1. Introduction

Blood is the most important body fluid, containing many kinds of proteins, carbohydrates, hormones, lipids, blood cells, and salts. Many biochemical variables in blood plasma, such as glucose and total cholesterol, are highly related to diseases such as diabetes and heart diseases. The concentration and molecular properties of most biochemical variables can only be examined ex-vivo. For example, triglycerides in plasma can only be examined by enzymatic colorimetric analysis [1]. In this method, certain enzymes are added into plasma to interact with the examined factor. The formed chemical compound absorbs visible light at a certain frequency, and the concentration of the factor can then be estimated according to the light absorption coefficient of the compound [1]. To avoid the visible light absorption by red blood cells, this method needs blood centrifuge. Recently, THz wave has been proved to have strong interactions with many biochemical molecules such as amino acids [2,3], proteins [4,5], and deoxyribonucleic acid (DNA) [6]. In aqueous solutions composed of water, THz wave has also been found to be a sensitive tool to investigate the collective bending vibration of hydrogen bond of water molecules [7–9]. Recent studies indicate that the THz absorption of water solution is highly sensitive to the concentration of various solutes [3–5,7,8] since the polarized solute would change the collective bending vibration of hydrogen bond. Blood has around 80% water content [10] and contains many kinds of polarized solutes, representing different risk factors of human. Comparing to existing enzymatic colorimetric examination methods, THz wave could be a potential candidate to monitor the concentrations of some critical biochemical variables in the human whole blood, and even to be applied for future in vivo wearable device examination [11,12] or in vivo molecular imaging applications [13–16].

In the past, there were two patents describing non-invasive THz blood examination devices [11,12] and three journal articles investigating the THz dielectric properties of the ex-vivo blood of human and rat [17–19]. In reference 17, the study was focused on blood plasma but not whole blood. In references [18,19], only less than 5 samples were studied. Since previous researches have indicated that proteins, ions, and lipids might all modify the THz absorption coefficients in aqueous solutions [4–9], and the concentrations of these factors in blood are with a huge variation by individuals, the investigated THz blood spectra of few samples would be more likely to have high deviations to the spectra of the general population. Conducting a clinical study which is designed to demonstrate the effect of a certain technique on a selected subset of the general population [20] is the most promising way to not only have a representative human blood spectra investigation, but also to find out the dominant factors that strongly affect the THz spectral characteristics of human whole blood.

In this work we have conducted a pilot clinical study to not only investigate the sub-THz spectra of ex-vivo fresh human whole blood of 28 patients following 8-hours fasting guideline, but also to find out the critical blood ingredients of which the concentration dominantly affects those spectra. A great difference between the sub-THz absorption properties of human blood among different people was observed, while the difference can be up to ~15% of the averaged absorption coefficient. With the observed smooth absorption spectral shape, we chose to statistically analyze the measured absorption coefficients at 270 and 820 GHz, to respectively representing the low sub-THz frequency region (between 100GHz and 500GHz) and the high sub-THz frequency region (between 500GHz and 1THz). The absorption coefficients at 270GHz of the 28 samples were found to be dominated by the red blood cell count with a significant negative correlation. At 820 GHz, the absorption coefficients were found to be significantly negative correlated to the concentration of triglyceride. For the other examined factors, no significant correlation was observed, probably due to the dominance of the red blood cell count fluctuation as well as the triglyceride concentration variation. Our study not only suggests the potential use of different sub-THz wavelengths to detect red blood cell counts and triglyceride concentration in human whole blood, but also suggests that these two dominant factors would need to be limited within a narrow range for future correlation investigation to the other biochemical factors in human whole blood.

2. Method

2.1 Experimental setup

A portable THz time-domain spectrometer (mini-Z, Zomega Inc.) was placed in the laboratory right next to the cardiac catheterization laboratory of National Taiwan University Hospital, to enable fresh human blood spectrum acquisition immediately after blood extraction from every experimental subject. The portable THz time-domain spectrometer used an external fiber-coupled 1.5 μm pulsed laser with a 140mW average power, with a less than 100fs pulse duration, and with a pulse repetition rate of 100 MHz as the pump source that was split into a pump and a probe beam for THz wave generation and detection. A large-aperture photoconductive antenna was used to generate the broadband THz wave and an electro-optic crystal was used to detect the THz waveform. The system measured the sub-THz spectrum from 0.1 THz to 3 THz. The peak dynamic range was greater than 48dB under the 0.5 second acquisition condition. To improve the signal to noise ratio, in this paper we present the 1-minute waveform by averaging 120 continuous 0.5-second waveforms in sequence numerically. The fluidic-sample chamber for transmission-type measurement was self-designed with 2 polyethylene (PE) windows, one was a flat PE cap and the other was with a 100μm-thick 1.5cm-wide fluidic channel at one side. The PE window and PE cap were sandwiched by two aluminum frames and locked by four nuts, and the injected liquid can thus be sealed. The photos of the chamber with human blood injection are shown in Fig. 1(a), 1(b) and 1(c). All the experiments were operated at a room temperature of 24°C with a stable humidity 60%. In each time of the measurement, the reference spectrum of the empty chamber was acquired one minute before the sample injection. Our measurement repeatability was tested by injecting bulk water, with the chamber assembled, fixed, injecting water, and disassembled three times. Figure 2 shows the measured absorption coefficient spectra of water for three measurements. The ratio of the standard deviation to the average value was less than 1% in all frequencies between 0.1 and 1.2THz. Due to high water absorption, no reliable data was acquired beyond 1.2 THz in this study.

 

Fig. 1 (a) Self-designed fluidic channel. (b) The PE cap with human blood. (c) The blood injection process, with a portable THz spectrometer.

Download Full Size | PDF

 

Fig. 2 Three-times measured absorption coefficients of bulk water shown as the top black, red, and blue dots. The bottom green dots show their standard deviations.

Download Full Size | PDF

2.2 Clinical protocol: blood spectra acquisition

This study was conducted according to the Declaration of Helsinki Principles, and the following protocol was approved by the Institutional Review Board of National Taiwan University Hospital. Informed consent was obtained from each subject prior to study entry. The human blood samples were obtained before cardiac catheterization, and all patients followed 8-hour fasting guidelines before the surgery. We started THz spectrum acquisition 3-4 minutes after blood extraction from patients, which was the shortest duration we could reach. 28 extracted blood samples were injected into heparin vacutainer (BD vacutainer). Heparin, having a high negative charge density, is the most common anticoagulant added in blood for examinations [21, 22]. The heparin concentration added was 15 USP (US Pharmacopeia) units of heparin per milliliter of blood [23].

In each measurement, after the reference spectrum of the empty chamber was acquired for one minute, the human whole blood was injected and its spectrum was recorded 2 spectra/second continuously also for 1 minute. Figure 3(a) shows two time-domain traces of the THz pulse measured after the empty chamber and two corresponding traces measured after the chamber filled with blood. To test the fluctuation of the TDS system, the spectrum of one of the samples was recorded continuously for 5 minutes. The time-domain traces of the two different blood samples (one was recorded for 1 minute and the other was recorded for 5 minutes) are shown in Fig. 3(b). As can be seen, the system fluctuation was much less than the sample difference. Figure 3(c) shows the corresponding frequency-domain spectra of the traces in Fig. 3(b). The absorption coefficients of sample 1 were 118cm⁻¹ at 270GHz and 206cm⁻¹ at 820GHz, and of sample 2 were 106cm⁻¹ at 270GHz and 190cm⁻¹ at 820GHz, respectively. According to the sample 2’s result, the system temporal fluctuation is negligible, when compared with the measured difference between samples.

 

Fig. 3 (a) Two time-domain traces of the THz pulse measured after the empty chamber and two corresponding traces measured after the chamber filled with blood. (b) The time-domain traces of two different blood samples. One was recorded for 1 minute and the other was recorded for 5 minutes. (c) The corresponding frequency-domain spectra of the traces shown in (b).

Download Full Size | PDF

2.3 Clinical protocol:examination items

For the examined variables, different biochemical risk factors were examined within 24 hours after blood extraction including HbA1c (%), glucose (mg/dl), total protein (g/dl), albumin (ALB) (g/dl), globulin (GLO) (g/dl), total cholesterol (T-CHO) (mg/dl), triglyceride (TG) (mg/dl), urea nitrogen (UN) (mg/dl), creatinine (CRE) (mg/dl), Ca²⁺ (mmol/L), Na⁺ (mmol/L), and K⁺ (mmol/L). Different hematological risk factors were examined one day before the cardiac catheterization, including red blood cell (RBC) count (M/μL), white blood cell (WBC) count (K/μL), platelet (PLT) count (K/μL), mean corpuscular hemoglobin (MCH) (pg), and mean corpuscular hemoglobin concentration (MCHC) (g/dL). The correlations between the spectra of all the 28 samples and these factors were analyzed. It is well known that these hematological risk factors do not fluctuate within one day [24].

The average age of the 28 patients were 62.5 years old. 3 of the 28 patients were women. We have checked the case histories of these patients. The two examined items, hyperlipidemia and diabetes mellitus which are directly related to blood contents were further investigated. 7 of them have both hyperlipidemia and diabetes mellitus, 8 of them only have hyperlipidemia, 3 of them only have diabetes mellitus, and 10 of them have no hyperlipidemia or diabetes mellitus. Since our blood samples were donated from the patients undergoing cardiac catheterization, these samples were more likely to have abnormal levels of HbA1c level, TG, and T-CHO. Thus, in this study we have another selected subset with these variables controlled within the normal range: HbA1c < = 5.7%, T-CHO level <200mg/dl, and TG level <200 mg/dl [25,26]. With these limitations, 13 samples were selected. It needs to be mentioned that these samples were not recruited from healthy donors, but we did analyze our data with the variables within the normal range.

2.4 Analysis

To analyze the measured THz dielectric properties of blood, simple calculations were applied and are described below. The spectrum of the THz electric field transmitted through the empty container E_ref(ω) is:

In our correlation analysis to find out the most dominant factor to affect the sub-THz absorption coefficient of human whole blood, we would calculate the Pearson correlation coefficient r between the values of the examined items and the measured absorption coefficients at 270GHz and 820 GHz. The Pearson correlation coefficient r is defined as [29]:

Since x and y are normally distributed random variables, Pearson correlation coefficient r also has a symmetric probability density function [32]. r has a higher probability density if its absolute value is smaller [32]. A statistical hypothesis called null hypothesis, which means r = 0, is set. P-value is a function defined as the probability, under the assumption of the null hypothesis existing, of obtaining a result equal to and more extreme than what was actually observed [33]. If this p-value is very small, usually less than or equal to a threshold value called the significance level α (traditionally α = 5%), it suggests that the observed data is inconsistent with the assumption that the null hypothesis is true [33]. In our analysis we apply the two-tail test, which allots half of the α in one direction (r>0 side) and half of the α in the other direction (r<0 side). As a result, we used p<0.05 as a threshold to define if the examined variable is correlated with the absorption of whole blood. If p<0.05 is satisfied, we then adopted the r value to state their correlation strength. If p>0.05, we say that no correlation is found between the examined variable and the absorption of whole blood.

2.5 The effect of the added heparin

To study the effect on the blood spectrum by heparin, we further recruited one healthy volunteer and split the donated 8-hour fasting blood into two parts: one without anticoagulant, and the rest one added into a heparin vacutainer tube (BD vacutainer) to measure the two blood spectra. The spectrum of the blood without adding anticoagulant was obtained 3-4 minutes after extraction before it became solid state. The spectra acquisition time both lasted for 1 minute respectively.

3. Result

Figure 4 shows the acquired 28 ex-vivo absorption coefficient spectra of fresh human whole blood with heparin added. The largest absorption coefficient difference between these samples was 17cm⁻¹ at 270GHz, which equaled to 15% of the average absorption coefficient (109.4cm⁻¹), and was 27cm⁻¹ at 820GHz, which equaled to 13% of the average absorption coefficient (201.6cm⁻¹). The absorption coefficient spectra of sample 1 and sample 2 introduced in Fig. 3 are also marked in Fig. 4. We can observe that the largest absorption coefficient between these samples was comparable to the absorption coefficient difference between water (the top line in Fig. 5) and whole blood. This strongly indicates that the sub-THz absorption coefficient should be sensitive to some factors in blood. Figure 5 shows the absorption coefficients of the blood sample with and without heparin added from the recruited healthy volunteer. The absorption coefficient showed no difference at 270GHz (< 1%), and the difference increased with frequency to 4cm⁻¹ (2% of the absorption coefficient) at 500GHz and to 10cm⁻¹ (5% of the absorption coefficient) at 820 GHz when heparin was added. As can be seen, the difference caused by heparin was not as high as the individual difference shown in Fig. 6. With a fixed heparin concentration added into all examined human whole blood, the heparin-induced spectral shift should be similar to all examined human whole blood shown in Fig. 6. Our statistical analysis is thus not heparin-related.

 

Fig. 4 The top line shows the absorption coefficient of water, and the other lines below show the different absorption coefficients of the whole blood samples, measured in the first minute, of the 28 patients with heparin added. Sample 1 and sample 2 marked in Fig. 5 correspond to the two samples introduced in Fig. 3(b) and 3(c).

Download Full Size | PDF

 

Fig. 5 The spectral characteristics of the single blood sample recruited from one healthy volunteer. Black open square: blood absorptions coefficient without anticoagulant. Red solid circle: with heparin added. The absorption coefficient showed no difference at 270GHz (< 1%), and the difference increased with frequency to 4cm⁻¹ (2% of the absorption coefficient) at 500GHz and to 10cm⁻¹ (5% of the absorption coefficient) at 820 GHz when heparin was added.

Download Full Size | PDF

 

Fig. 6 The sub-THz absorption coefficients of the 28 blood samples versus RBC count. (a) at 270GHz (negative correlation r = −0.383, two-tail p-value = 0.044) (b) at 820GHz (negative correlation r = 0.075, two-tail p-value = 0.736).

Download Full Size | PDF

With our analysis, RBC count was found to have a negative correlation with the absorption coefficient at 270GHz (r = −0.383, two-tail p-value = 0.044, n = 28, in Fig. 6(a)), but not at 820GHz (r = −0.067, two-tail p-value = 0.736, n = 28, in Fig. 6(b)), indicating the presence of other factors dominating the high sub-THz region over the red blood cell count. As a result, we found that for most whole blood samples having their TG levels within the normal range (<200mg/dl), the absorption coefficients at 820GHz were dominated by TG concentration with a negative correlation (r = −0.437, two-tail p-value = 0.037, n = 23, in Fig. 7(b)), but not that significant at 270GHz (r = −0.405, two-tail p-value = 0.056, n = 23, in Fig. 7(a)). For other examined factors, no correlation with the sub-THz absorption coefficients at both 270GHz and 820GHz was found. These results are summarized in Table 1 with the range of each examined item provided.

 

Fig. 7 The sub-THz absorption coefficients of the 23 blood samples with their TG levels less than 200mg/dl (a) at 270GHz (r = −0.405, p = 0.056) (b) at 820GHz (negative correlation r = −0.437, p = 0.037).

Download Full Size | PDF

Table 1. The correlations between the whole blood absorption coefficients and the examined variables. The range of each item is listed in parenthesis. r is the Pearson correlation coefficient, p is the two-tail p-value, and n is the sample number.

View Table | View all tables in this article

Then, by statistical analysis to the other subset composed of the selected 13 normal samples with the healthy range of HbA1c < = 5.7%, T-CHO level <200mg/dl, and TG level <200 mg/dl, we found that their absorption coefficients also had a negative correlation to the TG concentration at both 270GHz (r = −0.575, two-tail p-value = 0.04, in Fig. 8(a)) and 820GHz (r = −0.611, two-tail p-value = 0.027, in Fig. 8(b)), and had no significant correlation to other factors. These results are summarized in Table 2 with the range of each examined item provided. In Table 2, the smaller sample number results in a higher p-value of the negative correlation between the absorption coefficient at 270GHz and RBC count. The plot of the absorption coefficient at 270GHz and RBC count is exhibited in Fig. 9(a). On the other hand, as we categorized the absorption coefficients at both 270GHz and 820GHz into the 4 groups introduced in section 2.3 (with/without hyperlipidemia and with/without diabetes mellitus, according to the case histories), by calculating the mean and variance of the 4 groups, no significant difference between groups was found.

 

Fig. 8 TG concentration versus the absorption coefficients of the 13 blood samples having HbA1c < = 5.7%, T-CHO level <200mg/dl, and TG level <200mg/dl (a) at 270GHz (negative correlation r = −0.575, two-tail p-value = 0.04) (b) at 820GHz (negative correlation r = −0.611, two-tail p-value = 0.027).

Download Full Size | PDF

Table 2. The correlations between the whole blood absorption coefficients and the examined variables. The whole blood samples have the limitations of HbA1c < = 5.7%, T-CHO level <200mg/dl, and TG level <200 mg/dl. The range of each item is listed in parenthesis. r is the Pearson correlation coefficient and p is the two-tail p-value. For all the items n = 13.

View Table | View all tables in this article

 

Fig. 9 RBC count versus the absorption coefficients of the 13 blood samples having HbA1c < = 5.7%, T-CHO level <200mg/dl, and TG level <200mg/dl, (a) at 270GHz (r = −0.37, p = 0.213). (b) at 820GHz (r = 0.075, p = 0.807).

Download Full Size | PDF

4. Discussion

The negative correlation to the RBC count at 270GHz corresponds to the fact that RBC occupies the largest volume fraction of blood and varies a lot by individual between 37% to 52% [24], and to previous reports that RBC absorbs less THz power than plasma [17,18]. However, compared to the perfect inverse relation in reference 17 from only 3 rat examples, the p-value (0.044) of the inverse correlation was not that small as the sample size increased. This indicates that RBC count is not as dominant as shown in the previous animal study for sub-THz absorption in human whole blood and other factors also contribute to some extent. Moreover, we herein found that TG is another dominant factor for sub-THz absorption in human whole blood. The TG-dominant negative correlation in the sub-THz region has not been described in any previous researches. The most dominant lipids in blood include cholesterol and fatty acids [34]. As an essential blood lipid, a triglyceride is an ester derived from glycerol and three fatty acids [34]. Even though it is reasonable to speculate that the negative correlation between TG concentration and THz absorption could be due to the fact that lipids absorb much less THz power than water, it is however hard to apprehend why there was no significant correlation for T-CHO given that both T-CHO and TG have similar mass concentrations per unit volume of blood (most of the blood samples had their T-CHO within 100mg/dl and 200mg/dl, and TG within 50mg/dl and 200mg/dl). Both TG and cholesterol in the whole blood are carried by lipoproteins, including very-low-density lipoprotein (VLDL), low-density lipoprotein (LDL), high-density lipoprotein (HDL), and chylomicron with a density even lower than VLDL [34]. These four kinds of lipoproteins have different densities and different percentages of TG and cholesterol components, as summarized in Table 3. As shown in Table 3, the lipoprotein with a higher TG weight percentage has a lower density. While TG and CHO have similar mass concentrations per unit volume of blood, the volume faction of TG in the whole blood is higher than CHO. Furthermore, TG concentration in blood varies greater than T-CHO. These two features of TG might partly explain why TG dominates the correlation over T-CHO.

Table 3. The density and different percentages of TG and cholesterol composition of different kinds of lipoprotein

View Table | View all tables in this article

In summary, our results showed that human blood is a very complicated fluid which has the THz spectral characteristics different from the predicted ones of the simple aqueous solutions. To further study the correlations between the THz spectrum of the whole blood and other examined factors such as glucose, proteins and ions, our results indicate that both RBC count and TG concentration are the two primary parameters that needed to be limited within a narrow range.

5. Conclusion

In conclusion, a pilot clinical study of the sub-THz human blood spectroscopy was conducted. A great difference between the sub-THz absorption properties of human blood among different people was observed. We have found that TG concentration was the dominant factor to affect the whole blood absorption coefficient in the high sub-THz region and was with a significant negative correlation to the sub-THz absorption coefficient. On the other hand the RBC count was found to dominate the low sub-THz region and was with a significant negative correlation to the sub-THz absorption coefficient. For the concentration of the ions including Na⁺, Ca²⁺, and K⁺, the glucose level, and the concentration of the total proteins including albumin and globulin, no correlation to the absorption coefficient at the low and high sub-THz frequencies was observed in this clinical study. RBC count and TG concentration needs to be limited within a narrow range for future THz correlation investigation to the other factors in human whole blood.