Cancer detection based on electrical impedance spectroscopy: A clinical study

This study was performed with a bioimpedance measurement setup. The bio-impedance measurements, which will be explained in the following have been performed on both the left and right forearms and biceps of 21 subjects.

In-vivo experiments were performed on left and right forearm and bicep of eleven healthy and ten cancerous human subjects. For the in-vivo bioimpedance experiment, two Ag/AgCl electrodes were integrated into a probe. Electrical impedance electrodes, along with temperature sensors, were integrated into an enclosure by the computer aided design software SolidWorks. After that, the prototype was developed using a 3D printer and the rapid prototyping technique. In this design, which is shown in Figure 2, two Ag/AgCl electrodes, as well as six temperature sensors, were integrated into the probe [15]. The temperature sensors were used for another study, therefore the results of the temperature sensors are not presented in this paper [15]. The probe was designed in a way that the electrodes were at 2-centimeter distance. The input units for EIS play an important role in this system. Several factors that the input unit should cover are sensitivity, dynamic range, and speed. In order to address all of the preceding mentioned factors, the Model HF2TA trans-impedance amplifier from Zürich Instruments was used in this study. The frequency sweep range is selected from 300 Hz to 1 MHz with 50 step-points, which is enough to create fitted Nyquist plots and extract electrical properties of the tissue under test. Most of the patients in this study suffered from stage 2 or 3 skin cancer. Therefore, their tumors were 1-4 mm deep with ulceration or larger than 4 mm without ulceration (Figure 1).

Figure 1

Figure 2

In these cases, the area under study was the deeper living layers of the skin. As high frequency measurements reflects the deeper layers of skin [16], the chosen frequency range was suitable for capturing the cancer of the volunteer subjects.

The designed probe works in the safe current range of 20 μA. The probe’s supplier system has an isolation amplifier, which isolates the probe from power line and guarantees safety. The four features of accurate phase shift, high dynamic reserve, zero drift and orthogonality make this device a good choice over the other valid analog devices. The frequency response can also be derived from noisy and low voltage sets by using this setup.

A special area was considered for the gel (gel is used to fill the gap between sensors and skin) so that the used gel of each electrode is not mixed with the gel of the other electrode. In spite of the design of the electrode positions in such a way that the gels cannot be mixed, the electrodes were in a good contact with the tissue. As the focus of this study is not the stratum corneum, the gel influence on the measurements is neglected. The details of the probe design can be seen in Figure 3.

Figure 3

To have consistency in the results, the tests were performed on contralateral sites of the body. For bicep and forearm tests, the probe was placed at a distance 8 cm from elbow and wrist respectively. The finger TPS pressure sensor is designed to be worn over the examiner’s index finger to apply constant pressure over the probe (see Figure 4). The applied pressure was held constant by the examiner monitoring the pressure value from the Chameleon TVR software. The force level was chosen in a way that the pressure value was tolerable by the human subjects (2.5 N). The in-vivo admittances were measured at the force level in three trials and recorded for further analysis.

Figure 4

This section describes the experimental setup used for the in-vivo cancer detection study. The main purpose of this study was to utilize the electrical impedance spectroscopy as a means for cancer detection. Since the malignancy changes the tissue structure, the tissue electrical properties in the malignant tissue will also be different from the electrical properties of healthy tissue. This study was a collaborative research project with BC cancer agency (BCCA). The primary purpose of this study was to test EIS over breast cancerous subjects. However, the team oncologists’ suggestion was to test the method on skin tumors due to the sensitivity involved in breast cancer patients. This study was performed under the UBC BCCA Research Ethics Board under Dr. Paris Ann Ingledew as the Principal Investigator and REB Number H13- 02887. Ethics approval for conducting this study was given by the UBC BCCA Research Ethics Board and SFU Ethics Board. The title of the ethic sapproval is: “Electrical Impedance Analysis of Malignant and Benign Skin Tissue: Effects of Temperature and Pressure”.

In this study, ten cancerous subjects were recruited. The study was performed over malignant skin tumors with their contralateral healthy tissue as the control. The details of the skin cancerous subjects, their age, sex, type of tumor and tumor position are listed in Table 1.

Table 1

Case studies recruited for testing the proposed method on cancerous subjects and their information

The electrical properties of tissue change with compression [17]. Thus, in this study, the force sensors was utilized to monitor the amount of applied pressure over the tissue, so as to apply the same amount of pressure on all subjects and hence have consistency in results.

Various models and circuit theories have been proposed for modeling the multi-frequency electrical impedance data. These models consider the tested tissue as a circuit containing some resistances and capacitances. In this study parameter mapping of EIS will be presented using the popular model: the “Cole model” in the form of admittance. The structure of this circuit and its constitutive mathematical equations are explained as follows:

Cole model (in the form of admittance)

The circuit of the Cole model is illustrated in Figure 5. The admittance of human tissue can be expressed by the Cole equation (Eq. 1) in the form of admittance proposed by Liu et al. [18, 19]

Figure 5

Y=G+jB=G∞<!-- 8 -->+G0−<!-- - -->G∞<!-- 8 -->1+(jffyc)α<!-- a -->=1Z$$\begin{array}{} \displaystyle {\rm Y=G+jB=G_{\infty}+}\frac{\text G_{0}-\text G_{\infty}}{1+\bigg(\frac{\text j_{f}}{\text f_{yc}}\bigg)^{\alpha}}=\frac{1}{\text Z} \end{array}$$Eq.1

where Y is the whole admittance, G is the conductance and B is the susceptance. G₀ is the admittance at zero driving frequency, G_∞ is the admittance when the driving frequency is infinity (it should be known that G₀ and G_∞ are only valid within one single dispersion), f is the driving frequency, f_yc is the frequency at which the imaginary part of the admittance reaches its maximum and α is the dispersion parameter.

As stated by Liu et al. [20], the equation of admittance can form the equation of a circle, and thus plotting the imaginary part of admittance (susceptance) versus the real part of admittance (conductance) in a complex plane, results in a circular arc as shown in Figure 6.

Figure 6

The low-frequency region is affected by the extracellular environment and the high-frequency region is influenced by the intracellular space. In other words, since the cell membranes have a high capacitance, currents at low frequency cannot penetrate into the cell and must pass around the cells and go through the extracellular area. According to the Cole circuit, at low frequencies, open circuit occurs in the branch containing capacitance. Therefore, current flows only through the resistor called extracellular resistance. High frequency currents, on the other hand, have the ability to penetrate through cell membranes and other electronic barriers in the cell structure by polarisation. In the Cole circuit, the same behavior is observed at high frequencies in which the capacitor short circuit occurs and thus current flows through both resistances in parallel [21]. Based on these facts, the Cole model may represent the cellular properties of biological tissue in an efficient way, which is illustrated in Figure 7.

Figure 7

In this study, the use of contralateral sites of the body is proposed to be used as the control. Soft tissue is conductive, and the tissue conductivity is different in various types of tissues. Thus it is postulated that similar cellular structure is leading to similar electrical behavior [12, 23]. Moreover, the electrical properties of malignant tissues are altered compared to healthy tissue due to the increased cellular water and sodium content, altered membrane permeability, and changed packing density and orientation of cells [24]. As a result, in this study it is proposed to use the contralateral sites of the body as control. This idea suggests that if the tested tissue is healthy, the changes in electrical properties of that tissue on its contralateral site will not be significant and if the tested tissue is tumorous, a significant change in the electrical properties will be observed. In this section, the changes of the electrical properties of healthy and cancerous human tissues, about their contralateral parts, are studied.

To make a comparison, the bioimpedance measurements were conducted on the contralateral sites on biceps and forearms. The imaginary part of the admittance of each subject was plotted versus the real part of the admittance, thus the Nyquist plot of each subject was obtained. After that, the raw data were fitted into the Cole model and the corresponding electrical parameters were calculated. We compared the Nyquist plots of eleven healthy subjects’ biceps and forearms as well as their contralateral Nyquist plots. As a result of the comparison, the contralateral sites of the body have almost the same Nyquist bioimpedance results.

The same procedure in presenting the healthy subjects’ data was incorporated in this part; the imaginary part of admittance of each subject is plotted versus the real part of admittance, thus the Nyquist plot of each subject can be obtained. To compare the differences in admittances of tumorous and healthy tissues, the admittance of the tumorous part of each patient and the admittance of its contralateral healthy site were presented in the same Nyquist plot. After that, the raw data were fitted into the Cole model and the corresponding electrical parameters were calculated. In Figure 8, the Nyquist plots of the tumorous body parts of ten tumorous subjects as well as their contralateral healthy sites are illustrated and the results are visually comparable.

Figure 8

Figure 8 shows that the admittance Nyquist plots of contralateral sites in tumorous subjects significantly vary. As shown in this study, the use of electrical impedance of contralateral parts of body can be used as the control when performing tumor detection in human tissue. In this section, the electrical impedance spectra, measured from volunteer subjects, are used to train classification methods for automatic detection of malignancy.

Three classification methods are implemented on the raw data, the extracted features as well as the normalized features. Thereafter, the performance of each classification method on both raw and processed data are compared. Three different classifier techniques were used in this study: Naïve Bayes classifier with Gaussian distribution, Naïve Bayes classifier with Kernel distribution and support vector machine (SVM). Using electrical bioimpedance spectra of in-vivo tissue, the human subjects are classified into two classes: cancerous and healthy classes (10-fold cross validation, which is a popular choice of cross validation, was performed on the data).

A program was written in MATLAB for training and testing of NB classifiers (with Gaussian and Kernel distribution) and SVM classifier. The three classification methods have been applied on the raw data, the extracted features and the normalized extracted features. The accuracy, sensitivity, specificity, cross validation error and the misclassification error or resubstitution error of each method are illustrated in Figure 9 and Table 2.

Figure 9

Table 2

The performance of various classification methods on raw and processed data.

These results support the proposed idea of this research that the contralateral sites of body should have the same electrical properties unless there is a malignancy in one of the sites.