieve depth sensitive diffuse reflectance and fluorescence measurements on the early epithelial cancer model. After that experimental studies were performed by a lens based optical system to validate the MC results and confirm the findings obtained in simulation study. We further extended the lens based non-contact spectroscopy system to an imaging setup with a larger field of view to perform depth sensitive color imaging on an early epithelial cancer phantom in Chapter 7. In the proposed setup, a micro-lens array was used to induce multi-focal illumination and a tunable lens was utilized to map multiple light foci into the tissue phantom at a range of depths. Another imaging lens was used to image light into a 3-CCD camera. The study performed on the epithelial cancer phantoms demonstrated that our method could be potentially used as a clinical tool for the diagnosis of early epithelial cancer. In addition to the numerical and phantom studies in epithelial cancer diagnosis, an animal study was also performed in Chapter 8 to predict tissue viability in flap surgery using a dual-modal system capable of performing both visible diffuse reflectance and auto-fluorescence spectroscopy. The results showed that either visible diffuse reflectance spectroscopy or auto-fluorescence spectroscopy alone can predict the skin viability accurately; however, auto-fluorescence spectroscopy was more sensitive to tissue changes in the first two hours after the induction of ischemia. It was feasible to predict flap failures in the first two hours when using auto-fluorescence spectroscopy alone. Moreover, it is possible to predict flap failures even in the first 15 minutes with high accuracy when using diffuse reflectance and auto-fluorescence spectroscopy simultaneously. In Chapter 9, we summarized that UV-visible diffuse reflectance and auto-fluorescence spectroscopy or imaging are promising tools for early epithelial cancer diagnosis and tissue viability prediction. Further refinement of these techniques would help advance the use of optical spectroscopy and imaging in clinical settings for tissue characterization in an even larger range of clinical applications.

Background: Liver parenchymal injuries such as steatosis, steatohepatitis, fibrosis, and sinusoidal obstruction syndrome can lead to increased morbidity and liver failure after liver resection. Diffuse reflectance spectroscopy (DRS) is an optical measuring method that is fast, convenient, and established. DRS has previously been used on the liver with an invasive technique consisting of a needle that is inserted into the parenchyma. We developed a DRS system with a hand-held probe that is applied to the liver surface. In this study, we investigated the impact of the liver capsule on DRS measurements and whether liver surface measurements are representative of the whole liver. We also wanted to confirm that we could discriminate between tumor and liver parenchyma by DRS. Materials and Methods: The instrumentation setup consisted of a light source, a fiber-optic contact probe, and two spectrometers connected to a computer. Patients scheduled for liver resection due to hepatic malignancy were included, and DRS measurements were performed on the excised liver part with and without the liver capsule and alongside a newly cut surface. To estimate the scattering parameters and tissue chromophore volume fractions, including blood, bile, and fat, the measured diffuse reflectance spectra were applied to an analytical model. Results: In total, 960 DRS spectra from the excised liver tissue of 18 patients were analyzed. All factors analyzed regarding tumor versus liver tissue were significantly different. When measuring through the capsule, the blood volume fraction was found to be 8.4 ± 3.5%, the lipid volume fraction was 9.9 ± 4.7%, and the bile volume fraction was 8.2 ± 4.6%. No differences could be found between surface measurements and cross-sectional measurements. In measurements with/without the liver capsule, the differences in volume fraction were 1.63% (0.75-2.77), -0.54% (-2.97 to 0.32), and -0.15% (-1.06 to 1.24) for blood, lipid, and bile, respectively. Conclusion: This study shows that it is possible to manage DRS measurements through the liver capsule and that surface DRS measurements are representative of the whole liver. The results are consistent with data published earlier on the combination of liver chromophores. The results encourage us to proceed with in vivo measurements for further quantification of the liver's composition and assessment of parenchymal damage such as steatosis and fibrosis grade.

Hyperspectral imaging, image analysis and diffusion theory were used to visualize skin vasculature and to monitor the development of fresh skin bruises. Bruises were inflicted in a porcine model, and the development of the hemorrhage was monitored using white light hyperspectral imaging (400-1000 nm). Hyperspectral images from human volunteers were also included in the study. Statistical image analysis was used to classify bruised regions and to visualize the skin vasculature. Biopsies were collected from the animals to reveal the true depth of the bruising. A three-layer diffusion model and an analytic hemoglobin transport model were used to model the reflectance spectra from the images. The results show that hyperspectral images contain depth information, and that the approximate depth and extent of bruises can be retrieved using a combination of statistical image analysis and diffusion theory. This technique also shows potential to visualize vascular structures in human skin.

Caffeine is the most widely consumed psychoactive substance in the world. It affects many tissues and organs, in particular central nervous system, heart, and blood vessels. The effect of caffeine on vascular smooth muscle cells is an initial transient contraction followed by significant vasodilatation. In this study we investigate the use of diffuse reflectance spectroscopy (DRS) for monitoring of vascular changes in human skin induced by caffeine consumption. DRS spectra were recorded on volar sides of the forearms of ten healthy volunteers at time delays of 0, 30, 60, 120, and 180 minutes after consumption of caffeine, while one subject served as a negative control. Analytical diffusion approximation solutions for diffuse reflectance from three-layer structures were used to assess skin composition (e.g., dermal blood volume fraction and oxygen saturation) by fitting to experimental data. The results demonstrate that cutaneous vasodynamics induced by caffeine consumption can be monitored by DRS, while changes in the control subject not consuming caffeine were insignificant.

.SignificanceHyperspectral reflectance imaging can be used in medicine to identify tissue types, such as tumor tissue. Tissue classification algorithms are developed based on, e.g., machine learning or principle component analysis. For the development of these algorithms, data are generally preprocessed to remove variability in data not related to the tissue itself since this will improve the performance of the classification algorithm. In hyperspectral imaging, the measured spectra are also influenced by reflections from the surface (glare) and height variations within and between tissue samples.AimTo compare the ability of different preprocessing algorithms to decrease variations in spectra induced by glare and height differences while maintaining contrast based on differences in optical properties between tissue types.ApproachWe compare eight preprocessing algorithms commonly used in medical hyperspectral imaging: standard normal variate, multiplicative scatter correction, min–max normalization, mean centering, area under the curve normalization, single wavelength normalization, first derivative, and second derivative. We investigate conservation of contrast stemming from differences in: blood volume fraction, presence of different absorbers, scatter amplitude, and scatter slope—while correcting for glare and height variations. We use a similarity metric, the overlap coefficient, to quantify contrast between spectra. We also investigate the algorithms for clinical datasets from the colon and breast.ConclusionsPreprocessing reduces the overlap due to glare and distance variations. In general, the algorithms standard normal variate, min–max, area under the curve, and single wavelength normalization are the most suitable to preprocess data used to develop a classification algorithm for tissue classification. The type of contrast between tissue types determines which of these four algorithms is most suitable.

Early accurate diagnosis and assessment severity of the hypoxic state of diabetic foot is of paramount importance. In this paper, a noninvasive monitoring method of the hypoxic state of the diabetic foot based on diffuse reflectance spectroscopy is proposed. Monte Carlo simulation method is used to imitate the effect on diffuse reflectance of the foot from different blood volume fractions and blood oxygen saturations. Moreover, the in vivo experiments of the hypoxic state of the foot based on the binding method is carried out using an integrating sphere system. Monte Carlo simulation results show that the diffuse reflectance of the foot under normal state and hypoxic state is significantly different. In vivo experimental results are highly consistent with the Monte Carlo simulation results. In vivo experiments further indicate that six characteristic wavelengths (440, 469, 514, 540, 560, and 576 nm) of the diffuse reflectance are more able to distinguish the normal oxygen state and hypoxic states of the foot. The proposed method promotes the development and application of the diffuse reflectance spectroscopy method in the noninvasive detection of biological tissue.

Accurate diagnosis of premalignant or malignant oral lesions depends on the quality of the biopsy, adequate clinical information and correct interpretation of the biopsy results. The major clinical challenge is to precisely locate the biopsy site in a clinically suspicious lesion. Dips due to oxygenated hemoglobin absorption have been noticed at 545 and 575 nm in the diffusely reflected white light spectra of oral mucosa and the intensity ratio R545/R575 has been found suited for early detection of oral pre-cancers. A multi-spectral diffuse reflectance (DR) imaging system has been developed consisting of an electron multiplying charge coupled device (EMCCD) camera and a liquid crystal tunable filter for guiding the clinician to an optimal biopsy site. Towards this DR images were recorded from 27 patients with potentially malignant lesions on their tongue (dorsal, lateral and ventral sides) and from 44 healthy controls at 545 and 575 nm with the DR imaging system. False colored ratio image R545/R575 of the lesion provides a visual discerning capability that helps in locating the most malignant site for biopsy. Histopathological report of guided biopsy showed that out of the 27 patients 16 were cancers, 9 pre-cancers and 2 lichen planus. In this clinical trial DR imaging has correctly guided 25 biopsy sites, yielding a sensitivity of 93% and a specificity of 98%, thereby establishing the potential of DR imaging as a tool for guided biopsy.

.Significance: Physiological parameters extracted from diffuse reflectance spectroscopy (DRS) provide clinicians quantitative information about tissue that helps aid in diagnosis. There is a great need for an accurate and cost-effective method for extracting parameters from DRS measurements.Aim: The aim is to explore the accuracy and speed of physiological parameter extraction using machine learning models compared to that of the widely used Monte Carlo lookup table (MCLUT) inverse model.Approach: Diffuse reflectance spectra were simulated using a light transport model based on Monte Carlo simulations and weighted to six wavelengths. Deep learning (DL), random forest (RF), gradient boosting machine (GBM), and generalized linear model (GLM) machine learning models were built using a training set of 10,000 spectra from the simulated data. The MCLUT and machine learning models were used to predict physiological parameters from a separate test set of 30,000 simulated spectra. Mean absolute errors were calculated to evaluate the accuracy and compare it among MCLUT and machine learning models. In addition, the computational time to predict parameters from the test set was recorded to compare the speed among MCLUT and machine learning models.Results: The DL, RF, GBM, and GLM models all had significantly lower errors than the MCLUT inverse method for six wavelengths. The DL model proved to have the lowest errors, with all absolute percent errors under 10%. The DL model had much faster runtimes than the MCLUT.Conclusions: Machine learning is promising for extracting physiological parameters from six-wavelength DRS data, with both lower errors and a faster runtime than the widely used MCLUT model.

Caffeine is the most widely consumed psychoactive substance worldwide, affecting numerous tissues and organs, with notable impacts on the central nervous system, heart, and blood vessels. The effect of caffeine on vascular smooth muscle cells is an initial transient contraction followed by significant vasodilatation. In this study we investigate the use of diffuse reflectance spectroscopy (DRS) for monitoring of vascular changes in human skin induced by caffeine consumption. DRS spectra were recorded on volar sides of the forearms of eight healthy volunteers at time intervals of 0, 30, 60, 120, and 180 min after consumption of caffeine, while one subject served as a negative control. Analytical diffusion approximation solutions for diffuse reflectance from three-layer structures were used to assess skin composition (e.g. dermal blood volume fraction and oxygen saturation) by fitting these solutions to experimental data. The results demonstrate that cutaneous vasodynamics induced by caffeine consumption can be monitored by DRS, while changes in the control subject not consuming caffeine were insignificant.

Flap transfer has become a common technique in reconstructive surgery. However, a significant number of compromised skin flaps are not successfully salvaged because the current clinical method for flap assessment relies heavily on the clinician's experience. Vascular occlusion is the major reason for flap failure, thus the accurate and objective early prediction of blood vessel occlusion is vitally important. Our parallel point measurement study has demonstrated the great potential of joint diffuse reflectance and autofluorescence spectroscopy in the early detection and differentiation of venous and arterial occlusion in skin flaps. Unfortunately, the technique of point measurements is not suitable to examine a large skin flap when a high spatial resolution is required. In this study, we attempted to overcome this problem by performing spectral diffuse reflectance and autofluorescence imaging on a rat skin flap model. Both imaging data and reconstructed spectra were used to statistically differentiate control flaps, arterially occluded flaps and venously occluded flaps. Our preliminary results suggest that the technique of joint diffuse reflectance and autofluorescence spectroscopic imaging can achieve high classification accuracy thus could be used to detect and differentiate flaps with venous and arterial occlusion accurately at an early time point in a large skin flap. Typical reconstructed spectra of (a) diffuse reflectance and (b) autofluorescence after normalization.

Aim Gastrointestinal (GI) cancers account for 26% of global cancer incidence with prevalence projected to rise exponentially due to the ageing population and lifestyle choices. Surgical resection is the mainstay of treatment to remove the cancer in its entirety to achieve an R0 resection. Positive margins, when cancerous tissue has been left in situ, is associated with increased morbidity and mortality. Current margin assessment involves histopathological analysis, after resection of the specimen. Diffuse Reflectance Spectroscopy (DRS) and Hyperspectral Imaging (HSI) are novel imaging techniques that have the potential to provide real-time assessment of cancer margins intra-operatively to reduce the incidence of positive resection margins and improve patient outcomes. The aim of this review is to assess the current state of evidence for the use of novel imaging techniques in GI cancer margin assessment. Method A literature review was conducted of studies using DRS and HSI in GI cancers in adult patients, published from inception to October 2020. Results A total of 15 studies were analysed, nine of which used DRS and six used HSI and the majority of studies were performed ex-vivo. Current image acquisition techniques and processing algorithms vary greatly. The sensitivity and specificity of DRS ranged from 0.90-0.98 and 0.88-0.95 respectively and for HSI 0.63-0.98 and 0.69-0.98, respectively across five types of GI cancers. Conclusions DRS and HSI are novel imaging techniques, currently in their infancy but the outlook is promising. With further research focused on standardising methodology and in-vivo settings, DRS and HSI could transform intra-operative margin assessment in GI cancers.

A hyperspectral image in the near infrared contains thousands of position-referenced spectra. After imaging reference materials of known composition it is possible to build Partial Least Squares (PLS) regression models for predicting unknown compositions from new images or spectra. In this paper a comparison is made between spectra from a hyperspectral image and spectra from two spectrometers: a scanning grating instrument with rotating sample holders and an FT-NIR instrument utilizing a fiber-optic probe. The raw spectra and the quality of the PLS calibration models and predictions are compared. Two sample datasets consist of a set of 13 designed artificial mixtures of pure constituents and a selection of 13 sampled cheeses. The prediction error from the hyperspectral image spectra is between that of the two spectrometers. For a typical food sample, the average bias [and replicate standard deviation] was -0.6% [0.5%] for protein and -0.2% [1.3%] for fat. Comparable values for the best spectrometer were -0.2% bias for protein and -0.5% for fat. Some of the advantages of working with hyperspectral images are highlighted: the simultaneous exploration of representations of both spectral and spatial data, and the analysis of concentration profiles and concentration maps all contribute to better characterization of organic and biological materials.

Measurement of the optical properties of fruits and vegetables using spatially resolved hyperspectral diffuse reflectance imaging technique