Keywords

1 Introduction

Digital twins in healthcare commonly refer to virtual representations or models of physical patients, organs, or processes. These virtual replicas can be created using data collected from various sources, including e.g. medical records and images or physiological measurements. Hence, digital twins aim to accurately mimic the behavior, characteristics, and responses of their real-world counterparts and they are considered to have many potential and significant benefits in the research of medical diagnosis and therapies [1,2,3,4,5]. Digital twins serve for several different purposes: a) Simulation and predictive analysis, b) Personalized medicine, c) Medical training and education, d) Remote Monitoring and Telehealth, and e) Research and Development. For instance, digital twins have been proposed e.g., for research on drug delivery, arrythmia, detecting severity of carotid stenoses from head vibration [1, 5]. The proposed digital twins in the literature are mainly computer models without possibility to test diagnosis device or treatment in practice.

There exist few 3D phantom-based studies in the literature which aim at realistic appearance and tissue thickness [6, 7]. However, very limited number of studies have been presented how digital twins could be developed using magnetic resonance images (MRI) and how they could be used in realistic simulation and emulation platforms for assessments of new medical monitoring applications [8,9,10] Moreover, up to authors’ knowledge, none of the previous studies have focused on brain tumor modelling.

1.1 Microwave-Based Brain Tumor Monitoring

Brain tumors are abnormal growth of cells in the brain which can be categorized to benign (noncancerous) or malignant (cancerous) [11]. Conventionally, brain tumors are detected by magnetic resonance imaging (MRI) scanning, computed tomography (CT) scans, and positron emission tomography (PET) [12, 13]. Recently, interest in using microwave sensing/imaging technology for brain cancer detection has increased since it enables a safe, rapid, low-cost, noninvasive solution which involves nonionizing radiation, and it can be realized with portable devices [14]. The microwave technique is based on detecting differences in radio channel responses caused by abnormalities having different dielectric properties than the surrounding tissues [14]. Microwave technique has been proposed for several different brain monitoring applications such as stroke detection [15, 16], brain temperature monitoring [17, 18], and cerebral circulation monitoring [19,20,21]. Additionally, its suitability for detection of brain tumors has been studied in [23,24,25,26,27].

1.2 Objectives and Contribution

The first aim is to show how realistic digital brain tumor twins are developed from gadolium enhanced T1 weighted MRI images. The second aim is to utilize digital twins to prepare a realistic emulation and simulation environments to evaluate microwave sensing based brain tumor detection technique. The emulation environment developed with the brain tumor digital twin and realistic human tissue mimicking phantoms, i.e. replicas of human tissues in terms of size, shape, and dielectric properties. The simulation environment is created in electromagnetic simulation software platform with the developed tumor simulation model embedded with anatomically realistic human head model. The third aim is to carry out evaluations with developed emulation and simulation platforms to investigate microwave sensing technique in brain tumor detection.

This paper is organized as follows: Sect. 2 describes in detail how the brain tumor digital twins can be prepared from MRI images. Section 3 describes the realistic simulation and emulation platforms. Section 4 presents results for measurements and simulations. Finally, Conclusions and future works in given in Sect. 5.

2 Preparation of Digital Twin of Brain Tumor from MRI

2.1 Tumor Mold Preparation

In this section, the procedure to prepare digital brain tumor twins is explained. It includes description how to develop brain tumor models from gadolium enhanced T1 weighted MRI images and how to prepare brain tumor phantoms as well as other relevant phantoms to be used in the realistic emulation platform. Additionally, use of brain twins in realistic simulation platforms is explained as well.

Realistic-shaped and -sized brain tumor digital twin is developed by using MRI images of real brain tumor (permission obtained from the patient), which is presented using FSLeyes software image in Fig. 1. The tumor is approximately 5 cm large and located on the left temporal lobe of the brain. The aim is to convert the MRI-image to a 3D model, which can be used directly in the simulation model and/or further convert a negation of the model to obtain a tumor mold which could be used for brain tumor phantom development.

Fig. 1.
figure 1

Brain tumor image represented with the FSLeyes software.

Full size image

Firstly, tumor's mask is created on each MRI slice image, as shown in Fig. 2a. Tumor is cropped from each MRI-image slice manually to isolate it from the rest of the image, as shown in Fig. 2b. As a result, the tumor model shown in Fig. 2c is obtained. The model is in.nii format, which is a raster format, with files generally containing at least 3-dimensional data: voxels, or pixels with a width, height, and depth. To convert this file to smoother format and suitable for 3D printing, the second software, 3DSlicer is used.

With 3D slicer software, pixelization can be removed and unnecessary holes of the model can be filled. Figure 3 illustrates tumor model before and after smoothening with 3DSlicer. Additionally, the 3DSlicer converts the tumor model to.stl format which is suitable to be used for 3D printing or in simulation software.

Finally, Fusion360 software is used to create a mold for phantom by performing the negation of the 3D model, as shown in Fig. 4. The mold is composed of two interlocking parts including a hole which allows the insertion of the phantom mixture before solidification. Finally, the mold parts are printed with 3D printer. The mold halves are presented in Fig. 5.

Fig. 2.
figure 2

a) Creation of a tumor's mask on each MRI slice image, b) isolation of the tumor from the rest of the MRI image, c) volume visualization of the tumor on the FSLeyes software.

Full size image
Fig. 3.
figure 3

Tumor model (on the left) before and (on the right) after the smoothening with 3DSlicer software.

Full size image
Fig. 4.
figure 4

Different views of mold modeling in the Fusion 360 software.

Full size image
Fig. 5.
figure 5

3D printed tumor molds.

Full size image

2.2 Tumor Phantom Preparation

Tumor phantom mixture is prepared using the recipe presented in [7] and repeated in Table 1. The aim is to obtain the same dielectric properties as measured for brain tumor in [29]. The cooking procedure is the following: First, the water is heated to 65 ℃ before gelatine is added. The oil is heated separately till 50 ℃ and added to the mixture at the same time with dishwashing liquid. The mixture is stirred smoothly and heated until 65 ℃ is achieved. The mixture is cooled slightly and poured into the mold for solidification. The solidified phantom is presented in Fig. 6.

Fig. 6.
figure 6

Solidified brain tumor phantom with size of 5 cm.

Full size image

3 Realistic Emulation and Simulation Platforms

3.1 Realistic Emulation Platform: Phantoms, Antennas, and Measurement Setups

3.1.1 Phantoms

The realistic emulation platforms include human tissue phantoms (skin and brain), tumor phantom, real human skull (borrowed from Pathology department), Vector Network Analyzer (VNA8720ES), and flexible ultrawideband (UWB) antennas.

Brain phantom is prepared using a realistic sized and shaped brain model retrieved from public 3D organ library The mold for brain phantom, illustrated in Fig. 7a, is prepared using Fusion360 software similarly to the brain tumor mold is prepared as explained in Sect. 2. The tumor and brain phantoms are prepared using the recipe depicted in Table 1 and which are originally presented in [7]. The solidified brain phantom is presented in Fig. 7b. Besides this average brain phantom, we also prepare a tumorous brain phantom in which the earlier developed tumor phantom is inserted inside the brain in the same location as depicted by MRI image. The skin phantom is also prepared based on the recipe given in [7] and summarized in Table 1. Solidified skin phantom is presented in Fig. 7c. The dielectric properties of the skin and brain phantoms are designed to closely match those of average human tissue at the selected frequencies, as outlined in ITIS foundation public datasheets [30].

Table 1. Recipes for skin, brain, and tumor phantoms [7].
Full size table
Fig. 7.
figure 7

a) Brain mold prepared with Fusion 360 software, b) solidified brain phantom, c) solidified skin phantom.

Full size image

3.1.2 Antennas

The antenna used in this study is a flexible monopole designed for in-body sensing for the frequency range 2–10 GHz covering both ISM band 2.5 GHz and UWB 3.1–10.6 GHz. The antenna is a slightly larger but improved version of the flexible antenna introduced in [31]. The size of the antenna is x = 40 mm, y = 40 mm, and z = 0.125 mm; it is fabricated on a thin flexible substrate Rogers5880 and is designed to be attached to the skin surface. Figure 8a–b presents the simulation model and the prototype of the antenna, respectively.

Fig. 8.
figure 8

The flexible UWB antenna used in the evaluations: a) simulation model, b) prototype.

Full size image

3.1.3 Measurement Setups

In this study, two measurements setups are required. The first one, consisting of Vector Network Analyzer (VNA) and SPEG’s probe and depicted in Fig. 9a, is used to evaluate dielectric properties of the phantoms before using them in the measurements. The second measurement setup consists of VNA, UWB antennas and absorber pieces which are built to avoid additional reflections to antennas from the surrounding items. The antennas are located around the voxel model’s head similarly to a portable helmet or band type of scenario. Measurements with VNA are carried out for the frequency range 2–8 GHz to evaluate antenna reflection coefficients S11 and S22 as well as channel transfer function S21 and S12.

Fig. 9.
figure 9

a) measurements set up with VNA and SPEAG probe to evaluate dielectric properties of the developed phantoms, b) measurements set up with VNA, real human skull, phantoms, and antennas to evaluate microwave sensing technology in brain tumor detection.

Full size image

3.2 Realistic Simulation Setup

The simulations are carried out using electromagnetic simulation software CST [32] which has several anatomically realistic human voxel models resembling humans having different sizes and body constitutions. For these evaluations, we chose the voxel model Hugo since it has the most realistic brain model. The antennas are located around the voxel model’s head similarly to a portable helmet or band type of scenario. Firstly, the simulations are carried out in the reference case, i.e. without any abnormalities. Next, the realistic tumor model is inserted into the brain model in the same location as in MRI-image hence resembling fully realistic scenario. S-parameters simulations are conducted only up to 5.8GHz to save simulation time which usually increases remarkably with voxel models with higher frequencies.

Fig. 10.
figure 10

a) Realistic CST’s Hugo-voxel model, b) vertical crosscut of the voxel’s head, c) horizontal cross-section of the voxel’s head illustrating the location of the realistic tumor model and antennas 1 and 2.

Full size image

4 Measurement and Simulations Results

4.1 Dielectric Property Measurements

Firstly, the dielectric properties of developed phantoms are verified with SPEG’s probe. The dielectric properties of the phantoms at 2.5 GHz and 6 GHz are presented in Table 2. Also, the corresponding reference values of average human tissue, obtained from [30], are included for comparison. As stated earlier, the dielectric properties of brain tumor are obtained from [29].

It is found that the dielectric properties of the phantoms are relatively close to those of the average human tissue values. Small changes can be seen especially in conductivity values. However, as presented in [33], small changes will not impact significantly on practical scenarios, e.g., in S-parameter evaluations.

Table 2. Dielectric properties of the developed phantoms and average tissues [29, 30]
Full size table

4.2 Measurement Results with Realistic Platform

Next, the measurements are carried out with phantoms, human skull, UWB antennas and VNA. The S21 parameters, i.e. the radio channel between the antennas 1 and 2 located on the opposite sides of the head, are presented in Fig. 11. As it can be seen the tumor is clearly visible in the channel response between the antennas located on the opposite side of the sides of the skull. The impact on the channel characteristics is frequency-dependent: at certain frequency ranges the presence of tumor increases the channel attenuation and the certain ranges decrease it. A similar tendency has been observed also in the other microwave-based detection studies [16, 17]. This phenomenon is due to the several reasons: difference in dielectric properties of the tumor and brain tissue vary with the frequency and thus the diffraction caused by tumor on the propagation signal may vary clearly. Additionally, antenna radiation characteristics may vary with frequency: at certain ranges there might be a radiation null towards the tumor and at certain ranges there could be a stronger lobe towards the tumor or on the sides of the tumor. All these aspects affect together on the propagation in vicinity of tumor and hence in general inside the tissue. Hence, careful study on optimal frequency range and optimal antenna locations with realistic models is essential for microwave-based detection applications.

When comparing the reference and tumorous results obtained from the measurements, there is always a small possibility of some uncertainties, e.g., due to unintentional movements of cables, or antennas. Thus, it is good to carry out comparative simulations in which these uncertainties are removed. The simulation results for a similar setup are presented in the next section.

Fig. 11.
figure 11

Measured S21 results obtained with phantoms and real human skull.

Full size image

4.3 Simulation Results

Finally, the simulations are carried out with CST and realistic brain tumor. The location of the tumor is set the same as with the measurement model and in MRI-images. In this case, the simulations are carried out only till 5.8 GHz to save simulation time which increases drastically as the frequency increases. Besides, at higher frequencies, the penetration of the radio signal through the whole head is not possible within due to higher propagation loss. The S-parameter results are presented in Fig. 12. It is found that also in this case, there is clear difference between the reference and tumorous cases: even up to 15 dB at 2.5 GHz. Channel attenuation decreases significantly after 3 GHz since the propagation loss in the tissue increases. Additionally. The flexible antennas are not directional and thus, less power can be directed towards the body than compared to the directive antennas. However, the advantage of the flexible antennas is excellent feasibility to the practical applications.

When comparing simulation and measurement results, it is noted that the channel attenuation is clearly stronger in the simulations than in the measurements. For instance, at 2.5 GHz, the level of S21 parameters is −60 dB in the simulations whereas only around −50 dB in the measurements. One reason for this is that the size of Hugo-voxel’s head is larger than that of a real human skull. Additionally, as noted in Fig. 10b, Hugo -voxel clearly has thicker muscle and fat layers than average in that area of the head where antennas (and tumor) are located. Especially in muscle tissue, the propagation loss is high due to high relative permittivity [30] and thus excessively thick muscle layer effects on the results remarkably. However, the trend is similar in both results: tumor causes clear differences in the channel responses. At lower frequencies, 1–1.5 GHz, the channel attenuation is stronger in the presence of tumor, whereas between 2–4 GHz it is vice versa. In the measurement the trend was somewhat similar, but the S21 curves have more fluctuation and thus there are some exceptions in this trend.

The reason for this trend is that, at lower frequencies, where the microwave signal penetrates the whole head easily, the tumor tissue with higher relative permittivity than brain tissue impedes propagation compared to the reference case. This can be seen as increased channel attenuation. Instead, at higher frequencies, where the propagation inside the tissue is more challenging due to higher loss, significant amount of the radio signal propagates as on the skin surface as creeping waves [33] or though the fat layer. In this antenna location setup case, the presence of tumor causes stronger diffractions towards on-body area, and thus the part of the in-body signal is summed to the signal travelling on the skin surface which is more prominent in this case. Consequently, the channel appears to be stronger in the presence of tumor. However, the antenna radiation characteristics at different frequencies have a clear impact on this. More comprehensive analysis of this phenomenon is left for the future work with the evaluations of several different types of UWB antennas.

Fig. 12.
figure 12

S21 parameter results with Hugo-voxel model obtained with the simulation in the presence and absence of the realistic brain tumor model.

Full size image

5 Summary and Conclusions

This paper presents a study on the development of digital twins for microwave sensing based brain tumor monitoring applications. The first aim was to show how realistic digital brain tumor twins can be developed from MRI and CT images. The second aim was to utilize digital twins to prepare a realistic emulation and simulation environment to evaluate microwave sensing based brain tumor detection technique. The emulation environment was developed with the brain tumor digital twin and realistic human tissue mimicking phantoms, the simulation environment was created in electromagnetic simulation software with the developed tumor simulation model embedded with anatomically realistic human head model.

This paper also presented realistic evaluations with the developed emulation and simulation platforms to investigate microwave sensing technique in brain tumor detection. Evaluations were carried out using flexible UWB antennas which are beneficial for practical solutions. The realistic simulation and emulation results show that microwave sensing is efficient in brain tumor detection also with flexible antennas. In order to have fully comparable simulation and measurement results, thicknesses of the tissues and also dimensions of the skull should be equal, otherwise there are clear differences in the S-parameter results especially if the antennas are located on the opposite sides of the head. Nevertheless, the trend was found to be similar in both cases: the presence of tumor clearly changes the radio signal propagation inside the brain which can be detected in S21 results. The changes are frequency dependent as is found also in previous studies [12].

The results presented in this paper are remarkable in several ways: the development of digital twins for healthcare is an important and timely topic since they could revolutionize research on medical/healthcare applications taking steps towards the future’s customized and personalized health care. In addition, digital twins with authentic texture of target tissue could be used in the training of microsurgical skills to neurosurgery residents The results of this paper outline the importance of usability of adjustable digital twins in the evaluations: they help to understand how for instance frequency selection and antenna location has to planned carefully with realistic and adjustable models since the tissue thicknesses affect clearly on the results, especially if aiming to do whole head scanning using antennas located opposite sides of the head. Moreover, the detectability of brain tumor with flexible antennas is a promising result for the research on practical applications: flexible antennas are easily feasible even with a head band type of portable monitoring device.

In this paper, we presented only pure frequency domain S-parameter data. As our future work, we will analyze microwave sensing in different signal domains and study the efficiency of different imaging algorithms. Furthermore, we will start developing digital twins for multimodal monitoring applications which includes design and development of multimodal phantoms.