Real-time imaging of traumatic brain injury using magnetic induction tomography

Objective. Early diagnosis of traumatic brain injury (TBI) is crucial for its prognosis; however, traditional computed tomography diagnostic methods rely on large medical devices with an associated lag time to receive results. Therefore, an imaging modality is needed that provides real-time monitoring, can easily be carried out to assess the extent of TBI damage, and thus guides treatment. Approach. In the present study, an improved magnetic induction tomography (MIT) data acquisition system was used to monitor TBI in an animal model and distinguish the injury level. A pneumatically controlled cortical impactor was used to strike the parietal lobe of anesthetized rabbits two or three times under the same parameter mode to establish two different rabbit models of TBI. The MIT data acquisition system was used to record data and continuously monitor the brain for one hour without intervention. Main results. A target with increased conductivity was clearly observed in the reconstructed image. The position was relatively fixed and accurate, and the average positioning error of the image was 0.013 72 m. The normalized mean reconstruction value of all images increased with time. The slope of the regression line of the normalized mean reconstruction value differed significantly between the two models (p < 0.0001). Significance. This indicates that in the animal model, the unique features of MIT may facilitate the early monitoring of TBI and distinguish different degrees of injuries, thereby reducing the risk and mortality of associated complications.


Introduction
Traumatic brain injury (TBI) is a common, severe disease and a leading cause of death and disability. According to recent reports (Jiang et al 2019), over 50 million patients are affected by TBI worldwide annually, resulting in economic losses of approximately 400 billion U.S. dollars. Even with effective treatment, the mortality rate of TBI is 20%-30%, and the disability rate can be 50%-60% (Servadei et al 2000). It also results in tremendous economic and psychological burdens to society and families (Maas et al 2017). The primary imaging examination method for TBI is head computed tomography (CT). Examination methods such as the observation of vital signs and intracranial pressure measurements are also essential. Early diagnosis and evaluation of the injury, as well as continuous dynamic monitoring, will greatly contribute to reduced mortality and disability due to craniocerebral injury. CT, which requires large imaging equipment, is most commonly used to diagnose craniocerebral injury (Servadei et al 2000). Furthermore, CT is only available at medical centers and exposes patients to ionizing radiation. The transfer of patients from the trauma site to the medical center is a blank area for intracranial imaging, and CT cannot be used for continuous monitoring. Thus, an imaging modality that can provide real-time, non-invasive examinations to assess the extent of the injury and thereby guide treatment is urgently required.
In recent years, the rapid development of bioelectrical impedance technologies has enabled its application in medicine. Several commercialized and clinically approved electrical impedance tomography (EIT) systems are available for intensive care unit monitoring, particularly for the brain and chest (ventilation-perfusion monitoring) (Al-Zeibak and Saunders 1993, Jang et al 2019. Traditional EIT has a broad prospect in electrophysiological measurement, diagnosis, and real-time monitoring of brain disorders, such as epilepsy (Witkowska-Wrobel et al 2018), stroke (Dowrick et al 2016), and brain edema . However, the poor conductivity of the skull and the electrode-skin contact impedance limit the development of EIT for brain disease monitoring. Magnetic induction tomography (MIT) is an emerging non-contact EIT technology that uses the principle of eddy current induction to detect the distribution or change in conductivity inside the human body because of the strong permeability of magnetic fields to the skull (Ma and Soleimani 2018), and it has been proved that the magnetic induction method can be used to monitor TBI in animals . Therefore, MIT has advantages in monitoring intracranial injury. Furthermore, compared with current medical imaging technology, MIT is a new technology that has not been applied in a clinical setting to date, which would provide non-contact real-time dynamic monitoring (Mahdavi andRosell-Ferrer 2017, Xiang et al 2020), ensuring safety and convenience in future medical applications. The unique advantages of MIT make it promising for early TBI monitoring.

Materials and methods
2.1. Establishment of the controlled cortical impact (CCI) model Ten healthy adult rabbits (2.0-2.5 kg) were obtained from the Animal Experiment Center of the Fourth Military Medical University (FMMU) in Xi'an, China. All procedures were approved by the Institutional Animal Care and Use Committee of the FMMU and complied with animal experimental guidelines. The rabbits were fasted prior to the operation, and all surgeries were performed using an aseptic technique. The hair on the skull was removed, and 3.5% sodium pentobarbital (1 ml kg −1 ) was injected into the ear vein to induce anesthesia.
The rabbit was laid prone on the operating table and disinfected, after which the top of the forehead was shaved. A 5 cm midline incision of the head was used to bluntly separate the soft tissue and periosteum, exposing the skull. A dental drill was used to drill through the skull, layer by layer, to perform unilateral bone flap removal 5 mm to the right outside the sagittal suture and 5 mm behind the coronal suture. The diameter of the bone hole was 5 mm, and the dura remained intact. A gelatin sponge was used to stop the bleeding, and the operating field was cleared (figures 1(a), (b)). The rabbits were then fixed in a stereotactic frame. A pneumatically controlled cortex impactor (RWD Life Science Co., Ltd, Shenzhen, China) equipped with a firing pin 4 mm in diameter was used for the operation. The motion profile of the tip of the impactor was programmed to exit 20 mm from the surface, then provide a downward stroke of 25 mm (i.e. an indentation depth of 5.0 mm) at a preset speed of 3.0 m s −1 and an indentation duration of 0.5 s. The intact dura mater was impacted (Osier and Dixon, 2016), and the impact was performed twice in succession to establish injury in Group A. Three consecutive impacts with the same parameters were used to establish injury in Group B (figure 1(c)).
After the impact, the surgical site was sutured, and the rabbit was immediately placed on the monitoring table for examination using MIT (figure 1(d)), and data were continuously collected for one hour. During this period, the rabbits were kept in an anesthetized static state without any intervention. A series of continuously measured data frames were obtained.
After monitoring, each animal was deeply sedated and perfused with a frozen heparinized phosphate buffer and 4% frozen buffer paraformaldehyde. The brain was removed, fixed at 4°C for 2-5 h, and then stored frozen in 30% sucrose for 48 h. Coronal sections (25 μm thick) were cut in a cryostat, pasted on glass slides coated with gelatin, and dried overnight at 37°C. The samples were rehydrated in a descending alcohol series in distilled water and then soaked in 0.1% cresol violet acetate for 7 min. The brain slices were dehydrated in an ascending alcohol series, cleared with xylene, and covered with a mounting medium. Images were obtained using a BX60F microscope (Olympus Corporation, Tokyo, Japan) with a 1 × objective lens (Watts et al 2015). Microsoft ICE software (Microsoft Corporation, Redmond, WA, USA) was used to assemble individual images into the wholebrain image (Watts et al 2015). Six slices at the center of the brain damage were taken from each animal. The injured brain areas were measured, and the largest damaged area was selected for further analyses (figures 2(a), (b)).

MIT imaging
The MIT monitoring data system is an improved FMMU MIT system developed through cooperation between the FMMU and Hangzhou Utron Technology Co., Ltd (Zhang et al 2021). As shown in figure 3(a), the rabbit was placed right below the center of the device, and the head area was 5.6 cm away from the bottom coil. Figure 3 (b) shows that the MIT equipment primarily consists of 16 PCB-type coils uniformly arranged on the circumference of the 21 cm diameter, in which the maximum radius of the circular coil on the PCB is 20 mm. The excitation current of the coil is set to 420 mA and the frequency to 21 MHz. When one of the coils is the excitation coil, the remaining 15 are the measurement coils; therefore, one frame of measurement data contains 240 values. We set the first frame of all monitoring data as the reference frame and the subsequent data as the foreground frame. In the resulting images, we defined regions of interest (ROIs) that exceed the maximum reconstruction value of 60% (Zhang et al 2022). In the present study, for the inverse MIT problem, a circular 2D mesh for the coil plane comprising 800 triangle elements was introduced, as depicted in figure 3(c). This is a general standard circular finite element model with the same size as the sensing area of hardware equipment used for image reconstruction. Figure 3(d) is a schematic diagram of the ROI selection, and the positioning error of the imaging result is evaluated by calculating the difference between the geometric median of the ROI ( ) x y , ROI ROI and the The rabbit was fixed in the stereotactic frame and a pneumatically controlled cortical impactor (RWD Life Science Co., Ltd, Shenzhen, China) was used to impact the complete dura mater. (d) After the impact, the surgical site was sutured, and the rabbit was immediately placed on the monitoring table for examination using MIT. (The labels L, R, A and P represent left, right, anterior, and posterior, respectively.) Figure 2. MIT was used to monitor for one hour, after which the experimental animals were dissected. The brains of Group B rabbits (b) were more severely damaged than those of Group A rabbits (a). The red arrows indicate the damage.
. real real The positioning error (PE) is evaluated using the following equation: In the present study, the time-difference eigenvalue threshold regularization algorithm reconstructed the distribution of conductivity changes. The change in conductivity can be obtained using the following equation (Liu et al 2014): where s D is the conductivity change, j p is the foreground frame data, j b is the background frame data, S is the sensitivity matrix, -(·) 1 is the inverse of the matrix, and (·) T is the transposition of the matrix. Let = H S S T be the Hessian matrix. In the eigenvalue threshold method, eigenvalue decomposition is first performed on the Hessian matrix, and then the condition number for regularization to obtain a stable reconstruction solution is set (Liu et al 2008, Liu et al 2014, Chen et al 2020. We set the condition number to 5.0 × 10 6 for our reconstruction. During the monitoring imaging process, the background frame data remains unchanged, whereas the foreground frame is the current continuously collected data that changes over time. The red, green, and blue areas in the reconstruction results indicate increased conductivity, constant conductivity, and reduced conductivity, respectively.

MIT analysis
In the MIT reconstruction image, the mean reconstruction value (MRV) of the ROI was calculated as follows (Xu et al 2010): where s ij is the reconstruction value of the ROI and N is the area of the ROI. Considering individual differences, the MRV data of all rabbits were normalized using the mean of the first 10 MRVs. To study the injury effects of different CCI models, this study used the normalized first 300 frames of data (about 1 h) with SPSS software (version 19.0, SPSS Inc., Chicago, USA) to perform statistical analysis on both sets of rabbit data.

Statistical analyses
The data of the largest brain damage area were analyzed using Student's t-test. The average normalized MRVs of the two groups were analyzed using linear regression. Covariance was used to analyze the slope of the regression line between the two sets of data. The values are expressed as average ± standard error of the mean (s.e.m.). Statistical significance was accepted at the p < 0.05 level. Statistical analysis was performed using SPSS software (version 19.0, SPSS Inc., Chicago, USA).

Histological manifestations and grading after trauma
Ten rabbits were randomly divided into two groups: A (n = 5) and B (n = 5). Group A underwent two impacts, and Group B underwent three impacts. All animals were monitored, and their brains were removed after euthanasia. The results of brain tissue anatomy scans revealed visible brain contusions in both groups. Brain injury in group B was more severe than that in group A, and more extensive bleeding was observed in the cerebral cortex (figure 2). The brain tissue was fixed, and the extent of the injury was identified using Nissl staining. Figures 4(a) and (b) show the coronal views of the affected regions of the rabbit's brains. The cortical layers of each brain region were stained. Nissl staining showed that Group B experienced more serious brain injuries and greater extents of brain tissue edema. The average maximum brain injury area in the image analysis was determined using ImageJ (NIH, Bethesda, MD). The mean injured areas in Groups A and B were 5.03 ± 0.09 and 5.87 ± 0.16 mm 2 , respectively. Student's t-test was performed on both models, and the areas of the two groups differed significantly (p = 0.0020, figure 4(c)). The bar chart shows the difference between the two groups of animals regarding the maximum damaged area of the brain slices (mean ± s.e.m., n = 5 for each group; p = 0.0020). ** p < 0.01.

MIT imaging
Equation (2) was used to reconstruct the conductivity changes and represent them in images. The red and blue signals indicate increased and decreased conductivity, respectively. As shown in figure 6(a), after craniocerebral trauma, MIT can clearly locate the area of the trauma, the location of the target area in the reconstructed image is fixed, and the electrical conductivity increases. The red area representing the increase in electrical conductivity appears in the image in a manner consistent with the position of the injury in the rabbit's head ( figure 5(a)). According to equation (1), the PE value of all reconstructed images is lower than the radius of the target object (r = 0.02800 m; figure 5(b)), and the average positioning error of the image is 0.013 72 m. The overall positioning error is small, the position of the target area is relatively fixed, and no obvious changes are observed in other positions. The normalized MRV was used to evaluate the models with different injury levels, and the local change trend of the local area of the brain injury remained nearly the same. As shown in figure 6(a), the local highconductivity area increased. When the rabbit's brain was injured, intracranial hemorrhage and brain contusion appeared, and the normalized MRVs of all ten rabbits increased (figures 6(b), (c)).

Differences in normalized MRV for different injury levels
Normalized MRVs were used to reconstruct the regression of the two models with different injury levels. Figure 7 shows the relationship between the normalized MRV and data frame index and describes the normalized MRVs of the two models and their regression lines. R 2 was used to assess the goodness of fit. The equations of the two sets of regression lines are determined as follows: The regression lines are shown in figure 7. The electrical conductivity of Groups A and B exhibited a visible upward trend during one hour of monitoring, and the rates of electrical conductivity increases in the two groups during the first 20 min were higher than those in the last 40 min. However, during the entire monitoring process, the rate of increase in the electrical conductivity of Group B was less than that of Group A. The two groups differed significantly. The results of the covariance analysis showed that the slopes of the two regression lines were statistically different (p < 0.0001).

Discussion
In this study, the intracranial conductivity significantly increased during the acute phase of TBI. This is similar to the increase in blood conductivity of abdominal subcutaneous injections in rabbits monitored by Chen et al (2020). It is also similar to the use of EIT by Xu et al to monitor the increase in conductivity of intracranial blood injection in a pig model (Xu et al 2010). However, the pathophysiological reaction after TBI in a closed cranial cavity is more complex. When TBI occurs, the blood vessels in the subarachnoid space, pia mater, and cerebral parenchyma are ruptured, and the outflow of blood accumulates in the subarachnoid space, pia mater, and brain parenchyma to form an intracranial hematoma, which may be simultaneously accompanied by subarachnoid hemorrhage. When a TBI occurs, it leads to local brain tissue injury and cell rupture, resulting in intracellular fluid outflow coupled with increased intracranial pressure, insufficient blood perfusion, swelling of unruptured  During one hour of MIT monitoring, the slope of the fitted straight line of the average normalized MRV of the two groups of models at 20 min changed, and the slopes became smaller. The normalized MRV change rate of Group A remained higher than that of Group B throughout the MIT monitoring period, and the difference in the slopes was statistically significant (p < 0.0001).
nerve and glial cells, as well as tissue cells around the injury area, resulting in vasogenic and non-vasogenic brain edema. We know that at 21 MHz, the electrical conductivity values of cerebrospinal fluid, blood, gray matter, and cranial white matter are 2.0096, 1.1435, 0.381 32 and 0.205 83 S m −1 , respectively (Hasgall et al 2012), whereas the conductivity of edema tissue after TBI will be higher than that of normal tissue (Harting et al 2010), and the electrical conductivity of blood will be higher than that of the brain parenchyma and edematous brain tissue; therefore, intracranial conductivity continues to increase.
The animal model used in this study strictly ensures that the dura mater of the animal remains intact during the occurrence of traumatic craniocerebral injury, that is, to ensure that the cerebrospinal fluid is not artificially drained. The aim is the restoration of intracranial physiological and pathological conditions after craniocerebral injury. In the closed cranial cavity, cerebrospinal fluid has the highest intracranial electrical conductivity (Rojas et al 2008) and exists in large amounts in the brain. In a previous study, our team established a three-dimensional craniocerebral simulation model including the scalp, skull, cerebrospinal fluid, brain parenchyma, and ventricles. MIT was used for monitoring. This previous study showed that MIT can image and identify the location of intracranial injury after TBI (Zhang et al 2021). In the present study, the intracranial pressure increased following craniocerebral injury in rabbits, initiating compensatory mechanisms. When the local intracranial pressure increased, the local cerebrospinal fluid and arterial blood perfusion decreased and reflux increased, resulting in a decrease in cerebrospinal fluid in the intracranial local sulcus gyrus. This is combined with an increase in electrical conductivity caused by local cell fragmentation and intracellular fluid overflow, as well as an increase in local electrical conductivity caused by the increase in tissue edema and the formation of intracranial hematoma. In this study, the change in electrical conductivity was divided into two stages. In the first stage, the electrical conductivity changed rapidly within 20 min after the occurrence of TBI, mainly due to the increase in intracranial hematoma, formation of brain edema, and increased cell fragmentation. During the second stage, 20 min after the occurrence of TBI, the change in electrical conductivity decreased significantly, partly due to the decline in the increase in electrical conductivity caused by hematoma, edema, and cell damage, and partly due to a reduction in the local cerebrospinal fluid caused by increased intracranial pressure. This is consistent with the results reported by Yang et al on magnetic induction phase shift showing that the maximum mean phase shift was the turning point of hemorrhage from the acute phase to the chronic phase 20 min after head injury .
The advantage of using the CCI rabbit model to simulate TBI in this experiment is that the mechanical factors, such as the impact time, speed, depth, and number, can be controlled, and this method has high repeatability. We report the first use of multiple impact head injury. The study findings showed that the two degrees of TBI differed significantly and that the head injury in this animal model is closed; that is, the dura mater remains intact, and intracranial hemorrhage occurs.
Because of the circular arrangement and circumference of the MIT sensor array, the forward model was used as the standard circular domain in this study. Compared with EIT, a forward model with an accurate boundary shape must be built, and MIT shows its advantages. In addition, because differential imaging is used in this study, accurate prior information of conductivity distribution needs not to be known. The MRV used in this study, that is, the average reconstruction value of the ROI in the reconstructed image, can only be obtained by selecting the ROI after MIT was used to obtain the reconstructed image.
In this study, we used two or three impacts to simulate different degrees of TBI. Tissue staining revealed that the maximum bleeding areas in the brain tissues of the two models differed significantly. It is noteworthy that the normalized MRVs obtained using MIT monitoring continued to increase, the linear regression coefficients exceeded 0.98, and the slope of the linear regression was significantly different. We observed that MIT monitoring can distinguish between different degrees of TBI. Interestingly, in the two TBI models, the rate of conductivity change in group A was greater than that in group B; however, the degree of brain injury in group A was lower than that in group B. When severe TBI occurs, the local intracranial pressure changes drastically, the local tissue undergoes cell rupture, necrosis occurs, tissue edema becomes more severe, and the compensatory mechanisms of cerebrospinal fluid and blood flow appear earlier (Wykes and Vindlacheruvu, 2015).
Although the intracranial pressure may change after head injury, exhibiting a certain relationship with the change in conductivity, an intracranial pressure monitoring device was not used in this study due to the nature of the conditions. We used a closed head injury model, whereas an intracranial pressure monitor requires the dura to be opened and a probe implanted. This could cause loss of cerebrospinal fluid and affect the mechanism of intracranial pressure compensation after a closed head injury. Although the skull was opened in this model, the dura mater was kept intact, and an airtight cranial state was maintained throughout. Regarding the location of damage, the MIT device used in this study was limited by the size of the hardware system, and the monitoring range was large, but the rabbit brain is very small in comparison. Therefore, the device does not match the animal and cannot accurately reflect the location of brain injury in the tested animals. However, if a real human body was used, it would be better adapted to the size of the MIT hardware system as the brain is larger. In theory, it will achieve better imaging results than those obtained using experimental animals. Although the human spinal fluid layer will be thicker, previous simulation studies performed by our group show that better imaging can be performed even in the presence of cerebrospinal fluid (Zhang et al 2021). The imaging in this study is based on two-dimensional imaging, which mainly reflects the changes in electrical conductivity of the plane where the coil is located. Indeed, full-scale three-dimensional imaging can be more conducive to accurate diagnosis in theory, but because of the extremely poor conditioning of the MIT reconstruction image itself, achieving accurate three-dimensional imaging is difficult.
In addition, this is, to our knowledge, the first preliminary study of MIT for early monitoring of TBI. Only two injury models were used in this study, which can only verify that MIT can identify TBI with different levels of injury. The relationship between the changing trend of MIT-derived MRVs and the degree of brain damage requires further evaluation. It is hoped that the degree of brain damage can be inferred via real-time monitoring of changes in MRVs of MIT and that MIT could be used for long-term continuous monitoring of TBI.

Conclusion
In the present study, MIT was, to our knowledge, used for the first time to perform imaging diagnosis and evaluation in the early stages of TBI. A multiple-impact method was for the first time used to establish TBI models with different levels of injury, and this animal model was a closed TBI model. The study revealed that MIT can accurately locate the craniocerebral injury, reflect the regularity of electrical conductivity changes after craniocerebral injury, distinguish different levels of TBI, and perform imaging. It provides basic evidence for the application of MIT in the early diagnosis of craniocerebral injury, judgment of injury level, and change in injury during long-term monitoring.