Deep Learning-Based Development of Personalized Human Head Model With Non-Uniform Conductivity for Brain Stimulation

Comparison of Transcranial Magnetic Stimulation Dosimetry between Structured and Unstructured Grids Using Different Solvers

In recent years, the interest in transcranial magnetic stimulation (TMS) has surged, necessitating deeper understanding, development, and use of low-frequency (LF) numerical dosimetry for TMS studies. While various ad hoc dosimetric models exist, commercial software tools like SimNIBS v4.0 and Sim4Life v7.2.4 are preferred for their user-friendliness and versatility. SimNIBS utilizes unstructured tetrahedral mesh models, while Sim4Life employs voxel-based models on a structured grid, both evaluating induced electric fields using the finite element method (FEM) with different numerical solvers. Past studies primarily focused on uniform exposures and voxelized models, lacking realism. Our study compares these LF solvers across simplified and realistic anatomical models to assess their accuracy in evaluating induced electric fields. We examined three scenarios: a single-shell sphere, a sphere with an orthogonal slab, and a MRI-derived head model. The comparison revealed small discrepancies in induced electric fields, mainly in regions of low field intensity. Overall, the differences were contained (below 2% for spherical models and below 12% for the head model), showcasing the potential of computational tools in advancing exposure assessment required for TMS protocols in different bio-medical applications.

M2M-InvNet: Human Motor Cortex Mapping From Multi-Muscle Response Using TMS and Generative 3D Convolutional Network

Transcranial magnetic stimulation (TMS) is often applied to the motor cortex to stimulate a collection of motor evoked potentials (MEPs) in groups of peripheral muscles. The causal interface between TMS and MEP is the selective activation of neurons in the motor cortex; moving around the TMS 'spot' over the motor cortex causes different MEP responses. A question of interest is whether a collection of MEP responses can be used to identify the stimulated locations on the cortex, which could potentially be used to then place the TMS coil to produce chosen sets of MEPs. In this work we leverage our previous report on a 3D convolutional neural network (CNN) architecture that predicted MEPs from the induced electric field, to tackle an inverse imaging task in which we start with the MEPs and estimate the stimulated regions on the motor cortex. We present and evaluate five different inverse imaging CNN architectures, both conventional and generative, in terms of several measures of reconstruction accuracy. We found that one architecture, which we propose as M2M-InvNet, consistently achieved the best performance.

Study of multistep Dense U-Net-based automatic segmentation for head MRI scans

BACKGROUND

Despite extensive efforts to obtain accurate segmentation of magnetic resonance imaging (MRI) scans of a head, it remains challenging primarily due to variations in intensity distribution, which depend on the equipment and parameters used.

PURPOSE

The goal of this study is to evaluate the effectiveness of an automatic segmentation method for head MRI scans using a multistep Dense U-Net (MDU-Net) architecture.

METHODS

The MDU-Net-based method comprises two steps. The first step is to segment the scalp, skull, and whole brain from head MRI scans using a convolutional neural network (CNN). In the first step, a hybrid network is used to combine 2.5D Dense U-Net and 3D Dense U-Net structure. This hybrid network acquires logits in three orthogonal planes (axial, coronal, and sagittal) using 2.5D Dense U-Nets and fuses them by averaging. The resultant fused probability map with head MRI scans then serves as the input to a 3D Dense U-Net. In this process, different ratios of active contour loss and focal loss are applied. The second step is to segment the cerebrospinal fluid (CSF), white matter, and gray matter from extracted brain MRI scans using CNNs. In the second step, the histogram of the extracted brain MRI scans is standardized and then a 2.5D Dense U-Net is used to further segment the brain's specific tissues using the focal loss. A dataset of 100 head MRI scans from an OASIS-3 dataset was used for training, internal validation, and testing, with ratios of 80%, 10%, and 10%, respectively. Using the proposed approach, we segmented the head MRI scans into five areas (scalp, skull, CSF, white matter, and gray matter) and evaluated the segmentation results using the Dice similarity coefficient (DSC) score, Hausdorff distance (HD), and the average symmetric surface distance (ASSD) as evaluation metrics. We compared these results with those obtained using the Res-U-Net, Dense U-Net, U-Net++, Swin-Unet, and H-Dense U-Net models.

RESULTS

The MDU-Net model showed DSC values of 0.933, 0.830, 0.833, 0.953, and 0.917 in the scalp, skull, CSF, white matter, and gray matter, respectively. The corresponding HD values were 2.37, 2.89, 2.13, 1.52, and 1.53 mm, respectively. The ASSD values were 0.50, 1.63, 1.28, 0.26, and 0.27 mm, respectively. Comparing these results with other models revealed that the MDU-Net model demonstrated the best performance in terms of the DSC values for the scalp, CSF, white matter, and gray matter. When compared with the H-Dense U-Net model, which showed the highest performance among the other models, the MDU-Net model showed substantial improvements in the HD view, particularly in the gray matter region, with a difference of approximately 9%. In addition, in terms of the ASSD, the MDU-Net model outperformed the H-Dense U-Net model, showing an approximately 7% improvements in the white matter and approximately 9% improvements in the gray matter.

CONCLUSION

Compared with existing models in terms of DSC, HD, and ASSD, the proposed MDU-Net model demonstrated the best performance on average and showed its potential to enhance the accuracy of automatic segmentation for head MRI scans.

High-resolution EEG source localization in personalized segmentation-free head model with multi-dipole fitting

Objective. Electroencephalograms (EEGs) are often used to monitor brain activity. Several source localization methods have been proposed to estimate the location of brain activity corresponding to EEG readings. However, only a few studies evaluated source localization accuracy from measured EEG using personalized head models in a millimeter resolution. In this study, based on a volume conductor analysis of a high-resolution personalized human head model constructed from magnetic resonance images, a finite difference method was used to solve the forward problem and to reconstruct the field distribution.Approach. We used a personalized segmentation-free head model developed using machine learning techniques, in which the abrupt change of electrical conductivity occurred at the tissue interface is suppressed. Using this model, a smooth field distribution was obtained to address the forward problem. Next, multi-dipole fitting was conducted using EEG measurements for each subject (N= 10 male subjects, age: 22.5 ± 0.5), and the source location and electric field distribution were estimated.Main results.For measured somatosensory evoked potential for electrostimulation to the wrist, a multi-dipole model with lead field matrix computed with the volume conductor model was found to be superior than a single dipole model when using personalized segmentation-free models (6/10). The correlation coefficient between measured and estimated scalp potentials was 0.89 for segmentation-free head models and 0.71 for conventional segmented models. The proposed method is straightforward model development and comparable localization difference of the maximum electric field from the target wrist reported using fMR (i.e. 16.4 ± 5.2 mm) in previous study. For comparison, DUNEuro based on sLORETA was (EEG: 17.0 ± 4.0 mm). In addition, somatosensory evoked magnetic fields obtained by Magnetoencephalography was 25.3 ± 8.5 mm using three-layer sphere and sLORETA.Significance. For measured EEG signals, our procedures using personalized head models demonstrated that effective localization of the somatosensory cortex, which is located in a non-shallower cortex region. This method may be potentially applied for imaging brain activity located in other non-shallow regions.

How to assess the accuracy of volume conduction models? A validation study with stereotactic EEG data

Introduction

Volume conduction models of the human head are used in various neuroscience fields, such as for source reconstruction in EEG and MEG, and for modeling the effects of brain stimulation. Numerous studies have quantified the accuracy and sensitivity of volume conduction models by analyzing the effects of the geometrical and electrical features of the head model, the sensor model, the source model, and the numerical method. Most studies are based on simulations as it is hard to obtain sufficiently detailed measurements to compare to models. The recording of stereotactic EEG during electric stimulation mapping provides an opportunity for such empirical validation.

Methods

In the study presented here, we used the potential distribution of volume-conducted artifacts that are due to cortical stimulation to evaluate the accuracy of finite element method (FEM) volume conduction models. We adopted a widely used strategy for numerical comparison, i.e., we fixed the geometrical description of the head model and the mathematical method to perform simulations, and we gradually altered the head models, by increasing the level of detail of the conductivity profile. We compared the simulated potentials at different levels of refinement with the measured potentials in three epilepsy patients.

Results

Our results show that increasing the level of detail of the volume conduction head model only marginally improves the accuracy of the simulated potentials when compared to in-vivo sEEG measurements. The mismatch between measured and simulated potentials is, throughout all patients and models, maximally 40 microvolts (i.e., 10% relative error) in 80% of the stimulation-recording combination pairs and it is modulated by the distance between recording and stimulating electrodes.

Discussion

Our study suggests that commonly used strategies used to validate volume conduction models based solely on simulations might give an overly optimistic idea about volume conduction model accuracy. We recommend more empirical validations to be performed to identify those factors in volume conduction models that have the highest impact on the accuracy of simulated potentials. We share the dataset to allow researchers to further investigate the mismatch between measurements and FEM models and to contribute to improving volume conduction models.

Tensor-conductance model for reducing the computational artifact in target tissue for low-frequency dosimetry

Objective.In protecting human from low-frequency (<100 kHz) exposure, an induced electric field strength is used as a physical quantity for assessment. However, the computational assessment suffers from a staircasing error because of the approximation of curved boundary discretized with cubic voxels. The international guidelines consider an additional reduction factor of 3 when setting the limit of external field strength computed from the permissible induced electric field. Here, a new method was proposed to reduce the staircasing error considering the tensor conductance in human modeling for low-frequency dosimetry.Approach.We proposed a tensor-based conductance model, which was developed on the basis of the filling ratio and the direction of the tissue interface to satisfy the electric field boundary condition and reduce staircasing errors in the target tissue of a voxel human model.Main results.The proposed model was validated using two-layer nonconcentric cylindrical and spherical models with different conductivity contrasts. A comparison of induced electric field strengths with solutions obtained using an analytical formula and finite element method simulation indicated that for a wide range of conductivity ratios, staircasing errors were reduced compared with a conventional scalar-potential finite-difference method. The induced electric field in a simple anatomical head model using our approach was in good agreement with finite element method for exposure to uniform magnetic field exposure and that from coil, simulating transcranial magnetic stimulation.Significance.The proposed tensor-conductance model demonstrated that the staircasing error in an inner target tissue of a voxel human body can be reduced. This finding can be used for the electromagnetic compliance assessment and dose evaluation in electric or magnetic stimulation at low frequencies.

Neural stochastic differential equations network as uncertainty quantification method for EEG source localization

EEG source localization remains a challenging problem given the uncertain conductivity values of the volume conductor models (VCMs). As uncertain conductivities vary across people, they may considerably impact the forward and inverse solutions of the EEG, leading to an increase in localization mistakes and misdiagnoses of brain disorders. Calibration of conductivity values using uncertainty quantification (UQ) techniques is a promising approach to reduce localization errors. The widely-known UQ methods involve Bayesian approaches, which utilize prior conductivity values to derive their posterior inference and estimate their optimal calibration. However, these approaches have two significant drawbacks: solving for posterior inference is intractable, and choosing inappropriate priors may lead to increased localization mistakes. This study used the Neural Stochastic Differential equations Network (SDE-Net), a combination of dynamical systems and deep learning techniques that utilizes the Wiener process to minimize conductivity uncertainties in the VCM and improve the inverse problem. Results revealed that SDE-Net generated a lower localization error rate in the inverse problem compared to Bayesian techniques. Future studies may employ new stochastic dynamical systems-based techniques as a UQ technique to address further uncertainties in the EEG Source Localization problem. Our code can be found here:https://github.com/rrwabina/SDENet-UQ-ESL.

Attention-assisted hybrid 1D CNN-BiLSTM model for predicting electric field induced by transcranial magnetic stimulation coil

Deep learning-based models such as deep neural network (DNN) and convolutional neural network (CNN) have recently been established as state-of-the-art for enumerating electric fields from transcranial magnetic stimulation coil. One of the main challenges related to this electric field enumeration is the prediction time and accuracy. Despite the low computational cost, the performance of the existing prediction models for electric field enumeration is quite inefficient. This study proposes a 1D CNN-based bi-directional long short-term memory (BiLSTM) model with an attention mechanism to predict electric field induced by a transcranial magnetic stimulation coil. The model employs three consecutive 1D CNN layers followed by the BiLSTM layer for extracting deep features. After that, the weights of the deep features are redistributed and integrated by the attention mechanism and a fully connected layer is utilized for the prediction. For the prediction purpose, six input features including coil turns of single wing, coil thickness, coil diameter, distance between two wings, distance between head and coil position, and angle between two wings of coil are mapped with the output of the electric field. The performance evaluation is conducted based on four verification metrics (e.g. R2, MSE, MAE, and RMSE) between the simulated data and predicted data. The results indicate that the proposed model outperforms existing DNN and CNN models in predicting the induced electrical field with R2 = 0.9992, MSE = 0.0005, MAE = 0.0188, and RMSE = 0.0228 in the testing stage.

Closed-loop direct control of seizure focus in a rodent model of temporal lobe epilepsy via localized electric fields applied sequentially

Direct electrical stimulation of the seizure focus can achieve the early termination of epileptic oscillations. However, direct intervention of the hippocampus, the most prevalent seizure focus in temporal lobe epilepsy is thought to be not practicable due to its large size and elongated shape. Here, in a rat model, we report a sequential narrow-field stimulation method for terminating seizures, while focusing stimulus energy at the spatially extensive hippocampal structure. The effects and regional specificity of this method were demonstrated via electrophysiological and biological responses. Our proposed modality demonstrates spatiotemporal preciseness and selectiveness for modulating the pathological target region which may have potential for further investigation as a therapeutic approach.

Measurement and image-based estimation of dielectric properties of biological tissues —past, present, and future—

The dielectric properties of biological tissues are fundamental pararmeters that are essential for electromagnetic modeling of the human body. The primary database of dielectric properties compiled in 1996 on the basis of dielectric measurements at frequencies from 10 Hz to 20 GHz has attracted considerable attention in the research field of human protection from non-ionizing radiation. This review summarizes findings on the dielectric properties of biological tissues at frequencies up to 1 THz since the database was developed. Although the 1996 database covered general (normal) tissues, this review also covers malignant tissues that are of interest in the research field of medical applications. An intercomparison of dielectric properties based on reported data is presented for several tissue types. Dielectric properties derived from image-based estimation techniques developed as a result of recent advances in dielectric measurement are also included. Finally, research essential for future advances in human body modeling is discussed.

A deep learning model to classify and detect brain abnormalities in portable microwave based imaging system

Automated classification and detection of brain abnormalities like a tumor(s) in reconstructed microwave (RMW) brain images are essential for medical application investigation and monitoring disease progression. This paper presents the automatic classification and detection of human brain abnormalities through the deep learning-based YOLOv5 object detection model in a portable microwave head imaging system (MWHI). Initially, four hundred RMW image samples, including non-tumor and tumor(s) in different locations are collected from the implemented MWHI system. The RMW image dimension is 640 × 640 pixels. After that, image pre-processing and augmentation techniques are applied to generate the training dataset, consisting of 4400 images. Later, 80% of images are used to train the models, and 20% are used for testing. Later, from the 80% training dataset, 20% are utilized to validate the models. The detection and classification performances are evaluated by three variations of the YOLOv5 model: YOLOv5s, YOLOv5m, and YOLOv5l. It is investigated that the YOLOv5l model performed better compared to YOLOv5s, YOLOv5m, and state-of-the-art object detection models. The achieved accuracy, precision, sensitivity, specificity, F1-score, mean average precision (mAP), and classification loss are 96.32%, 95.17%, 94.98%, 95.28%, 95.53%, 96.12%, and 0.0130, respectively for the YOLOv5l model. The YOLOv5l model automatically detected tumor(s) accurately with a predicted bounding box including objectness score in RMW images and classified the tumors into benign and malignant classes. So, the YOLOv5l object detection model can be reliable for automatic tumor(s) detection and classification in a portable microwave brain imaging system as a real-time application.

Evaluation of Peripheral Electrostimulation Thresholds in Human Model for Uniform Magnetic Field Exposure

The external field strength according to the international guidelines and standards for human protection are derived to prevent peripheral nerve system pain at frequencies from 300-750 Hz to 1 MHz. In this frequency range, the stimulation is attributable to axon electrostimulation. One limitation in the current international guidelines is the lack of respective stimulation thresholds in the brain and peripheral nervous system from in vivo human measurements over a wide frequency range. This study investigates peripheral stimulation thresholds using a multi-scale computation based on a human anatomical model for uniform exposure. The nerve parameters are first adjusted from the measured data to fit the peripheral nerve in the trunk. From the parameters, the external magnetic field strength to stimulate the nerve was estimated. Here, the conservativeness of protection limits of the international guidelines and standards for peripheral stimulation was confirmed. The results showed a margin factor of 4-6 and 10-24 times between internal and external protection limits of Institute of Electrical and Electronics Engineers standard (IEEE C95.1) and International Commission on Non-Ionizing Radiation Protection guidelines, with the computed pain thresholds.

High-Resolution EEG Source Localization in Segmentation-Free Head Models Based on Finite-Difference Method and Matching Pursuit Algorithm

Electroencephalogram (EEG) is a method to monitor electrophysiological activity on the scalp, which represents the macroscopic activity of the brain. However, it is challenging to identify EEG source regions inside the brain based on data measured by a scalp-attached network of electrodes. The accuracy of EEG source localization significantly depends on the type of head modeling and inverse problem solver. In this study, we adopted different models with a resolution of 0.5 mm to account for thin tissues/fluids, such as the cerebrospinal fluid (CSF) and dura. In particular, a spatially dependent conductivity (segmentation-free) model created using deep learning was developed and used for more realist representation of electrical conductivity. We then adopted a multi-grid-based finite-difference method (FDM) for forward problem analysis and a sparse-based algorithm to solve the inverse problem. This enabled us to perform efficient source localization using high-resolution model with a reasonable computational cost. Results indicated that the abrupt spatial change in conductivity, inherent in conventional segmentation-based head models, may trigger source localization error accumulation. The accurate modeling of the CSF, whose conductivity is the highest in the head, was an important factor affecting localization accuracy. Moreover, computational experiments with different noise levels and electrode setups demonstrate the robustness of the proposed method with segmentation-free head model.

Electrocardiographic Imaging for Atrial Fibrillation: A Perspective From Computer Models and Animal Experiments to Clinical Value

Electrocardiographic imaging (ECGI) is a technique to reconstruct non-invasively the electrical activity on the heart surface from body-surface potential recordings and geometric information of the torso and the heart. ECGI has shown scientific and clinical value when used to characterize and treat both atrial and ventricular arrhythmias. Regarding atrial fibrillation (AF), the characterization of the electrical propagation and the underlying substrate favoring AF is inherently more challenging than for ventricular arrhythmias, due to the progressive and heterogeneous nature of the disease and its manifestation, the small volume and wall thickness of the atria, and the relatively large role of microstructural abnormalities in AF. At the same time, ECGI has the advantage over other mapping technologies of allowing a global characterization of atrial electrical activity at every atrial beat and non-invasively. However, since ECGI is time-consuming and costly and the use of electrical mapping to guide AF ablation is still not fully established, the clinical value of ECGI for AF is still under assessment. Nonetheless, AF is known to be the manifestation of a complex interaction between electrical and structural abnormalities and therefore, true electro-anatomical-structural imaging may elucidate important key factors of AF development, progression, and treatment. Therefore, it is paramount to identify which clinical questions could be successfully addressed by ECGI when it comes to AF characterization and treatment, and which questions may be beyond its technical limitations. In this manuscript we review the questions that researchers have tried to address on the use of ECGI for AF characterization and treatment guidance (for example, localization of AF triggers and sustaining mechanisms), and we discuss the technological requirements and validation. We address experimental and clinical results, limitations, and future challenges for fruitful application of ECGI for AF understanding and management. We pay attention to existing techniques and clinical application, to computer models and (animal or human) experiments, to challenges of methodological and clinical validation. The overall objective of the study is to provide a consensus on valuable directions that ECGI research may take to provide future improvements in AF characterization and treatment guidance.

Influence of segmentation accuracy in structural MR head scans on electric field computation for TMS and tES

In several diagnosis and therapy procedures based on electrostimulation effect, the internal physical quantity related to the stimulation is the induced electric field. To estimate the induced electric field in an individual human model, the segmentation of anatomical imaging, such as magnetic resonance image (MRI) scans, of the corresponding body parts into tissues is required. Then, electrical properties associated with different annotated tissues are assigned to the digital model to generate a volume conductor. However, the segmentation of different tissues is a tedious task with several associated challenges specially with tissues appear in limited regions and/or low-contrast in anatomical images. An open question is how segmentation accuracy of different tissues would influence the distribution of the induced electric field. In this study, we applied parametric segmentation of different tissues to exploit the segmentation of available MRI to generate different quality of head models using deep learning neural network architecture, named ForkNet. Then, the induced electric field are compared to assess the effect of model segmentation variations. Computational results indicate that the influence of segmentation error is tissue-dependent. In brain, sensitivity to segmentation accuracy is relatively high in cerebrospinal fluid (CSF), moderate in gray matter (GM) and low in white matter for transcranial magnetic stimulation (TMS) and transcranial electrical stimulation (tES). A CSF segmentation accuracy reduction of 10% in terms of Dice coefficient (DC) lead to decrease up to 4% in normalized induced electric field in both applications. However, a GM segmentation accuracy reduction of 5.6% DC leads to increase of normalized induced electric field up to 6%. Opposite trend of electric field variation was found between CSF and GM for both TMS and tES. The finding obtained here would be useful to quantify potential uncertainty of computational results.

Development of anatomically accurate brain phantom for experimental validation of stimulation strengths during TMS

Transcranial magnetic stimulation (TMS) is a non-invasive technique for diagnosis and treatment of various neurological conditions. However, the lack of realistic physical models to test the safety and efficacy of stimulation from magnetic fields generated by the coils has hindered the development of new TMS treatment and diagnosis protocols for several neurological conditions. We have developed an anatomically and geometrically accurate brain and head phantom with an adjustable electrical conductivity matching the average conductivity of white matter and grey matter of the human brain and the cerebrospinal fluid. The process of producing the phantom starts with segmenting the MRI images of the brain and then creating shells from the segmented and reconstructed model ready for 3-D printing and serving as a mold for the conductive polymer. Furthermore, we present SEM images and conductivity measurements of the conductive polymer composite as well as confirmation of the anatomical accuracy of the phantom with computed tomography (CT) images. Finally, we show the results of induced voltage measurements obtained from TMS on the brain phantom.