Photoacoustic and absorption spectroscopy imaging analysis of human blood

Wei-Yun Tsai; Stephan Breimann; Tsu-Wang Shen; Dmitrij Frishman

doi:10.1371/journal.pone.0289704

Abstract

Photoacoustic and absorption spectroscopy imaging are safe and non-invasive molecular quantification techniques, which do not utilize ionizing radiation and allow for repeated probing of samples without them being contaminated or damaged. Here we assessed the potential of these techniques for measuring biochemical parameters. We investigated the statistical association between 31 time and frequency domain features derived from photoacoustic and absorption spectroscopy signals and 19 biochemical blood parameters. We found that photoacoustic and absorption spectroscopy imaging features are significantly correlated with 14 and 17 individual biochemical parameters, respectively. Moreover, some of the biochemical blood parameters can be accurately predicted based on photoacoustic and absorption spectroscopy imaging features by polynomial regression. In particular, the levels of uric acid and albumin can be accurately explained by a combination of photoacoustic and absorption spectroscopy imaging features (adjusted R-squared > 0.75), while creatinine levels can be accurately explained by the features of the photoacoustic system (adjusted R-squared > 0.80). We identified a number of imaging features that inform on the biochemical blood parameters and can be potentially useful in clinical diagnosis. We also demonstrated that linear and non-linear combinations of photoacoustic and absorption spectroscopy imaging features can accurately predict some of the biochemical blood parameters. These results demonstrate that photoacoustic and absorption spectroscopy imaging systems show promise for future applications in clinical practice.

Citation: Tsai W-Y, Breimann S, Shen T-W, Frishman D (2023) Photoacoustic and absorption spectroscopy imaging analysis of human blood. PLoS ONE 18(8): e0289704. https://doi.org/10.1371/journal.pone.0289704

Editor: Giuseppe Chirico, Universita degli Studi di Milano-Bicocca, ITALY

Received: December 15, 2022; Accepted: July 25, 2023; Published: August 4, 2023

Copyright: © 2023 Tsai et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All human data, statistical analysis code, polynomial regression models are available on Zenodo at link http://doi.org/10.5281/zenodo.5893390. Powered.

Funding: This research was supported by Tzu Chi Foundation under the grants TCMMP104-01-03-01.There was no additional external funding received for this study The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Photoacoustic and absorption spectroscopy imaging are emerging non-invasive imaging modalities with a broad spectrum of potential biomedical applications. In photoacoustic imaging, a tissue sample is irradiated with a laser pulse. Depending on the optical absorption properties of the chromophore, the absorbed photon energy is converted to heat and generates an acoustic wave, which is detected by an ultrasonic transducer [1]. In absorption spectroscopy imaging, biomolecules absorb light pulses emitted by light-emitting diodes (LED). The scattered light is detected by a balanced detector and the signal is captured and processed by a digital oscilloscope [2].

Both photoacoustic and absorption spectroscopy imaging are easy-to-use, low-cost, non-ionizing and non-radioactive methods capable of detecting absorption spectroscopy contrast in a variety of biological tissues [3, 4]. Here, we apply photoacoustic/ absorption spectroscopy imaging systems to detect biochemical parameters from blood samples in vitro. Blood reflects the status of internal organs and tissues throughout the body and is therefore important in detecting, monitoring, and treating many diseases. Photoacoustic signals have been applied in many blood-related studies [5], such as the observation of size and shape of red blood cells [6–9], measurement of blood glucose concentration [10, 11], early malaria diagnosis [12, 13], diagnosis of bacteria in the blood [14], detection of circulating tumor cells [15, 16], investigation of hemoglobin oxygen and carboxyhemoglobin saturation [17, 18], bilirubin monitoring [19], and pharmaceuticals monitoring [20–22]. In particular, light scattering by blood cells can be effectively used to assess the general state of health and catch the early signs of multiple diseases [23]. Changes of glucose concentration can be used to diagnose diabetes [24], urea, creatinine, eGFR, and uric acid levels are related to renal function [25], total cholesterol, triglycerides, as well as high-density and low-density lipoproteins have relevance to cardiovascular disease [26, 27]. Alanine aminotransferase and albumin are indicators of liver and bile duct function [28], while Na, K, and Cl levels are sensitive to hypertension [29, 30]. Those compounds may provide different spectrums of ultrasound echoes in photoacoustic measurements.

In this study, we analyzed photoacoustic and absorption spectroscopy imaging signals with both the time domain and frequency domain. The time domain graph displays a signal over time, whereas the frequency domain graph displays how much signal is in each specific frequency band over a range of frequencies. The time domain graph displays the changes of the signal over time, whereas the frequency domain graph reflects the distribution of the signal over frequency bands. The time domain graph can thus be directly used to observe the features of the waveform without transformation, while the frequency domain graph needs to be transformed. The frequency domain analysis reveals some additional features, such as PASA slope, midband fit, and intercept. Many research groups attempted to quantify the histologic information by specific features of photoacoustic and absorption spectroscopy imaging signals, both in the time domain (amplitude [10], rising and descending trend [31], area [10]) and in the frequency domain (slope, Midband fit, intercept [3], full width at half maximum (FWHM) [32], prominence, and area [10]). However, a comprehensive feature analysis of photoacoustic and absorption spectroscopy imaging signals applied to biochemical parameters has been lacking so far. In the present study, we investigated the connection between 31 imaging features from the time and frequency domains and a broad spectrum of biochemical parameters of human blood. We have also constructed polynomial regression models to predict biochemical parameters based on imaging signal features.

Materials and methods

Blood samples

Sample collection and analysis followed the protocol approved by the Tzu Chi university. This study was performed according to the regulation of the Institution Review Board of the Tzu Chi Hospital (Taiwan), ID number IRB105-137-B. All participants received a written informed consent containing a detailed explanation of the human experiments and signed the informed consent before the experiments. All experiments were performed in accordance with relevant guidelines and regulations. Photoacoustic imaging, absorption spectroscopy imaging, and biochemical measurements were performed at the Tzu Chi Hospital on blood drawn from 14 individuals (6 male and 8 female) aged 30–71. There were 12 individuals for which both photoacoustic and absorption spectroscopy imaging measurements were performed, 1 individual with only a photoacoustic measurement and 1 with only an absorption spectroscopy imaging measurement.

Biochemical blood parameters

For each patient, the total of 19 biochemical parameters were measured: ALT (alanine aminotransferase), TP (total protein), ALB.BCG (albumin), GLO (globulin), BUN (blood urea nitrogen), UA (uric acid), CRE (creatinine), TCH (total cholesterol), TG (triglyceride), GLU.AC (glucose), eGFR (estimated glomerular filtration rate), Na, K, Cl, Ca, Fe, TIBC (total iron binding capacity), HDL.C (high-density lipoprotein), LDL.C (low-density lipoprotein).

Photoacoustic imaging by laser pulses

The experimental setup for photoacoustic imaging consists of the following parts (Fig 1):

Class IIIB laser (LECC, Taiwan) with the wavelength of 658 nm, which is within the traditional optical window used for biological tissues [33, 34], output power of 80 mW, pulse width of 0.5 ms, energy of 32.5 mJ, sampling duration of 0.002 ms, and pulse repetition rate of 1k Hz. The total time of the laser activation is 0.5 ms.
The power supply (MOTECH LPS-305) as a tool to control the laser.
An ultrasound detector (piezoelectric sensor) for detecting laser-induced photoacoustic signals and converting them into electrical signals. We used the ultrasound transducer (Unictron Technologies Crops, Taiwan) with the resonant frequency 3.20MHz ± 3‰, resonant impedance less than 0.65 Ω, and electromechanical coupling coefficient larger than 49%. The piezoelectric ceramic piece has a length of 21.9 mm and a width of 15.9 mm with four strip holes of 12 mm in length and 2 mm in width, as shown in Fig 1B.
A low noise amplifier (LNA), which can amplify very low-power signals without significantly degrading their signal-to-noise ratio, was connected to the ultrasound transducer. The gain of the LNA is 271 times and the input referred noise of the operational amplifier (Texas instruments Incorporated OPA847, USA) is 1 nV/√Hz.
A digital oscilloscope (TDS 2024C, Tektronix) for recording PA signals from blood samples with the resampling rate of 500000.
A computer for extracting and processing features from the averaged PA signal.
Plastic containers (Sorenson BioScience Inc., 0.2 mL PCR tubes with attached caps, Salt Lake City, UT, USA) for holding samples on an ultrasonic detector, sealed with glue in order to minimize instability from external factors. The laser was focused on the tip position of the plastic containers but not on the ultrasound detector to avoid affecting the photoacoustic signal.

Download:

Fig 1. Schematic diagram of the photoacoustic imaging system and the placement of class IIIB laser, ultrasound detector and blood sample.

(A) The blood sample is irradiated with a laser pulse and the absorbed photon energy is converted to heat and generates an acoustic wave, which is detected by an ultrasonic detector. The signal is amplified by an amplifier and recorded on a digital oscilloscope. Finally, the computer is used to process the features and extract the features from the signal. (B) The structure of the piezoelectric ceramic piece. (C) The placement of the blood sample, the class IIIB laser and the ultrasonic detector. (D) The amplifier circuit diagram. (This figure has been updated).

https://doi.org/10.1371/journal.pone.0289704.g001

The voltage of the photoacoustic signals generated by laser pulses was in the approximate range between -1 V and 0.5 V.

The beginning voltage of the signals was negative. In order to make measurements more easily comparable between samples we shifted the photoacoustic signal values such that they start from 0 and lie in the approximate range between 0–1.5 a.u.

Absorption spectroscopy imaging by light emitting diodes

The experimental setup for absorption spectroscopy imaging consists of the following parts (Fig 2):

Light emitting diodes (Kolin, Taiwan) generating 0.45 ms pulse width at 650±10nm, coupled into the setup using an optical fiber.
The Thorlabs SKSAS (Saturated Absorption Spectroscopy Systems) kit, which includes a fiber collimator, prisms, mirrors, and a balanced detector. A fiber collimator (F220FC-780, Thorlabs) collimates the light, which is then passed through a half wave plate and a polarizing beam splitter cube in order to split off power for the “pump” beam. The remaining light is split into two parallel beams, the “probe” beam, and the “reference” beam, which pass through the blood sample and water, respectively. The experiment must ensure that the path lengths (blue and red arrows) of these two beams are the same. The two photodiodes of the balanced detector (PDB210A) with a power supply (LDS1212, Thorlabs) of 115 mA detect the probe and reference beams.
A digital oscilloscope (TDS 2024C, Tektronix) attached to the radio frequency OUTPUT connector of the balanced detector to record signals from samples.
A computer for processing and extracting features from the averaged PA signal.
Plastic containers (Sorenson BioScience Inc., 0.2 mL PCR Tubes with Attached Caps, Salt Lake City, UT, USA) for holding samples and water. The light with a beam diameter 1.5 mm was focused on the tip position of the plastic containers.

Download:

Fig 2. Schematic diagram of the absorption spectroscopy imaging system with light emitting diodes, the path of the light from the light emitting diode, and the experimental setup for the absorption spectroscopy imaging system.

(A) The blood sample absorbs light pulses emitted by light–emitting diodes (LED). The scattering of light is detected by a balanced detector. Finally, the signals are captured and processed by digital oscilloscope and the computer is used to process the features and extract the features from the signals. (B) The placement of the components in the optical imaging system. (This figure has been updated).

https://doi.org/10.1371/journal.pone.0289704.g002

The voltage of the absorption spectroscopy signals generated by light pulses was approximately in the range between -100 V to 20 V. The beginning voltage of the signals was negative. In order to make measurements more easily comparable between samples we shifted the range of absorption spectroscopy signal values to start from 0. Therefore, the intensity of the absorption spectroscopy signals generated by light pulses was approximately in the range between 0 a.u. to 120 a.u.

Photoacoustic and absorption spectroscopy signal features

In order to make measurements more easily comparable between samples we shifted the range of both photoacoustic (Fig 3A) and absorption spectroscopy (Fig 3B) signal values to start from 0 a.u. In all experiments 2500 samples of the signal were collected over 5 milliseconds with a sampling interval of 0.002 milliseconds. For each individual, photoacoustic and absorption spectroscopy experiments were repeated 10 and 5 times, respectively (Figs 4A and 5A). At each time point the signal intensity was averaged over the number of experiments to reduce the noise (Figs 4B and 5B). The trigger has an error in the switching time of each experiment, and the error time is about 0.5 ms. To reduce analysis error, we only use signals in the steady state for further analysis. Subsequently, the total of 31 time and frequency domain features were extracted from the signal, summarized in Table 1.

Download:

Fig 3.

Example of signals from photoacoustic (A) and absorption spectroscopy (B) experiments. In all experiments 2500 samples of the signal were collected over 5 milliseconds with a sampling interval of 0.002 milliseconds.

https://doi.org/10.1371/journal.pone.0289704.g003

Download:

Fig 4. Pre–and post–averaging of PA signal traces from laser samples.

(A) Each experiment includes 2500 signal samples in 5 milliseconds with a sampling interval of 0.002 milliseconds. The photoacoustic experiment was repeated 10 times for each individual. Different colors represent 10 repetitions of photoacoustic experiments. (B) To reduce the noise, the signal intensity at each time point was averaged.

https://doi.org/10.1371/journal.pone.0289704.g004

Download:

Fig 5. Pre–and post–averaging absorption spectroscopy signal traces from light samples.

(A) Each experiments includes 2500 signal samples in 5 milliseconds with a sampling interval of 0.002 milliseconds. The absorption spectroscopy experiment was repeated 5 times for each individual. Different colors represent 5 repetitions of absorption spectroscopy experiments. (B) To reduce the noise, we averaged the signal intensity at each time point.

https://doi.org/10.1371/journal.pone.0289704.g005

Download:

Table 1. The list of imaging features.

https://doi.org/10.1371/journal.pone.0289704.t001

Time domain features

Following previous research [4, 23], we calculated 3 types of amplitude features: peak-to-peak amplitude and amplitude of the positive and negative peak. The first step in calculating the average amplitude was to find the top 10 abrupt changes in the signal using the findchangepts [35] function in MATLAB (Fig 6A and 6B). Subsequently we defined the windows on the left (part A) and right (part B) sides of the window containing the largest signal intensity and averaged the intensity over these two windows. Peak-to-peak amplitude was calculated as the range between the maximum and minimum intensity, while the amplitude of the positive and negative peaks is the difference between the maximum and the average intensity as well as between the average and minimum intensity, respectively (Fig 6C and 6D).

Download:

Fig 6. Definition of two windows and the amplitude–related features in the photoacoustic and absorption spectroscopy signal.

(A) and (B) are signals generated by photoacoustic and absorption spectroscopy systems, respectively. The green lines indicate the top 10 abrupt changes determined by the findchangepts function in MATLAB. Part A and part B are located on the left and the right sides of the window containing the largest signal intensity. (C) The definition of peak–to–peak amplitude and amplitude of the positive and negative peak were according to the average intensity (red line). (D) The definition of peak–to–peak amplitude and amplitude of the positive and negative peak were according to the average intensity (red line).

https://doi.org/10.1371/journal.pone.0289704.g006

Positive and negative slopes

These measures have been previously applied for photoacoustic detection of protein coagulation [31] and for analyzing the hemodynamic response by functional near-infrared spectroscopy [36]. We conducted a search for slopes in the region bounded by the end of part A and the beginning of part B (see above and Fig 7). The region to calculate the positive slope is from the end of part A to the maximum of the signal intensity. The region to calculate the negative slope is from the maximum of the signal intensity to the beginning of part B. The relationship between the intensity of the optical absorption spectra y and the time x was modeled in form of a linear regression y = bx+a, where n is the signal sample size and the intercept a and the slope b were calculated as (1) (2)

Download:

Fig 7. Illustration of the positive and negative slopes.

(A) and (B) are signals generated by photoacoustic and absorption spectroscopy systems, respectively. The green lines indicate the end of part A and the beginning of part B, respectively. The red and yellow lines indicate the positive and negative slopes, respectively.

https://doi.org/10.1371/journal.pone.0289704.g007

Frequency domain features

Frequency domain features are suitable for evaluating the morphology of optically absorbing objects [37] and investigating the microscopic architecture of biological tissues [38]. The time to frequency domain conversion of the signal was affected by the Welch’s method (as implemented by the pwelch function in MATLAB) [39, 40]. Power spectra were estimated by dividing the time signal into successive Hamming windows of 1/8 of the total signal length overlapping by 50%, forming the periodogram for each block, and then averaging the periodograms over time.

PASA slope, midband fit and intercept

Photoacoustic spectrum analysis (PASA) is a non-invasive technique to investigate the microstructures in biological tissues [41]. It has been applied to assessing the Gleason scores of prostate cancers [38], quantifying the lipid content of liver [42], and differentiating between tumor types [43]. Power spectra were fit by a linear regression model using the fitlm function in MATLAB and then the slope, intercept and midband fit values were determined (Figs 8 and 9), which reflect the magnitude of high vs low frequency components, the zero frequency, and the average magnitude of the power spectrum, respectively [41]. These three parameters have been shown to correlate with the microstructural properties of tissues [3, 44]. In order to differentiate between tissues more precisely, we divided the signal into 4 frequency bands [45] (0–1 MHz,1–2 MHz, 2–3 MHz, and 0–3 MHz) (Fig 9).

Download:

Fig 8. Frequency domain analysis.

The PASA slope (red lines), the values of intercept (orange circles) and midband fit (green circles) were showed in photoacoustic (A) and absorption spectroscopy (B) signals, respectively.

https://doi.org/10.1371/journal.pone.0289704.g008

Download:

Fig 9. Different frequency bands used for frequency domain analysis.

(A) and (B) are signals generated by photoacoustic and absorption spectroscopy systems, respectively. In order to differentiate between tissues more precisely, we divided the signal into 4 frequency bands (0–1 MHz,1–2 MHz, 2–3 MHz, and 0–3 MHz). The red lines indicate the PASA slopes.

https://doi.org/10.1371/journal.pone.0289704.g009

Prominence and FWHM

Prominence is the height of a peak relative to the lowest contour line around that peak. For peaks with significant prominence, we also computed full width at half maximum (FWHM), the distance between the points where the magnitude is half of the maximal value. FWHM is a parameter describing pulse waveforms which was previously used to characterize blood viscosity in frequency-resolved photoacoustic measurements [32]. FWHM and prominence were calculated using the findpeaks function in MATLAB. We utilized prominence = 10 as a threshold, then only selected peaks with the maximum prominence in each frequency bands (0–0.5 MHz, 0.5–1 MHz, 1–1.5 MHz, 1.5–2 MHz, 2–2.5 MHz, and 2.5–3 MHz) (Fig 10).

Download:

Fig 10. The definition of Prominence and FWHM in frequency domain analysis from photoacoustic signals and absorption spectroscopy signals.

(A) Peaks of the photoacoustic signals with a prominence larger than 10 are identified. (B) A prominent peak of the photoacoustic signals in the 2–2.5 MHz band. (C) Peaks from absorption spectroscopy signals with a prominence larger than 10 are identified. (D) A prominent peak of the absorption spectroscopy signals in the 1–1.5 MHz band.

https://doi.org/10.1371/journal.pone.0289704.g010

Area under the photoacoustic signal waveform and frequency spectrum.

Area-based features have been shown to correlated with glucose concentration in whole blood samples [10]. We calculated the area under the signals in the time and frequency domain by the trapz function in MATLAB.

Correlation analysis

We computed Spearman’s correlations between biochemical and imaging features and assessed their significance by the R function rcorr [46]. Pairs of features with the absolute correlation coefficient > 0.5 and p-values ≤ 0.05 were considered to be significantly associated.

Model building

We investigated the performance of photoacoustic/absorption spectroscopy imaging features in predicting biochemical blood parameters based on polynomial regression, as implemented in the lm function of the R project. Only significantly associated features were considered, based on Spearman’s correlation. To test for multicollinearity between explanatory variables [47, 48] we entered biomedical parameters and imaging features into a polynomial regression model with degree 1. Multicollinearity was diagnosed based on the variance inflation factor (VIF) calculated using the vif function from the “car” R package [49]. Following the approach described in [50], the feature with the maximal VIF value (VIF>10) was removed and the diagnostic process was continued until only features with VIF ≤ 10 remained.

Next, we determined whether imaging features had a linear or a non-linear relationship with the blood parameters. Validation of the linear model assumptions [51] was performed using the gvlma function from the “gvlma” R package. This function performs a single global test to assess the linear model assumption as well as four specific directional tests designed to detect skewness, kurtosis, a nonlinear link function, and heteroscedasticity. For the global test the null hypothesis is that the relationship between two features is linear. The null hypothesis for the skewness and kurtosis tests is that the residuals are normally distributed. The null hypothesis of the link function is that the variable is continuous. Finally, for heteroscedasticity the null hypothesis is that the variance of the model residuals remains constant. We also utilized the Durbin-Watson test, as implemented in the “car” R package [49], to determine the independence of error terms, the null hypothesis being that the residuals are independent. If all the above null hypotheses were not rejected, an imaging feature was defined as linearly related with a biochemical parameter, otherwise a non-linear relationship was assumed.

Subsequently, linear relationships were fitted using polynomial regression models with a degree 1, while for non-linear relationships degrees 2 or 3 were chosen based on the Bayesian information criterion (BIC) [52], calculated as (3) where n is the sample size, k is the number of parameters to be estimated, and is the maximize value of the likelihood function of the model. The model with the lowest BIC value was selected.

The quality of the fit was assessed based on the adjusted R-squared and the p-value of the F-statistic. Adjusted R-squared values were computed using the following equation: (4) with the sample R-squared given by: (5) where n is the number of data points, y is the actual value, y^ is the predicted value, y_mean is the mean of y and p is the total number of explanatory variables in the model.

Results

Overview of the study

The goal of the study was to investigate to which extent photoacoustic and absorption spectroscopy imaging features are correlated with and predictive of biochemical parameters (Fig 11). Significantly correlated features were tested for multicollinearity. Subsequently it was determined whether imaging features were related to the biochemical parameters linearly or non-linearly. In the former case, polynomial models with degree 1 were used, while in the latter case the choice between polynomial models with degree 2 or 3 was guided by the Bayesian information criterion. The quality of the fit was assessed based on adjusted R². See Materials and Methods for a detailed explanation of experimental procedures, imaging and biochemical parameters, and statistical models.

Download:

Fig 11. Workflow of the study.

The study was based on different conditions to construct the polynomial models, which include Spearman’s correlation analysis, multicollinearity, linear regression diagnosis, determination of linearly or non–linearly, degree selection. Polynomial models were constructed with these conditions and the adjusted R–square of the models was calculated.

https://doi.org/10.1371/journal.pone.0289704.g011

Feature selection and correlation analysis

We found that 17 out of 31 photoacoustic imaging features were significantly correlated with at least one biochemical blood parameter (Fig 12A). Time domain features (peak-to-peak amplitude, amplitude of the positive peak, and amplitude of the negative peak, positive slope and the area under the signal waveform of time domain) tend to exhibit positive correlation while frequency domain features (PASA slope, Midband fit, Intercept, FWHM, and Prominence) are more frequently negatively correlated. The results for the absorption spectroscopy imaging features (Fig 12B) were qualitatively similar, with two time domain features (Peak-to-Peak Amplitude and Negative slope) and thirteen frequency domain features (PASA slope [2–3 MHz], Midband fit [1–2 MHz], Midband fit [2–3 MHz], Midband fit [0–3 MHz], Intercept [2–3 MHz], Intercept [0–3 MHz], FWHM [0–0.5 MHz], FWHM [1–1.5 MHz], FWHM [2.5–3 MHz], Prominence [0–0.5 MHz], Prominence [1.5–2 MHz], Prominence [2.5–3 MHz] and Frequency domain area) being mostly positively and negatively correlated with the blood parameters, respectively.

Download:

Fig 12. Heatmap representation of statistically significant (p–value≤0.05, r_s > 0.5) associations between imaging features (columns) and biochemical blood parameters (rows).

(A) Photoacoustic signals, (B) Absorption spectroscopy signals.

https://doi.org/10.1371/journal.pone.0289704.g012

Blood urea nitrogen (BUN) was the only biochemical parameter that did not show any correlation to imaging features in both photoacoustic and absorption spectroscopy systems. It is interesting to note that the photoacoustic and absorption spectroscopy imaging detection methods are complementary: estimated glomerular filtration rate (eGFR), glucose (GLU.AC), triglyceride (TG) and alanine aminotransferase (ALT) were not significantly associated with photoacoustic imaging features, while Fe was not significantly associated with absorption spectroscopy imaging features. Upon ranking imaging features according to their correlation with the biochemical parameters we found that in the photoacoustic system albumin (ALB.BCG) and time domain area had highest Spearman’s correlation coefficient (r_s = 0.909) (Fig 13), while uric acid (UA) and FWHM [2.5–3 MHz] had the highest Spearman’s correlation coefficient (r_s = 0.888) in absorption spectroscopy system (Fig 14).

Download:

Fig 13. The correlation ranking between biochemical parameters and imaging features in photoacoustic system.

Only significantly correlated features are shown (p–value≤0.05, r_s > 0.5). ALB.BCG (albumin), Ca, Cl, CRE (creatinine), Fe, GLO (globulin), HDL.C (high–density lipoprotein), K, LDL.C (low–density lipoprotein), Na, TCH (total cholesterol), TIBC (total iron binding capacity), TP (total protein), UA (uric acid).

https://doi.org/10.1371/journal.pone.0289704.g013

Download:

Fig 14. The correlation ranking between biochemical parameters and imaging features in absorption spectroscopy system.

Only significantly correlated features are shown (p–value≤0.05, r_s > 0.5). ALB.BCG (albumin), ALT (alanine aminotransferase), Ca, Cl, CRE (creatinine), eGFR (estimated glomerular filtration rate), GLO (globulin), GLU.AC (glucose), HDL.C (high–density lipoprotein), K, LDL.C (low–density lipoprotein), Na, TCH (total cholesterol), TG (triglyceride), TIBC (total iron binding capacity), TP (total protein), UA (uric acid).

https://doi.org/10.1371/journal.pone.0289704.g014

Model building

The total of 14 individual biochemical parameters significantly correlated to photoacoustic imaging features (ALB.BCG, Ca, Cl, CRE, Fe, GLO, HDL.C, K, LDL.C, Na, TCH, TIBC, TP and UA, see Fig 13) as well as 17 individual biochemical parameters significantly correlated to absorption spectroscopy imaging features (ALB.BCG, ALT, Ca, Cl, CRE, eGFR, GLO, GLU.AC, HDL.C, K, LDL.C, Na, TCH, TG, TIBC, TP and UA, see Fig 14) were used to construct statistical models. Multicollinear features were eliminated using the step-wise approach described in Methods. If the variance inflation factor (VIF) of some features was above 10, the feature with the maximum VIF value was removed and the procedure repeated until all VIF values were smaller or equal 10 (S1 and S2 Tables). Multicollinearity analysis was not performed when there was no significant correlation between imaging features and biochemical parameters or when a prediction model was based on only one imaging feature. In the S1 and S2 Tables, “NA” indicates that the multicollinearity analysis was not conducted for the model. For the remaining imaging features, we determined whether they had a linear or a non-linear relationship with the blood parameters based on the linear regression diagnosis (S3 and S4 Tables). Both for linear and non-linear relationships, the optimal degree was identified based on the Bayesian information criterion (BIC). These parameters were used to construct the biomedical prediction models by polynomial regression.

For the photoacoustic imaging features, we identified seven polynomial regression models with an adjusted R-squared values higher than 0.5 and significant p-values of F-statistic (<0.05) involving the following biochemical parameters: LDL.C, Cl, Na, TCH, CRE, UA and ALB.BCG (Fig 15 and S3 Table). Interestingly, five of parameters (LDL.C, TCH, CRE, UA and ALB.BCG) were significantly correlated with the amplitude-related features (Peak-to-Peak Amplitude, Amplitude of the positive peak and Amplitude of the negative peak). CRE, UA and ALB.BCG appear to be especially well explainable by polynomial regression models as they have very high adjusted R-squared values (0.80, 0.81 and 0.97, respectively). For the absorption spectroscopy imaging features, we identified six significant polynomial regression models with the adjusted R-squared values >0.5 and p-values of F-statistic <0.05 involving the following biochemical parameters: HDL.C, Ca, Cl, UA, ALB.BCG and TP (Fig 16 and S4 Table), whereby the adjusted R-squared values of UA and ALB.BCG were higher than 0.7.

Download:

Fig 15. Polynomial regression models for the photoacoustic imaging system with better prediction.

Only better prediction models are shown (adjusted R–squared values>0.5, p–values of F–statistic < 0.05). LDL.C (low–density lipoprotein), Cl, Na, TCH (total cholesterol), CRE (creatinine), UA (uric acid), ALB.BCG (albumin).

https://doi.org/10.1371/journal.pone.0289704.g015

Download:

Fig 16. Polynomial regression models for the absorption spectroscopy imaging system with better prediction.

Only better prediction models are shown (adjusted R–squared values>0.5, p–values of F–statistic < 0.05). HDL.C (high–density lipoprotein), Ca, Cl, UA (uric acid), ALB.BCG (albumin), TP (total protein).

https://doi.org/10.1371/journal.pone.0289704.g016

We found that UA and ALB.BCG can be accurately predicted by polynomial regression models constructed both for the photoacoustic and the absorption spectroscopy system (adjusted R-squared > 0.7 and p < 0.05) while CRE was only associated with a model for photoacoustic imaging features (R-squared = 0.83). Furthermore, although Fe, eGFR, GLU.AC, TG and ALT had significant correlation with photoacoustic or absorption spectroscopy imaging features, they could not be well explained by polynomial regression models. BUN had no significant correlation with either photoacoustic or absorption spectroscopy imaging features.

Discussion and conclusions

Previous research demonstrated the ability of the photoacoustic imaging system to extract tissue architecture, optical contrast, and molecular distribution based on the optical absorption characteristics of chromophores, such as hemoglobin, melanin, water or lipids [53]. Photoacoustic signals have been applied in many blood-related fields, such as probing the size and shape of red blood cells [6, 7] and measuring blood glucose concentration [10]. However, so far the full spectrum of biochemical blood parameters has not been explored by imaging techniques.

Furthermore, in our research, we utilized a 0.5 ms pulse width, which is larger than usual (10 ns). Generally, low pulse widths are used to detect the tissue texture. Meanwhile, to avoid the influence of the plastic tube on the propagating acoustic wave, we utilized a large pulse width and the laser was focused on the tip position of the plastic containers to excite temperature change A previous study has shown the enhancement of PA signals as the laser pulse width increased. At the same time, the temperature of the focal point also increased [54]. In the absorption spectroscopy imaging experiment, we compared the reference and the probe beam produced by the light passing through the blood sample and water, respectively, thus mitigating the influence of the plastic tube on the signal.

In this study, we investigated the extent to which time and frequency domain features derived from photoacoustic and absorption spectroscopy imaging signals reflect the values of biochemical blood parameters. Our results indicate that a number of imaging features are indeed informative in this respect and can therefore be potentially useful in the clinical diagnosis. We have also demonstrated that linear and non-linear combinations of photoacoustic/absorption spectroscopy imaging features can accurately predict some of the biochemical blood parameters. However, there are still some biochemical parameters that cannot be well predicted. In future studies, we intend to test different wavelengths for analyzing photoacoustic/absorption spectroscopy features of blood.

Another obvious limitation of the presented report is that it is based on a small number of samples (drawn from 14 individuals). We anticipate that future studies involving more samples will be able to achieve higher prediction accuracy. In addition, the same approach could also be applied to saliva analysis since most of the compounds found in blood are also present in saliva [55]. It would be very interesting to compare the results of the prediction models trained on blood and saliva samples. Finally, imaging-based technologies could be used for in vivo tests, thus reducing the need for conventional blood biopsies.

Supporting information

S1 Table. Multicollinearity analysis for the photoacoustic imaging system.

NA, no multicollinearity analysis was performed.

https://doi.org/10.1371/journal.pone.0289704.s001

(PDF)

S2 Table. Multicollinearity analysis for the absorption spectroscopy imaging system.

NA, no multicollinearity analysis was performed.

https://doi.org/10.1371/journal.pone.0289704.s002

(PDF)

S3 Table. Polynomial regression models for the photoacoustic imaging system.

https://doi.org/10.1371/journal.pone.0289704.s003

(PDF)

S4 Table. Polynomial regression models for the absorption spectroscopy imaging system.

https://doi.org/10.1371/journal.pone.0289704.s004

(PDF)

Acknowledgments

Special thanks to Ting-Ku Ou for drawing Fig 1B and 1D.

References

1. Leng X, Chapman W Jr., Rao B, Nandy S, Chen R, Rais R, et al. Feasibility of co-registered ultrasound and acoustic-resolution photoacoustic imaging of human colorectal cancer. Biomed Opt Express. 2018;9(11):5159–72. pmid:30460120
- View Article
- PubMed/NCBI
- Google Scholar
2. Moreira MC, Prado RN, Campos A. Application of High Brightness LEDs in the Human Tissue and Its Therapeutic Response. Appl Biomed Eng 1(InTech). 2011;3–20.
- View Article
- Google Scholar
3. Kumon Ronald E., Deng Cheri X., Wang Xueding. Frequency-Domain Analysis of Photoacoustic Imaging Data From Prostate Adenocarcinoma Tumors in a Murine Model. Ultrasound in Medicine & Biology. 2011;37(5):834–9. pmid:21376447
- View Article
- PubMed/NCBI
- Google Scholar
4. Pan QS, Qiu SY, Zhang SY, Zhang JB, Zhu S. Application of Photoacoustic Spectroscopy to Human Blood. Springer Series in Optical Sciences. 1988; 58: 542–5.
- View Article
- Google Scholar
5. Veverka M, Menozzi L, Yao J. The sound of blood: photoacoustic imaging in blood analysis. Medicine in Novel Technology and Devices. 2023;18:100219.
- View Article
- Google Scholar
6. Strohm EM, Berndl ES, Kolios MC. Probing red blood cell morphology using high-frequency photoacoustics. Biophys J. 2013;105(1):59–67. pmid:23823224
- View Article
- PubMed/NCBI
- Google Scholar
7. Fadhel MN, Hysi E, Strohm EM, Kolios MC. Optical and photoacoustic radiofrequency spectroscopic analysis for detecting red blood cell death. J Biophotonics. 2019;12(9):e201800431. pmid:31050867
- View Article
- PubMed/NCBI
- Google Scholar
8. Biswas D, Gorey A, Chen GCK, Sharma N, Vasudevan S. Investigation of diseases through red blood cells’ shape using photoacoustic response technique. SPIE. 2015;9322:93220K
- View Article
- Google Scholar
9. Cai C, Nedosekin DA, Menyaev YA, Sarimollaoglu M, Proskurnin MA, Zharov VP. Photoacoustic Flow Cytometry for Single Sickle Cell Detection In Vitro and In Vivo. Anal Cell Pathol (Amst). 2016;2016:2642361. pmid:27699143
- View Article
- PubMed/NCBI
- Google Scholar
10. Pai PP, Kumar Sanki P, De A, Banerjee S. NIR photoacoustic spectroscopy for non-invasive glucose measurement. Conf Proc IEEE Eng Med Biol Soc. 2015;7978–81. pmid:26738143
- View Article
- PubMed/NCBI
- Google Scholar
11. Tanaka Y, Tajima T, Seyama M, Waki K, Differential Continuous Wave Photoacoustic Spectroscopy for Non-Invasive Glucose Monitoring. IEEE Sensors Journal. 2020;20(8):4453–58.
- View Article
- Google Scholar
12. Cai C, Carey KA, Nedosekin DA, Menyaev YA, Sarimollaoglu M, Galanzha EI, et al. In vivo photoacoustic flow cytometry for early malaria diagnosis. Cytometry A. 2016;89(6):531–42. pmid:27078044
- View Article
- PubMed/NCBI
- Google Scholar
13. Menyaev YA, Carey KA, Nedosekin DA, Sarimollaoglu M, Galanzha EI, Stumhofer JS, et al. Preclinical photoacoustic models: application for ultrasensitive single cell malaria diagnosis in large vein and artery. Biomed Opt Express. 2016;7(9):3643–58. pmid:27699126
- View Article
- PubMed/NCBI
- Google Scholar
14. Galanzha EI, Shashkov E, Sarimollaoglu M, Beenken KE, Basnakian AG, Shirtliff ME, et al. In vivo magnetic enrichment, photoacoustic diagnosis, and photothermal purging of infected blood using multifunctional gold and magnetic nanoparticles. PLoS One. 2012;7(9):e45557. pmid:23049814
- View Article
- PubMed/NCBI
- Google Scholar
15. Lusk JF, Miranda C, Howell M, Chrest M, Eshima J, Smith BS. Photoacoustic Flow System for the Detection of Ovarian Circulating Tumor Cells Utilizing Copper Sulfide Nanoparticles. ACS Biomater Sci Eng. 2019;5(3):1553–60. pmid:33405628
- View Article
- PubMed/NCBI
- Google Scholar
16. Viator JA, Gupta S, Goldschmidt BS, Bhattacharyyal K, Kannan R, Shukla R, et al. Gold nanoparticle mediated detection of prostate cancer cells using photoacoustic flowmetry with optical reflectance. J Biomed Nanotechnol. 2010;6(2):187–91. pmid:20738074
- View Article
- PubMed/NCBI
- Google Scholar
17. Chen Z, Yang S, Xing D. In vivo detection of hemoglobin oxygen saturation and carboxyhemoglobin saturation with multiwavelength photoacoustic microscopy. Opt Lett. 2012;37(16):3414–16. pmid:23381275
- View Article
- PubMed/NCBI
- Google Scholar
18. Peters KS, Watson T, Logan T. Photoacoustic calorimetry study of human carboxyhemoglobin. J Am Chem Soc, 1992;114:4276–78.
- View Article
- Google Scholar
19. Zhou Y, Zhang C, Yao DK, Wang LV. Photoacoustic microscopy of bilirubin in tissue phantoms. J Biomed Opt. 2012;17(12):126019. pmid:23235894
- View Article
- PubMed/NCBI
- Google Scholar
20. Jeevarathinam AS, Pai N, Huang K, Hariri A, Wang J, Bai Y, et al. A cellulose-based photoacoustic sensor to measure heparin concentration and activity in human blood samples. Biosens Bioelectron. 2019;126:831–7. pmid:30602265
- View Article
- PubMed/NCBI
- Google Scholar
21. Wang J, Jeevarathinam AS, Humphries K, Jhunjhunwala A, Chen F, Hariri A, et al. A Mechanistic Investigation of Methylene Blue and Heparin Interactions and Their Photoacoustic Enhancement. Bioconjug Chem. 2018;29(11):3768–75. pmid:30281976
- View Article
- PubMed/NCBI
- Google Scholar
22. Wang J, Chen F, Arconada-Alvarez SJ, Hartanto J, Yap LP, Park R, et al. A Nanoscale Tool for Photoacoustic-Based Measurements of Clotting Time and Therapeutic Drug Monitoring of Heparin. Nano Lett. 2016;16(10):6265–71. pmid:27668964
- View Article
- PubMed/NCBI
- Google Scholar
23. Mauer J, Peltomäki M, Poblete S, Gompper G, Fedosov DA. Static and dynamic light scattering by red blood cells: A numerical study. PLoS One. 2017;12(5):e0176799. pmid:28472125
- View Article
- PubMed/NCBI
- Google Scholar
24. Abdul-Ghani MA, DeFronzo RA. Plasma glucose concentration and prediction of future risk of type 2 diabetes. Diabetes Care. 2009;32 Suppl 2:S194–8. pmid:19875551
- View Article
- PubMed/NCBI
- Google Scholar
25. Giordano C, Karasik O, King-Morris K, Asmar A. Uric Acid as a Marker of Kidney Disease: Review of the Current Literature. Dis Markers. 2015;2015:382918. pmid:26106252
- View Article
- PubMed/NCBI
- Google Scholar
26. Chapman MJ, Ginsberg HN, Amarenco P, Andreotti F, Borén J, Catapano AL, et al. Triglyceride-rich lipoproteins and high-density lipoprotein cholesterol in patients at high risk of cardiovascular disease: evidence and guidance for management. Eur Heart J. 2011;32(11):1345–61. pmid:21531743
- View Article
- PubMed/NCBI
- Google Scholar
27. Zhu Y, Lu JM, Yu ZB, Li D, Wu MY, Shen P, et al. Intra-individual variability of total cholesterol is associated with cardiovascular disease mortality: A cohort study. Nutr Metab Cardiovasc Dis. 2019;29(11):1205–1213. pmid:31383502
- View Article
- PubMed/NCBI
- Google Scholar
28. Levick C. How to interpret liver function tests. South Sudan Medical Journal. 2017;10:40–3.
- View Article
- Google Scholar
29. Perez V, Chang ET. Sodium-to-potassium ratio and blood pressure, hypertension, and related factors. Adv Nutr. 2014;5(6):712–41. pmid:25398734
- View Article
- PubMed/NCBI
- Google Scholar
30. McCallum L, Lip S, Padmanabhan S. The hidden hand of chloride in hypertension. Pflugers Arch. 2015;467(3):595–603. pmid:25619794
- View Article
- PubMed/NCBI
- Google Scholar
31. Robin FC, William MW, Michael CK. Photoacoustic detection of protein coagulation in albumen-based phantoms. Proceedings of the SPIE, 2008;6856:685626.
- View Article
- Google Scholar
32. Zhao Y, Yang S, Wang Y, Yuan Z, Qu J, Liu L. In vivo blood viscosity characterization based on frequency-resolved photoacoustic measurement. Applied Physics Letters. 2018; 113: 143703–143707.
- View Article
- Google Scholar
33. Alabugin A. Near-IR Photochemistry for Biology: Exploiting the Optical Window of Tissue. Photochem Photobiol. 2019;95(3):722–32. pmid:30536737
- View Article
- PubMed/NCBI
- Google Scholar
34. Salo D, Zhang H, Kim DM, Berezin MY. Multispectral measurement of contrast in tissue-mimicking phantoms in near-infrared spectral range of 650 to 1600 nm. J Biomed Opt. 2014;19(8):086008. pmid:25104414
- View Article
- PubMed/NCBI
- Google Scholar
35. Killick R, Fearnhead P, Eckley IA. Optimal Detection of Changepoints With a Linear Computational Cost. Journal of the American Statistical Association. 2012;107:1590–98.
- View Article
- Google Scholar
36. Karamzadeh N, Amyot F, Kenney K, Anderson A, Chowdhry F, Dashtestani H, et al. A machine learning approach to identify functional biomarkers in human prefrontal cortex for individuals with traumatic brain injury using functional near-infrared spectroscopy. Brain Behav. 2016;6(11):e00541. pmid:27843695
- View Article
- PubMed/NCBI
- Google Scholar
37. Feng T, Li Q, Zhang C, Xu G, Guo LJ, Yuan J, et al. Characterizing cellular morphology by photoacoustic spectrum analysis with an ultra-broadband optical ultrasonic detector. Opt Express. 2016;24(17):19853–62. pmid:27557261
- View Article
- PubMed/NCBI
- Google Scholar
38. Xu G, Davis MC, Siddiqui J, Tomlins SA, Huang S, Kunju LP, et al. Quantifying Gleason scores with photoacoustic spectral analysis: feasibility study with human tissues. Biomed Opt Express. 2015;6(12):4781–9. pmid:26713193
- View Article
- PubMed/NCBI
- Google Scholar
39. Hayes MH. Statistical Digital Signal Processing and Modeling. New York: John Wiley & Sons; 1996.
40. Stoica P, Moses R. Spectral Analysis of Signals. Upper Saddle River, NJ: Prentice Hall;2005.
41. Xu G, Fowlkes JB, Tao C, Liu X, Wang X. Photoacoustic spectrum analysis for microstructure characterization in biological tissue: analytical model. Ultrasound Med Biol. 2015;41(5):1473–80. pmid:25748521
- View Article
- PubMed/NCBI
- Google Scholar
42. Zhang H, Chao WY, Cheng Q, Huang S, Wang X, Wu D, et al. Interstitial photoacoustic spectral analysis: instrumentation and validation. Biomed Opt Express. 2017;8(3):1689–97. pmid:28663857
- View Article
- PubMed/NCBI
- Google Scholar
43. Xu G, Xue Y, Ozkurt ZG, Slimani N, Hu Z, Wang X, et al. Photoacoustic imaging features of intraocular tumors: Retinoblastoma and uveal melanoma. PLoS One. 2017;12(2):e0170752. pmid:28231293
- View Article
- PubMed/NCBI
- Google Scholar
44. Xu G, Meng ZX, Lin JD, Yuan J, Carson PL, Joshi B, et al. The functional pitch of an organ: quantification of tissue texture with photoacoustic spectrum analysis. Radiology. 2014;271(1):248–54. pmid:24475855
- View Article
- PubMed/NCBI
- Google Scholar
45. Yang Y, Wang S, Tao C, Wang X, Liu X. Photoacoustic tomography of tissue subwavelength microstructure with a narrowband and low frequency system. Applied Physics Letters. 2012;101, 034105.
- View Article
- Google Scholar
46. Harrell FE, Dupont C. Hmisc: harrell miscellaneous. R package version. 2008; 3:437.
- View Article
- Google Scholar
47. Shrestha N. Detecting Multicollinearity in Regression Analysis. American Journal of Applied Mathematics and Statistics. 2020;8(2):39–42.
- View Article
- Google Scholar
48. Vatcheva KP, Lee M, McCormick JB, Rahbar MH. Multicollinearity in Regression Analyses Conducted in Epidemiologic Studies. Epidemiology (Sunnyvale). 2016;6(2):227. pmid:27274911
- View Article
- PubMed/NCBI
- Google Scholar
49. Fox J, Weisberg S. An R Companion to Applied Regression. 3rd ed. Thousand Oaks CA: Sage;2019.
50. Tu YK, Kellett M, Clerehugh V, Gilthorpe MS. Problems of correlations between explanatory variables in multiple regression analyses in the dental literature. Br Dent J. 2005;199(7):457–61. pmid:16215581
- View Article
- PubMed/NCBI
- Google Scholar
51. Pena EA, Slate EH. Global Validation of Linear Model Assumptions. J Am Stat Assoc. 2006;101(473):341. pmid:20157621
- View Article
- PubMed/NCBI
- Google Scholar
52. Schwarz G. Estimating the dimension of a model. Ann Stat., 1978;6(2):461–4.
- View Article
- Google Scholar
53. Jnawali K, Chinni B, Dogra V, Rao N. Automatic cancer tissue detection using multispectral photoacoustic imaging. Int J Comput Assist Radiol Surg. 2020;15(2):309–320. pmid:31865531
- View Article
- PubMed/NCBI
- Google Scholar
54. Gao F, Feng X, Zhang R, Liu S, Ding R, Kishor R, et al. Single laser pulse generates dual photoacoustic signals for differential contrast photoacoustic imaging. Sci Rep. 2017;7(1):626. pmid:28377616
- View Article
- PubMed/NCBI
- Google Scholar
55. Yoshizawa JM, Schafer CA, Schafer JJ, Farrell JJ, Paster BJ, Wong DT. Salivary biomarkers: toward future clinical and diagnostic utilities. Clin Microbiol Rev. 2013;26(4):781–91. pmid:24092855
- View Article
- PubMed/NCBI
- Google Scholar

[ref1] 1. Leng X, Chapman W Jr., Rao B, Nandy S, Chen R, Rais R, et al. Feasibility of co-registered ultrasound and acoustic-resolution photoacoustic imaging of human colorectal cancer. Biomed Opt Express. 2018;9(11):5159–72. pmid:30460120
View Article
PubMed/NCBI
Google Scholar

[2] View Article

[3] PubMed/NCBI

[4] Google Scholar

[ref2] 2. Moreira MC, Prado RN, Campos A. Application of High Brightness LEDs in the Human Tissue and Its Therapeutic Response. Appl Biomed Eng 1(InTech). 2011;3–20.
View Article
Google Scholar

[6] View Article

[7] Google Scholar

[ref3] 3. Kumon Ronald E., Deng Cheri X., Wang Xueding. Frequency-Domain Analysis of Photoacoustic Imaging Data From Prostate Adenocarcinoma Tumors in a Murine Model. Ultrasound in Medicine & Biology. 2011;37(5):834–9. pmid:21376447
View Article
PubMed/NCBI
Google Scholar

[9] View Article

[10] PubMed/NCBI

[11] Google Scholar

[ref4] 4. Pan QS, Qiu SY, Zhang SY, Zhang JB, Zhu S. Application of Photoacoustic Spectroscopy to Human Blood. Springer Series in Optical Sciences. 1988; 58: 542–5.
View Article
Google Scholar

[13] View Article

[14] Google Scholar

[ref5] 5. Veverka M, Menozzi L, Yao J. The sound of blood: photoacoustic imaging in blood analysis. Medicine in Novel Technology and Devices. 2023;18:100219.
View Article
Google Scholar

[16] View Article

[17] Google Scholar

[ref6] 6. Strohm EM, Berndl ES, Kolios MC. Probing red blood cell morphology using high-frequency photoacoustics. Biophys J. 2013;105(1):59–67. pmid:23823224
View Article
PubMed/NCBI
Google Scholar

[19] View Article

[20] PubMed/NCBI

[21] Google Scholar

[ref7] 7. Fadhel MN, Hysi E, Strohm EM, Kolios MC. Optical and photoacoustic radiofrequency spectroscopic analysis for detecting red blood cell death. J Biophotonics. 2019;12(9):e201800431. pmid:31050867
View Article
PubMed/NCBI
Google Scholar

[23] View Article

[24] PubMed/NCBI

[25] Google Scholar

[ref8] 8. Biswas D, Gorey A, Chen GCK, Sharma N, Vasudevan S. Investigation of diseases through red blood cells’ shape using photoacoustic response technique. SPIE. 2015;9322:93220K
View Article
Google Scholar

[27] View Article

[28] Google Scholar

[ref9] 9. Cai C, Nedosekin DA, Menyaev YA, Sarimollaoglu M, Proskurnin MA, Zharov VP. Photoacoustic Flow Cytometry for Single Sickle Cell Detection In Vitro and In Vivo. Anal Cell Pathol (Amst). 2016;2016:2642361. pmid:27699143
View Article
PubMed/NCBI
Google Scholar

[30] View Article

[31] PubMed/NCBI

[32] Google Scholar

[ref10] 10. Pai PP, Kumar Sanki P, De A, Banerjee S. NIR photoacoustic spectroscopy for non-invasive glucose measurement. Conf Proc IEEE Eng Med Biol Soc. 2015;7978–81. pmid:26738143
View Article
PubMed/NCBI
Google Scholar

[34] View Article

[35] PubMed/NCBI

[36] Google Scholar

[ref11] 11. Tanaka Y, Tajima T, Seyama M, Waki K, Differential Continuous Wave Photoacoustic Spectroscopy for Non-Invasive Glucose Monitoring. IEEE Sensors Journal. 2020;20(8):4453–58.
View Article
Google Scholar

[38] View Article

[39] Google Scholar

[ref12] 12. Cai C, Carey KA, Nedosekin DA, Menyaev YA, Sarimollaoglu M, Galanzha EI, et al. In vivo photoacoustic flow cytometry for early malaria diagnosis. Cytometry A. 2016;89(6):531–42. pmid:27078044
View Article
PubMed/NCBI
Google Scholar

[41] View Article

[42] PubMed/NCBI

[43] Google Scholar

[ref13] 13. Menyaev YA, Carey KA, Nedosekin DA, Sarimollaoglu M, Galanzha EI, Stumhofer JS, et al. Preclinical photoacoustic models: application for ultrasensitive single cell malaria diagnosis in large vein and artery. Biomed Opt Express. 2016;7(9):3643–58. pmid:27699126
View Article
PubMed/NCBI
Google Scholar

[45] View Article

[46] PubMed/NCBI

[47] Google Scholar

[ref14] 14. Galanzha EI, Shashkov E, Sarimollaoglu M, Beenken KE, Basnakian AG, Shirtliff ME, et al. In vivo magnetic enrichment, photoacoustic diagnosis, and photothermal purging of infected blood using multifunctional gold and magnetic nanoparticles. PLoS One. 2012;7(9):e45557. pmid:23049814
View Article
PubMed/NCBI
Google Scholar

[49] View Article

[50] PubMed/NCBI

[51] Google Scholar

[ref15] 15. Lusk JF, Miranda C, Howell M, Chrest M, Eshima J, Smith BS. Photoacoustic Flow System for the Detection of Ovarian Circulating Tumor Cells Utilizing Copper Sulfide Nanoparticles. ACS Biomater Sci Eng. 2019;5(3):1553–60. pmid:33405628
View Article
PubMed/NCBI
Google Scholar

[53] View Article

[54] PubMed/NCBI

[55] Google Scholar

[ref16] 16. Viator JA, Gupta S, Goldschmidt BS, Bhattacharyyal K, Kannan R, Shukla R, et al. Gold nanoparticle mediated detection of prostate cancer cells using photoacoustic flowmetry with optical reflectance. J Biomed Nanotechnol. 2010;6(2):187–91. pmid:20738074
View Article
PubMed/NCBI
Google Scholar

[57] View Article

[58] PubMed/NCBI

[59] Google Scholar

[ref17] 17. Chen Z, Yang S, Xing D. In vivo detection of hemoglobin oxygen saturation and carboxyhemoglobin saturation with multiwavelength photoacoustic microscopy. Opt Lett. 2012;37(16):3414–16. pmid:23381275
View Article
PubMed/NCBI
Google Scholar

[61] View Article

[62] PubMed/NCBI

[63] Google Scholar

[ref18] 18. Peters KS, Watson T, Logan T. Photoacoustic calorimetry study of human carboxyhemoglobin. J Am Chem Soc, 1992;114:4276–78.
View Article
Google Scholar

[65] View Article

[66] Google Scholar

[ref19] 19. Zhou Y, Zhang C, Yao DK, Wang LV. Photoacoustic microscopy of bilirubin in tissue phantoms. J Biomed Opt. 2012;17(12):126019. pmid:23235894
View Article
PubMed/NCBI
Google Scholar

[68] View Article

[69] PubMed/NCBI

[70] Google Scholar

[ref20] 20. Jeevarathinam AS, Pai N, Huang K, Hariri A, Wang J, Bai Y, et al. A cellulose-based photoacoustic sensor to measure heparin concentration and activity in human blood samples. Biosens Bioelectron. 2019;126:831–7. pmid:30602265
View Article
PubMed/NCBI
Google Scholar

[72] View Article

[73] PubMed/NCBI

[74] Google Scholar

[ref21] 21. Wang J, Jeevarathinam AS, Humphries K, Jhunjhunwala A, Chen F, Hariri A, et al. A Mechanistic Investigation of Methylene Blue and Heparin Interactions and Their Photoacoustic Enhancement. Bioconjug Chem. 2018;29(11):3768–75. pmid:30281976
View Article
PubMed/NCBI
Google Scholar

[76] View Article

[77] PubMed/NCBI

[78] Google Scholar

[ref22] 22. Wang J, Chen F, Arconada-Alvarez SJ, Hartanto J, Yap LP, Park R, et al. A Nanoscale Tool for Photoacoustic-Based Measurements of Clotting Time and Therapeutic Drug Monitoring of Heparin. Nano Lett. 2016;16(10):6265–71. pmid:27668964
View Article
PubMed/NCBI
Google Scholar

[80] View Article

[81] PubMed/NCBI

[82] Google Scholar

[ref23] 23. Mauer J, Peltomäki M, Poblete S, Gompper G, Fedosov DA. Static and dynamic light scattering by red blood cells: A numerical study. PLoS One. 2017;12(5):e0176799. pmid:28472125
View Article
PubMed/NCBI
Google Scholar

[84] View Article

[85] PubMed/NCBI

[86] Google Scholar

[ref24] 24. Abdul-Ghani MA, DeFronzo RA. Plasma glucose concentration and prediction of future risk of type 2 diabetes. Diabetes Care. 2009;32 Suppl 2:S194–8. pmid:19875551
View Article
PubMed/NCBI
Google Scholar

[88] View Article

[89] PubMed/NCBI

[90] Google Scholar

[ref25] 25. Giordano C, Karasik O, King-Morris K, Asmar A. Uric Acid as a Marker of Kidney Disease: Review of the Current Literature. Dis Markers. 2015;2015:382918. pmid:26106252
View Article
PubMed/NCBI
Google Scholar

[92] View Article

[93] PubMed/NCBI

[94] Google Scholar

[ref26] 26. Chapman MJ, Ginsberg HN, Amarenco P, Andreotti F, Borén J, Catapano AL, et al. Triglyceride-rich lipoproteins and high-density lipoprotein cholesterol in patients at high risk of cardiovascular disease: evidence and guidance for management. Eur Heart J. 2011;32(11):1345–61. pmid:21531743
View Article
PubMed/NCBI
Google Scholar

[96] View Article

[97] PubMed/NCBI

[98] Google Scholar

[ref27] 27. Zhu Y, Lu JM, Yu ZB, Li D, Wu MY, Shen P, et al. Intra-individual variability of total cholesterol is associated with cardiovascular disease mortality: A cohort study. Nutr Metab Cardiovasc Dis. 2019;29(11):1205–1213. pmid:31383502
View Article
PubMed/NCBI
Google Scholar

[100] View Article

[101] PubMed/NCBI

[102] Google Scholar

[ref28] 28. Levick C. How to interpret liver function tests. South Sudan Medical Journal. 2017;10:40–3.
View Article
Google Scholar

[104] View Article

[105] Google Scholar

[ref29] 29. Perez V, Chang ET. Sodium-to-potassium ratio and blood pressure, hypertension, and related factors. Adv Nutr. 2014;5(6):712–41. pmid:25398734
View Article
PubMed/NCBI
Google Scholar

[107] View Article

[108] PubMed/NCBI

[109] Google Scholar

[ref30] 30. McCallum L, Lip S, Padmanabhan S. The hidden hand of chloride in hypertension. Pflugers Arch. 2015;467(3):595–603. pmid:25619794
View Article
PubMed/NCBI
Google Scholar

[111] View Article

[112] PubMed/NCBI

[113] Google Scholar

[ref31] 31. Robin FC, William MW, Michael CK. Photoacoustic detection of protein coagulation in albumen-based phantoms. Proceedings of the SPIE, 2008;6856:685626.
View Article
Google Scholar

[115] View Article

[116] Google Scholar

[ref32] 32. Zhao Y, Yang S, Wang Y, Yuan Z, Qu J, Liu L. In vivo blood viscosity characterization based on frequency-resolved photoacoustic measurement. Applied Physics Letters. 2018; 113: 143703–143707.
View Article
Google Scholar

[118] View Article

[119] Google Scholar

[ref33] 33. Alabugin A. Near-IR Photochemistry for Biology: Exploiting the Optical Window of Tissue. Photochem Photobiol. 2019;95(3):722–32. pmid:30536737
View Article
PubMed/NCBI
Google Scholar

[121] View Article

[122] PubMed/NCBI

[123] Google Scholar

[ref34] 34. Salo D, Zhang H, Kim DM, Berezin MY. Multispectral measurement of contrast in tissue-mimicking phantoms in near-infrared spectral range of 650 to 1600 nm. J Biomed Opt. 2014;19(8):086008. pmid:25104414
View Article
PubMed/NCBI
Google Scholar

[125] View Article

[126] PubMed/NCBI

[127] Google Scholar

[ref35] 35. Killick R, Fearnhead P, Eckley IA. Optimal Detection of Changepoints With a Linear Computational Cost. Journal of the American Statistical Association. 2012;107:1590–98.
View Article
Google Scholar

[129] View Article

[130] Google Scholar

[ref36] 36. Karamzadeh N, Amyot F, Kenney K, Anderson A, Chowdhry F, Dashtestani H, et al. A machine learning approach to identify functional biomarkers in human prefrontal cortex for individuals with traumatic brain injury using functional near-infrared spectroscopy. Brain Behav. 2016;6(11):e00541. pmid:27843695
View Article
PubMed/NCBI
Google Scholar

[132] View Article

[133] PubMed/NCBI

[134] Google Scholar

[ref37] 37. Feng T, Li Q, Zhang C, Xu G, Guo LJ, Yuan J, et al. Characterizing cellular morphology by photoacoustic spectrum analysis with an ultra-broadband optical ultrasonic detector. Opt Express. 2016;24(17):19853–62. pmid:27557261
View Article
PubMed/NCBI
Google Scholar

[136] View Article

[137] PubMed/NCBI

[138] Google Scholar

[ref38] 38. Xu G, Davis MC, Siddiqui J, Tomlins SA, Huang S, Kunju LP, et al. Quantifying Gleason scores with photoacoustic spectral analysis: feasibility study with human tissues. Biomed Opt Express. 2015;6(12):4781–9. pmid:26713193
View Article
PubMed/NCBI
Google Scholar

[140] View Article

[141] PubMed/NCBI

[142] Google Scholar

[ref39] 39. Hayes MH. Statistical Digital Signal Processing and Modeling. New York: John Wiley & Sons; 1996.

[ref40] 40. Stoica P, Moses R. Spectral Analysis of Signals. Upper Saddle River, NJ: Prentice Hall;2005.

[ref41] 41. Xu G, Fowlkes JB, Tao C, Liu X, Wang X. Photoacoustic spectrum analysis for microstructure characterization in biological tissue: analytical model. Ultrasound Med Biol. 2015;41(5):1473–80. pmid:25748521
View Article
PubMed/NCBI
Google Scholar

[146] View Article

[147] PubMed/NCBI

[148] Google Scholar

[ref42] 42. Zhang H, Chao WY, Cheng Q, Huang S, Wang X, Wu D, et al. Interstitial photoacoustic spectral analysis: instrumentation and validation. Biomed Opt Express. 2017;8(3):1689–97. pmid:28663857
View Article
PubMed/NCBI
Google Scholar

[150] View Article

[151] PubMed/NCBI

[152] Google Scholar

[ref43] 43. Xu G, Xue Y, Ozkurt ZG, Slimani N, Hu Z, Wang X, et al. Photoacoustic imaging features of intraocular tumors: Retinoblastoma and uveal melanoma. PLoS One. 2017;12(2):e0170752. pmid:28231293
View Article
PubMed/NCBI
Google Scholar

[154] View Article

[155] PubMed/NCBI

[156] Google Scholar

[ref44] 44. Xu G, Meng ZX, Lin JD, Yuan J, Carson PL, Joshi B, et al. The functional pitch of an organ: quantification of tissue texture with photoacoustic spectrum analysis. Radiology. 2014;271(1):248–54. pmid:24475855
View Article
PubMed/NCBI
Google Scholar

[158] View Article

[159] PubMed/NCBI

[160] Google Scholar

[ref45] 45. Yang Y, Wang S, Tao C, Wang X, Liu X. Photoacoustic tomography of tissue subwavelength microstructure with a narrowband and low frequency system. Applied Physics Letters. 2012;101, 034105.
View Article
Google Scholar

[162] View Article

[163] Google Scholar

[ref46] 46. Harrell FE, Dupont C. Hmisc: harrell miscellaneous. R package version. 2008; 3:437.
View Article
Google Scholar

[165] View Article

[166] Google Scholar

[ref47] 47. Shrestha N. Detecting Multicollinearity in Regression Analysis. American Journal of Applied Mathematics and Statistics. 2020;8(2):39–42.
View Article
Google Scholar

[168] View Article

[169] Google Scholar

[ref48] 48. Vatcheva KP, Lee M, McCormick JB, Rahbar MH. Multicollinearity in Regression Analyses Conducted in Epidemiologic Studies. Epidemiology (Sunnyvale). 2016;6(2):227. pmid:27274911
View Article
PubMed/NCBI
Google Scholar

[171] View Article

[172] PubMed/NCBI

[173] Google Scholar

[ref49] 49. Fox J, Weisberg S. An R Companion to Applied Regression. 3rd ed. Thousand Oaks CA: Sage;2019.

[ref50] 50. Tu YK, Kellett M, Clerehugh V, Gilthorpe MS. Problems of correlations between explanatory variables in multiple regression analyses in the dental literature. Br Dent J. 2005;199(7):457–61. pmid:16215581
View Article
PubMed/NCBI
Google Scholar

[176] View Article

[177] PubMed/NCBI

[178] Google Scholar

[ref51] 51. Pena EA, Slate EH. Global Validation of Linear Model Assumptions. J Am Stat Assoc. 2006;101(473):341. pmid:20157621
View Article
PubMed/NCBI
Google Scholar

[180] View Article

[181] PubMed/NCBI

[182] Google Scholar

[ref52] 52. Schwarz G. Estimating the dimension of a model. Ann Stat., 1978;6(2):461–4.
View Article
Google Scholar

[184] View Article

[185] Google Scholar

[ref53] 53. Jnawali K, Chinni B, Dogra V, Rao N. Automatic cancer tissue detection using multispectral photoacoustic imaging. Int J Comput Assist Radiol Surg. 2020;15(2):309–320. pmid:31865531
View Article
PubMed/NCBI
Google Scholar

[187] View Article

[188] PubMed/NCBI

[189] Google Scholar

[ref54] 54. Gao F, Feng X, Zhang R, Liu S, Ding R, Kishor R, et al. Single laser pulse generates dual photoacoustic signals for differential contrast photoacoustic imaging. Sci Rep. 2017;7(1):626. pmid:28377616
View Article
PubMed/NCBI
Google Scholar

[191] View Article

[192] PubMed/NCBI

[193] Google Scholar

[ref55] 55. Yoshizawa JM, Schafer CA, Schafer JJ, Farrell JJ, Paster BJ, Wong DT. Salivary biomarkers: toward future clinical and diagnostic utilities. Clin Microbiol Rev. 2013;26(4):781–91. pmid:24092855
View Article
PubMed/NCBI
Google Scholar

[195] View Article

[196] PubMed/NCBI

[197] Google Scholar

Figures

Abstract

Introduction

Materials and methods

Blood samples

Biochemical blood parameters

Photoacoustic imaging by laser pulses

Absorption spectroscopy imaging by light emitting diodes

Photoacoustic and absorption spectroscopy signal features

Time domain features

Positive and negative slopes

Frequency domain features

PASA slope, midband fit and intercept

Prominence and FWHM

Area under the photoacoustic signal waveform and frequency spectrum.

Correlation analysis

Model building

Results

Overview of the study

Feature selection and correlation analysis

Model building

Discussion and conclusions

Supporting information

S1 Table. Multicollinearity analysis for the photoacoustic imaging system.

S2 Table. Multicollinearity analysis for the absorption spectroscopy imaging system.

S3 Table. Polynomial regression models for the photoacoustic imaging system.

S4 Table. Polynomial regression models for the absorption spectroscopy imaging system.

Acknowledgments

References