I. INTRODUCTION
The advances in microwave technologies have contributed to their widespread usage from industry and engineering to healthcare. Being contactless, label-free, and reasonably inexpensive, microwave-based techniques are particularly attractive for biomedical applications. An essential advantage is non-destructive monitoring in real time being most important for living biological systems. Permittivity measurements, commonly over a relatively wide range of frequencies, provide valuable information on structural and functional properties of various biological objects. Thus, the existing techniques and associated metrology permit to accurately discriminate pathological conditions [Reference Grenier1, Reference Polevaya, Ermolina, Schlesinger, Ginzburg and Feldman2] and to observe different biological processes [Reference Irimajiri, Ando, Matsuoka, Ichinowatari and Takeuchi3, Reference Hayashi, Oshige, Katsumoto, Omori, Yasuda and Asami4]. Still, there is no optimal or universal tool for that. Evidently, any implementation as well as a sensor design should always match the specific requirements of a study and sample properties. So the development of new analyzing instrumentation and their real-world application are highly challenging. It may concern certain difficulties in usability during routine manipulations by medical staff, accurate calibration of sensors to be used in actual clinical practice, and other requirements on a test such as desired accuracy and sensitivity that are strongly influenced by the effect size, and data reproducibility.
Another big challenge when dealing with data derived from biomedical studies is that they are commonly non-normally distributed [Reference Sainani5, Reference Kitchen6] and may contain lots of missing values. Actually, it depends on a specific character of the parameters being studied. For instance, some values may contain outliers, or may be skewed due to the bias and confounding effects by multiple factors (age, gender, individual differences). For such non-normal variables, or small samples with large deviations from normality, standard statistical analysis may lead to incorrect inferences, so one should consider alternative approaches: data transformations, non-parametric tests, or statistical bootstrapping [Reference Sainani5].
The main focus of this work is the applicability of dielectric blood sensing in biomedical research and, most significantly, acquisition of clinically meaningful data derived from the dielectric measurements.
II. CONCEPT OF DIELECTRIC BLOOD SENSING
Specifically for blood measurements, the extensive results on dielectric properties of whole blood and its components are widely covered in literature. However, most results are from spectroscopy rather than at a single frequency. The approach to sensing we describe here supposes the permittivity measurements of red blood cells (RBCs) at a fixed frequency of 39.5 GHz, which corresponds to the γ-dispersion domain. This domain is attributed to a relaxation of free water molecules and occupies a large portion of the microwave range, but the changes in dielectric properties of biological objects are recognized throughout whole spectrum [Reference Schwan7]. Thus, we can implement our measurements at any single frequency within the 10–60 GHz range. In this particular case, the selection of such a frequency was influenced by: (1) waveguide size (5.2 × 2.6 mm) used in sensor design for measuring very small sample volumes (microliter); (2) operating only on the fundamental mode; and (3) a wavelength in waveguide, which should be in consistent with manufacturing tolerances of sensor elements. The main point is that we aim to detect clinically significant effect of Δε* rather than a frequency band where it occurs.
The idea to apply millimeter-wave dielectrometry specifically for clinical purposes goes back a long way, from the late 1980s onwards, when a series of control experiments revealed correlation between the dielectric behavior of different biological objects and physiological conditions [Reference Shchegoleva8]. Cumulative results and numerous observations on the dielectric response of RBCs to pharmaceuticals (hormones and blockers) motivated us to develop the technique for in vitro monitoring of adrenergic-receptor activity, which plays a crucial role in the pathogenesis of cardiovascular diseases. The interdisciplinary approach we use involves the analysis of Δε* of RBCs induced by the effect of β-blockers (BB) – the drugs commonly used in treatment of certain cardiovascular disorders – using the combination of theoretical and experimental techniques.
A) Single-frequency waveguide dielectrometry
The technique is based on the measurement of the complex reflection coefficient R* from a short-circuited multilayer structure (waveguide sensor) containing a sample by means of the six-port reflectometer (see Fig. 1) under the single-frequency mode (f = 39.5 GHz). Signal power in the waveguide transmission line does not exceed 10 mW; measurement time is 4 s per sample. A more detailed discussion of the measurement setup can be found in [Reference Krasov and Arkhypova9].
Fig. 1. Measuring setup based on a six-port measuring line and waveguide sensor.
The sensor is a short-circuited segment of rectangular waveguide (5.2 × 2.6 mm2), which holds up to 5–7 µl of sample with thickness of about 100 µm (Fig. 2). A distinguishing feature of the sensor is its measuring sensitivity improvement at the operating frequency by additional dielectric inserts of different shapes (Figs 2(c)and 2(d)) that has been realized specifically for the repeated microliter measurements of blood samples [Reference Krasov10].
Fig. 2. Waveguide measuring sensor (a) and its simulation models: basic structure (b) and modified by dielectric inserts (c, d).
Such sensor modifications have been studied with use of the two-dimensional finite-difference time-domain technique [Reference Sirenko, Ström and Yashina11] and the software package written in Python both for optimization and calibration of the measuring structure. We used pure water and ethanol/water mixtures with mole fractions of 0.04 and 0.08 as reference liquids [Reference Sato and Buchner12] both for simulation and measurements. Standard deviation of 30 measurements of each reference liquids is ~1.2% in average and ~2% at a maximum. Our results showed that the required accuracy of the calibration, which should be at least 2.5%, can be achieved by means of the proposed procedure. Further details concerning the calibration procedure can be found in [Reference Krasov and Arkhypova13].
B) Parallel method
To evaluate receptor-membrane reaction and to validate the results of dielectric study, we have opted for the osmotic fragility test. It refers to the degree of susceptibility to hemolysis (i.e., a disintegration of RBCs with the release of hemoglobin into the intercellular medium) that occurs when RBCs are subjected to osmotic stress by being placed in a hypotonic solution. The method we used exploits a principle of hemolysis inhibition in the presence of BB propranolol, which binds to β-adrenergic receptors (β-ARs) on red-cell membranes and thereby reduces damage degree [Reference Stryuk and Dlusskaya14]. The β-ARs activity is numerically expressed by β-adrenergic reactivity of membranes (ARM) index; according to the guideline in [Reference Stryuk and Dlusskaya14], about 93% of healthy individuals have reference values in the range 2–20 units. The index values and receptors activity have reciprocal correspondence: β-ARM index will be numerically increased as the osmotic resistance is decreased, and vice versa. This method is commonly used for the diagnosis of hyper-adrenergic states, prognosis of cardiovascular diseases, and assessment of individual sensitivity to BB in therapy.
C) Research protocol
The research protocol is based on tracing drug-induced changes in RBC permittivity measured before and after the effect of BB. Measurements were performed using RBC samples (hematocrit = 0.85) extracted from heparinized venous blood (5 ml) pretherapy collected from patients with neurological disorders. More details about the sample preparation can be found in [Reference Arkhypova15]. The samples were subdivided into two groups (A, from n = 31 with chronic encephalopathy; and B, from n = 29 with acute stroke). The comparison group (control) comprised healthy individuals (n = 27, aged 55.2 ± 9.4) without any medical intervention whose indicators were baseline. To characterize β-ARs activity, the RBC samples were incubated with BB propranolol (C = 7.5 g/l) for 30 min at room temperature (25°C). The measurements of permittivity are carried out immediately after the incubation. All manipulations with blood were conducted during 3 h after its sampling (Fig. 3).
Fig. 3. Infographics of the research protocol.
Data handling and statistical analysis were performed using free software based on programming language R [Reference Fellows16]. It also allowed us to create a multiparameter database, which in addition to permittivity measurements contains other indicators: patient demographics, clinical data, sample temperature, the results of repeated measurements when necessary, etc., thus enabling us to consider all important information during analysis.
III. EXPERIMENTAL RESULTS
The results of the study are presented as box plots (Figs 4 and 5). They include summarized data of permittivity measurements of RBC samples as part of the case–control study (i.e., comparison of dielectric response in health and diseased patient samples) and information on the statistical inferences (P-values). We opted for non-parametric tests since some raw data were skewed and did not follow a normal distribution. Hence, comparing medians (ranked data) rather than mean values is preferred because the median is not affected by extreme values. Within-group differences (Fig. 4) were analyzed using the Wilcoxon sign-rank test, whereas between-group differences (Fig. 5) using the Wilcoxon rank-sum test (also known as the Mann–Whitney U test).
Fig. 4. Complex permittivity of human erythrocytes (f = 39.5 GHz, T = 25°C) in the control and the diseased groups (A and B) before and after the drug effect.
Fig. 5. Comparison of the RBC dielectric response (Δε*) to the BB in health and disease.
Fig. 4 shows the distribution of measured absolute values of RBC permittivity (ε′ and ε″ separately) in all study groups, where their variability outside the upper and lower quartiles (whiskers), the medians and means (marked with a dot) may be observed. Here, we analyze the response of erythrocytes to the BB effect, i.e., the comparison of permittivity values of the samples of the same individual with and without the drug (with equivalent adding of physiological saline instead).
It should be also noted that since ε* measurements have been conducted at different times, the correction for temperature of samples was made. Thus, we adjusted all absolute values to 25°C (average value); the procedure is partially described in [Reference Arkhypova, Krasov and Fisun17].
As illustrated (in Fig. 4), there are statistically significant differences in the permittivity of RBCs of the control group and those patients with mild form of neurological disorder (group A), whereas in those with an acute stroke (group B), the change in ε′ to the drug is statistically non-significant. The same effect can be observed in imaginary part of the permittivity.
However, the observable effects and the main idea of the developed test are more clearly illustrated in Fig. 5. It demonstrates how exactly the dielectric response (Δε*) of red cells to BB effect is distributed in health and disease.
As can be seen from the graph (Fig. 5), drug-induced Δε* depends on physiological condition of blood donors and can be detected with use of microwave dielectrometry. This particular finding indicates a high potential of the method for diagnostic purposes.
The data derived from the dielectric study are in good agreement with the results of the parallel test (β-ARM indices) represented in Table 1.
Table 1. Measured values and statistical inferences.
*P-value < 0.05 is statistically significant, **NS denotes statistically non-significant difference.
The degree of β-AR activity of erythrocytes is essentially reduced in the diseased groups (A and B) compared with the control; this is apparent from the indices, which exceed the reference ones (2–20 units), which, in turn, correspond to low-level receptors activity caused by a disease [Reference Stryuk and Dlusskaya14]. Thus, the statistically significant (r = +0.3, p = 0.02, the Spearman rank) correlation between the magnitude and sign of dielectric response and β-ARM indices is observed. A lowering of the permittivity is observed only in healthy individuals, which confirms the effect previously observed in [Reference Arkhypova15] with the lesser sample size.
IV. CONCLUSION
The applicability of dielectric techniques in the biomedical area is highly dependent on study design, specific properties of samples, and other technical and methodological aspects that should be considered. The described multidisciplinary approach to the characterization of cellular membrane activity is based on blood cell permittivity measurements in conjunction with the parallel biochemical testing and is intended to supplement current cellular analysis used for healthcare. We imply that the approach may be helpful for specialists dealing with dielectric measurements and the developers of clinically oriented instrumentation based on microwave dielectrometry.
Kateryna Arkhypova received the M.Sc. degree in Physics from V.N. Karazin Kharkiv National University (KhNU), Kharkiv, Ukraine in 2006, and earned her Ph.D. in Biophysics from KhNU in 2012. She currently holds an appointment as a senior research scientist at the O.Ya. Usikov Institute for Radiophysics and Electronics NAS of Ukraine. Her research interests are in clinically oriented microwave applications for diagnostic and therapeutic purposes.
Pavlo Krasov received his M.Sc. degree in Physics from V.N. Karazin Kharkiv National University (KhNU), Kharkiv, Ukraine, in 2003, and his Ph.D. in Radiophysics at the Kharkiv National University of Radio Electronics in 2011. He is currently serving as a senior research scientist at O.Ya. Usikov Institute for Radiophysics and Electronics NAS of Ukraine. His research interests are in design of experimental devices and measuring sensors for biomedical applications, scientific, and hardware engineering.
Prof. Anatolii Fisun received a Master's degree in Physics from the Kharkiv State University in 1967 and received his Ph.D. degree in Radiophysics in 1977. He currently holds an appointment as chief researcher of Solid-State Electronics Department at O.Ya. Usikov Institute for Radiophysics and Electronics NAS of Ukraine. His main research interests are design and optimization of resonant quasi-optical systems and high-stable solid-state generators in millimeter-wave band.
Andriy Nosatov, MD, Head of the Department of Neurology of Public Health Institution “Kharkiv Municipal Hospital No 7”, Kharkiv, Ukraine. The research interests are mainly in neuroscience, developing and optimizing existing techniques in treatment and rehabilitation of cerebrovascular patients.
Volodymyr Lychko, MD, Ph.D. In 2000, he graduated from the Medical Department of Sumy State University. He earned his Ph.D. degree in Medical Sciences in 2010. He currently holds an appointment as a research scientist at the Department of Neurosurgery and Neurology of the Medical Institute of Sumy State University. His main research interests are molecular and cellular neurobiology and neuroscience.
Volodymyr Malakhov, MD, Ph.D., received his Ph.D. degree in Medicine from Kharkiv Medical Academy of Post-Graduate Education in 1990. Since 2007 he holds the chair of the Department of Rehabilitation, Sports Medicine and Exercise Therapy at the Kharkiv Medical Academy of Post-Graduate Education. His main research interests are molecular and cellular neurobiology, development of new therapy strategies, and optimizing existing treatments for acute and chronic neurological disorders in rehabilitation of cerebrovascular patients and sportsmen.
