Treatment of a patient using a TTField with electrode positions optimized with a deformable template

Abstract

Embodiments receive an image of a region of a patient's body; identify abnormal tissue in the image; generate a dataset masking the abnormal tissue; warp a model template in space so that features in the warped model template align with corresponding features in the dataset; place data representing the abnormal tissue back into the warped model template; generate an electrical property model of the tissue in the body region based on the warped and modified model template; and determine an electrode placement layout that maximizes field intensity in the abnormal tissue by using the electrical property model to simulate electromagnetic field distribution in the body region caused by simulated electrodes placed at a plurality of different candidate locations corresponding to the body region. This layout can then be used as guidance for placing electrodes corresponding to the body region of the patient to apply a TTField to the body region.

Description

[0001] Cross-reference of related applications

[0002] This application claims the benefit of U.S. Provisional Application 62 / 433,501 (filed December 13, 2016), the entire contents of which are incorporated herein by reference. Background Technology

[0003] The use of electric fields and currents to treat neurological disorders and brain diseases is becoming increasingly common. Examples of such treatments include, but are not limited to, transcranial direct current stimulation (TDCS), transcranial magnetic stimulation (TMS), and tumor therapeutic fields (TTFields). These treatments rely on delivering low-frequency electromagnetic fields to target areas within the brain. See, for example, Woods et al., Clinical Neurophysiology, 1271031-1048 (2016), which reviews the technical aspects of TDCS; and Thielscher et al., Conference Proceedings, Institute of Electrical and Electronics Engineers (IEEE), Institute of Medical and Biological Engineering, 222-225 (2015), which teaches methods for simulating TMS. For example, Miranda et al., *Physics in Medicine and Biology*, 594137-4147 (2014), taught the creation of a computational head model of a healthy individual to simulate the transmission of TTField using a magnetic resonance imaging (MRI) dataset, where model creation was performed in a semi-automatic manner. Furthermore, Wenger et al., *Physics in Medicine and Biology*, 607339-7357 (2015), taught a method for creating a computational head model of a healthy individual to simulate the transmission of TTField, where the model was created from an MRI dataset of healthy individuals.

[0004] In the cases of TDCS and TMS, treatment requires delivering electromagnetic fields to target regions of the brain where they stimulate specific neurons. In the case of TTField, the location of the transducer array on the patient's head is optimized to deliver maximum field strength to the tumor region. See, for example, Wenger et al., *International Journal of Radiation Oncology, Biology & Physics*, 941137-43 (2016), which teaches how to incorporate diffusion tensor imaging (DTI) data into a model used to simulate the delivery of TTField to the head. The DTI data is used to derive the anisotropic conductivity tensor for each voxel in the head model.

[0005] TTField is a low-intensity (e.g., 1 V / cm to 3 V / cm) alternating electric field in the mid-frequency range (100 kHz to 300 kHz) that can be used, for example, to treat tumors, as described in U.S. Patent 7,565,205, the entire contents of which are incorporated herein by reference. TTField therapy is an approved monotherapy for recurrent glioblastoma (GBM) and an approved combination therapy with chemotherapy for newly diagnosed patients. These alternating electric fields are placed directly on the patient's scalp (e.g., using Novocure Optune). TM The transducer array (i.e., capacitively coupled electrode array) of the system is non-invasively induced. TTField also appears to be beneficial for treating tumors in other parts of the body.

[0006] In vivo and in vitro studies have shown that the efficacy of TTField therapy increases with the increase of electric field strength in the target area, the strength of which depends on the placement of the transducer array on the patient's scalp.

[0007] One way to optimize transducer array placement is through computer simulation. As described in this paper, the large amounts of imaging data processed and the simulation / optimization process are computationally intensive and complex, making the use of computers essential. Typically, when performing simulations, an anatomically accurate computational model is built, and electrical properties are assigned to various tissue types. Once the model is established, the simulation model electrodes are positioned on the head model, and appropriate boundary conditions, such as voltages on the electrodes, are applied. The electric field within the head is then calculated. Using various computer-implemented and computationally intensive optimization schemes, the electrode layout and boundary conditions that produce the optimal electromagnetic field distribution within the head (and specifically, the target region) can then be found. However, the anatomical details of individual patients differ, and these variations affect the field distribution in an individual's head. Therefore, in order to use simulations to optimize processes involving the delivery of electromagnetic fields to the target region, until this point, a personalized computational model must be built for each individual.

[0008] The conventional method for constructing a head model is as follows. First, a set of medical images is acquired. Typically, these images include MRI and / or computed tomography (CT) images. Next, the images are segmented to determine which parts correspond to each of the different possible tissue types (e.g., white matter, gray matter, cerebrospinal fluid (CSF), skull, etc.). Next, a series of meshes is constructed for each tissue type in the segmented images and incorporated into the model, with representative conductivity values ​​assigned to each tissue type. Finally, electrodes are positioned on the model, and the field distribution is solved using appropriate numerical techniques (such as the finite element method or finite difference method) (based on the location of each tissue type in 3D space and the conductivity assigned to each of those tissue types).

[0009] Although many steps in the above process are computer-implemented, it still requires significant human intervention because automated algorithms for segmenting medical images of the head, especially those containing tumors, are not robust and often require user intervention to obtain reliable results. See, for example, Menze et al., IEEE Transactions on Medical Imaging, 341993-2024 (2014), who investigated the performance of various algorithms for automated tumor segmentation. Furthermore, mesh regularization is a time-consuming process requiring user supervision, as described, for example, in Miranda et al., Physics in Medicine and Biology, 594137-4147 (2014), Wenger et al., Physics in Medicine and Biology, 607339-7357 (2015), and Wenger et al., International Journal of Radiation Oncology, Biology and Physics, 941137-43 (2016). Specifically, when creating a finite element model of a volume, the volume is meshed into volume elements. To ensure the transformation of numerical solutions, it is desirable that all elements have high quality (the definition of quality varies depending on the type of mesh created). Furthermore, it is important to verify that the elements are disjoint and that the mesh quality is generally sufficient. Regularization is the process of treating the mesh to improve the conditioning of its elements and its overall quality. For a basic discussion, see S. Makarow et al., “Low Frequency Electromagnetic Modelling For Electrical and Biological Systems Using Matlab,” John Wiley & Son, 2010, pp. 36–81.

[0010] Between segmentation and mesh regularization, the time required to create a single model can range from a few hours to several days, depending on the quality of the image and the complexity of the model being created. Summary of the Invention

[0011] One aspect of the present invention relates to a first method for improving tumor treatment using a Tumor Therapeutic Field (TTField). The first method includes receiving a three-dimensional image of a patient's body region by a processor of a computer system; identifying image portions corresponding to anomalous tissue; and generating a dataset corresponding to the image, wherein the anomalous tissue is masked. The first method further includes retrieving a model template from a memory device of the computer system, the model template including a tissue probability map specifying the locations of multiple tissue types in a healthy version of the patient's body region; and deforming the model template in space such that features in the deformed model template are aligned with corresponding features in the dataset. The first method further includes modifying portions of the deformed model template corresponding to the masked portions of the dataset such that the modified portions represent anomalous tissue; and generating an electrical property model of the tissue in the body region based on (a) the locations of the multiple tissue types in the deformed and modified model template and (b) the locations of the anomalous tissue in the deformed and modified model template. The first method further includes determining an electrode placement layout that maximizes the field strength in at least a portion of the anomalous tissue using the electrical property model to simulate the electromagnetic field distribution in the body region caused by simulated electrodes placed at multiple different candidate sets of locations corresponding to the body region, and selecting one of the sets. The first method also includes placing electrodes corresponding to the patient's body area based on a determined electrode placement layout; and applying the TTField to the body area using the placed electrodes.

[0012] Another aspect of the invention relates to a second method for improving electrotherapy. The second method includes receiving a three-dimensional image of a patient's body region by a processor of a computer system; identifying image portions corresponding to abnormal tissue; and generating a dataset corresponding to the image, wherein the abnormal tissue is masked. The second method further includes retrieving a model template from a memory device of the computer system, the model template including a tissue probability map specifying the locations of multiple tissue types in a healthy version of the patient's body region; and deforming the model template in space such that features in the deformed model template are aligned with corresponding features in the dataset. The second method further includes modifying portions of the deformed model template corresponding to the masked portions of the dataset such that the modified portions represent abnormal tissue; and generating an electrical property model of the tissue in the body region based on (a) the locations of the multiple tissue types in the deformed and modified model template and (b) the locations of the abnormal tissue in the deformed and modified model template. The second method further includes determining an electrode placement layout that maximizes the field strength in at least a portion of the abnormal tissue using the electrical property model to simulate the electromagnetic field distribution in the body region caused by simulated electrodes placed at multiple different candidate sets of locations corresponding to the body region, and selecting one of the sets. The second method also includes outputting a defined electrode placement layout for subsequent use as a guide to place the electrodes in the corresponding areas of the patient's body before using the electrodes for electrotherapy treatment.

[0013] In some embodiments of the second method, the variation of the model template includes determining a mapping that maps the dataset to the coordinate space of the model template; and applying the inverse of the mapping to the model template. Optionally, in these embodiments, the mapping is determined for points in the dataset that fall outside the masked portion. Optionally, in these embodiments, the model template includes an organization probability map, wherein the mapping maps the dataset to the organization probability map.

[0014] Optionally, in these embodiments, the tissue probability map is derived from images of healthy individuals from which a model template has already been derived. Optionally, in these embodiments, the tissue probability map is derived by simultaneously registering and segmenting images of healthy individuals using an existing tissue probability map, wherein the existing tissue probability map is derived from images of multiple individuals.

[0015] Optionally, in these embodiments, the tissue probability map is an existing tissue probability map derived from images of multiple individuals.

[0016] Optionally, in these embodiments, the inversion of the mapping is applied to each of the tissue probability maps, wherein the inverse-mapped tissue probability maps are combined into a segmented image including a deformable model template. Optionally, in these embodiments, combining the inverse-mapped tissue probability maps includes assigning each voxel the tissue type with the highest probability of occupying that voxel on the inverse-mapped tissue probability map. Optionally, in these embodiments, combining the inverse-mapped tissue probability maps includes using a lookup table to assign tissue types to each voxel that has been assigned more than one tissue type on the inverse-mapped tissue probability map.

[0017] In some embodiments of the second method, the identification of image portions corresponding to abnormal tissue includes performing image segmentation. In some embodiments of the second method, the electrical property model of the tissue includes a model of conductivity or resistivity. In some embodiments of the second method, the images include MRI images, CT images, or a combination of MRI and CT images. In some embodiments of the second method, the body region includes the patient's head. In some embodiments of the second method, the image portion corresponding to abnormal tissue corresponds to a tumor. In some embodiments of the second method, the electrotherapy treatment includes TTField.

[0018] In some embodiments of the second method, determining the electrode placement layout includes applying boundary conditions to the simulated electrodes in each of at least two electrode placement layouts to resolve the field distribution in the volume region for each of the at least two electrode placement layouts, and selecting the electrode placement layout that produces the strongest field in the anomalous region. Optionally, in these embodiments, the boundary conditions correspond to the voltage or current applied to the simulated electrodes.

[0019] In some embodiments of the second method, a model template is selected from multiple model templates based on the similarity between the image and each of the model templates.

[0020] Some embodiments of the second method also include placing electrodes corresponding to the patient's body area based on a determined electrode placement layout, and applying the TTField to the body area using the electrodes.

[0021] Another aspect of the invention relates to an electrotherapy device comprising a processor configured to execute instructions stored in one or more memory devices to perform electrotherapy. In these embodiments, the treatment includes receiving a three-dimensional image of a patient's body region by the processor; identifying portions of the image corresponding to anomalous tissue; and generating a dataset corresponding to the image, wherein the anomalous tissue is masked. The treatment also includes retrieving a model template from one or more memory devices, wherein the model template specifies the locations of multiple tissue types in a healthy version of the patient's body region; deforming the model template in space such that features in the deformed model template are aligned with corresponding features in the dataset; and modifying portions of the deformed model template corresponding to the masked portions of the dataset such that the modified portions represent anomalous tissue. The treatment also includes generating an electrical property model of the tissue in the body region based on (a) the locations of the multiple tissue types in the deformed and modified model template and (b) the locations of the anomalous tissue in the deformed and modified model template. The treatment further includes determining an electrode placement layout that maximizes the field strength in at least a portion of the anomalous tissue using the electrical property model to simulate the electromagnetic field distribution in the body region caused by simulated electrodes placed at multiple different candidate sets of locations corresponding to the body region, and selecting one of the sets. The treatment also includes outputting a defined electrode placement layout for subsequent use as a guide to place the electrodes in the appropriate areas of the patient's body before using them for electrotherapy. Attached Figure Description

[0022] Figure 1 This is a flowchart of an embodiment of performing electrotherapy by creating a realistic head model of a patient using a deformable template.

[0023] Figure 2 The original MRI images obtained from patients with abnormalities (e.g., tumors) are depicted.

[0024] Figure 3 Depicting Figure 2 The MRI images show that the abnormalities are masked.

[0025] Figure 4 Describes generation Figure 3 The standardization / registration process of mapping and inverse mapping between a model deformable template and a healthy individual.

[0026] Figure 5 Depicting Figure 4 How the deformable template can deform to match the shape of the patient's MRI image.

[0027] Figure 6 It describes how to implant anomalies back into the deformable model.

[0028] Figure 7 A system for electrotherapy is described according to one embodiment.

[0029] Figure 8 This is another flowchart of an embodiment of performing electrotherapy by creating a realistic head model of the patient using a deformable template. Detailed Implementation

[0030] The embodiments described herein generate customized realistic head models for each individual patient by applying non-rigid deformation to a pre-existing realistic head model template, thereby reducing the time and manpower required to create head models. After generating a customized head model for each patient, conventional simulation methods are used to determine the optimal position of the transducer on the patient's body. Optionally, the pre-existing realistic head model template for healthy patients may include a tissue probability map (TPM). A TPM provides a model in which each point is represented by a corresponding probability of belonging to various tissue types, such as white matter, gray matter, CSF, etc.

[0031] Optionally, patient images may be supplemented with other MRI data, such as diffusion tensor imaging (DTI) data or water content electrical impedance tomography (Wept) data, to obtain a more accurate representation of conductivity in the patient's head, for example, as disclosed by E. Michel, D. Hernandez, and SYLee in "Electrical conductivity and permittivity maps of brain tissues derived from water content based on T1-weighted acquisition," Magnetic Resonance in Medicine, 2016. MRI imaging techniques such as DTI or Wept are known to provide information about tissue conductivity, for example, as disclosed in U.S. Application No. 15 / 336,660, the entire contents of which are incorporated herein by reference.

[0032] Figure 1 and Figure 8The embodiments described herein depict a workflow for creating personalized, realistic head models for each patient with reduced user intervention, and using these head models to optimize the patient's Tumor Treatment Field (TTField) array layout. Once a realistic model has been built for any given patient, optimization can be performed in a fully or semi-automatic manner using a series of algorithms also described herein. While these workflows are described in the TTField scenario, they can also be used in alternative scenarios.

[0033] Figure 1 and Figure 8 The embodiments begin with a deformable template, which is a realistic head model of a healthy individual (as opposed to a realistic head model of an actual patient). This head model can be obtained using any conventional method. For example, a realistic head model can be created in a standard coordinate system such as the Montreal Neurological Institute (MNI) or Talairach space. For example, Holmes et al., *Journal of Computer-Aided Computed Tomography*, 22324-333 (1998), which is incorporated herein by reference, teach the mapping and averaging of MRI images in the standard space of the MNI. If the model does not exist in the desired standard coordinate space, the transformation from the standard coordinate space to the head model is preferably known and can be used to map the model to the standard coordinate space. An example of a real head model built in standard coordinate space is based on the COLIN27 dataset created by Miranda et al. (as described in Holmes et al., *Journal of Computer-Aided Tomography*, 22324-333 (1998)) (as described in Miranda et al., *Physics in Medicine and Biology*, 594137-4147 (2014), which is incorporated herein by reference). However, various alternative real head models for healthy individuals can be used instead of the Miranda model. MRI scans from which models are intended to be created can also be used for the purposes described below.

[0034] In some embodiments, the real head model template of a healthy individual provides a TPM for each tissue type. That is, each point in the model is represented by a probability corresponding to a point belonging to various tissue types, such as white matter, gray matter, CSF, etc. In some embodiments, the real head model template of a healthy individual provides a TPM for each tissue type (e.g., six TPMs for the six tissue types: white matter, gray matter, skull, scalp, CSF, and air).

[0035] Figure 1 The process of creating a realistic head model for any given patient by using an existing head model as a deformable template and a real head model of a healthy individual is described 100.

[0036] Process 100 begins with step S1, which involves acquiring an appropriate set of MRI images. In step S1, an MRI dataset of the individual patient is acquired using any conventional method. This dataset preferably includes MRIs carrying structural data, such as structural data obtained from T1 or T2 MRI sequences. Optionally, additional sequences, such as DTI or perfusion imaging, may also be acquired, which may carry additional information that can be used for model creation, as described below. In some cases, the parameters of the MRI sequences are optimized to increase contrast between specific tissue types. Enhancing contrast is useful for image segmentation followed in the steps described below, for example, sequences as described in Windhoff et al., Mapping the Human Brain, 34923-935 (2013), which is incorporated herein by reference.

[0037] Preferably, the MRI is acquired at the highest possible resolution. Typically, a resolution better than 1 mm × 1 mm × 1 mm is desired. However, lower resolution images may also be used.

[0038] Optionally, DTI or diffusion-weighted magnetic resonance imaging (DWI) data may also be acquired. This data can be used to map conductivity (or conductivity tensor) within each voxel, as described in Wenger et al., *International Journal of Radiation Oncology Biology & Physics*, 941137-43 (2016), and Basser et al., *Journal of Biophysics*, 66259-267 (1994), which are incorporated herein by reference. In alternative embodiments, different imaging modalities may be used instead of MRI images, such as CT images, combinations of MRI and CT images, etc.

[0039] Process 100 continues in step S2, which is image preprocessing. However, in some cases, preprocessing is not required, and step S2 can be skipped. In step S2, image preprocessing is performed on the data obtained in step S1 to obtain a cleaner image. Figure 2An example of an MRI image 200 obtained after image preprocessing in step S2 is shown. Preprocessing can be implemented using any conventional method. In some embodiments, the image preprocessing step includes image alignment and distortion correction. For example, image alignment can be performed to remove artifacts caused by motion from the image using any conventional method. Any suitable conventional method, such as Statistical Parametric Mapping (SPM) implemented in the SPM 8.0 toolbox, can be used to perform realignment using affine registration. This toolbox was developed to construct and evaluate spatial extension statistical processes for testing hypotheses about functional imaging data. Additionally, image distortion (e.g., caused by induced eddy currents) can be corrected at this stage. When a model is created using more than one dataset, image realignment is required; in this case, the multiple datasets need to be aligned. For example, when axial and coronal image sets are used for super-resolution, they need to be aligned. As another example, when DTI data is used in addition to T1 data, alignment of the DTI and T1 data may be required.

[0040] In some embodiments, an additional preprocessing step (e.g., in a Neuroimaging Information Technology Initiative (NifTI) format) is performed to manipulate the head of the MRI image, such that the origin of the file is matched with the origin of the template TPM. The origin of the file refers to the origin of the axis in the file. This step helps to register the MRI image into deformable space, as described in step S4 below. In some embodiments, the origins of the axes in the patient's MRI image and in the file associated with the deformable template are located at similar voxels to facilitate the execution of step S4.

[0041] Optionally, super-resolution algorithms can be used to combine several MRI datasets of a single patient into a single image. These algorithms can be used to create a dataset displaying the patient's complete head when all other datasets truncate the head at different points, or to create images with high resolution (or slice spacing) when the original data has low resolution. High-resolution datasets and datasets displaying the complete 3D head are very useful for creating accurate head models. Woo et al. describe an example of a super-resolution algorithm. "Reconstruction of high-resolution tongue volumes from MRI," IEEE Transactions on Medical Imaging, Vol. 59, No. 12 (2012). This algorithm employs several preprocessing steps, including motion correction and intensity normalization, followed by a region-based maximum a posteriori (MAP) Markov random field (MRF) method to combine three orthogonal image volumes from the MRI dataset into a single super-resolution isotropic volume reconstruction of the tongue. In terms of signal-to-noise ratio (SNR) and resolution, the output super-resolution image outperforms the input image.

[0042] In many cases, background noise and aliasing can be present and can degrade the quality of head models created using deformable templates. In particular, when background noise is present, the skull contours obtained during model creation are often inaccurate and include portions of the background. Therefore, some embodiments implement various thresholding schemes known to those skilled in the art to remove background noise and aliasing. The aliasing mentioned herein refers to artifacts in MRI images that cause weak “shadows” of the imaged object to appear in the background (i.e., shadows caused by aliasing). These shadows are often inverted and directly attached to the main image. In this case, a thresholding scheme can be used to remove the weak shadows in the background. An example of a thresholding scheme that can be used to enhance image quality is a semi-automatic method where the user selects a single value representing background noise, and the software applies that value as a threshold to automatically detect the contours of the scalp and slice by slice to zero the background noise intensity. As those skilled in the art will understand, a variety of alternative methods are also available.

[0043] Alternatively or additionally, scanner-specific preprocessing may be applied. For example, images may be converted from Medical Digital Imaging and Communications (DICOM) format to NifTI.

[0044] Process 100 continues in step S3, which involves masking abnormal areas in the head. Step S3 is performed only if a tumor or other abnormality (e.g., skull defect / cranial flap) is present in the patient's MRI images. In step S3, as... Figure 3 As shown in image 300, these abnormal areas are masked. Optionally, if desired, the masked areas may extend beyond the tumor / abnormality to include all areas of the brain whose normal structure is severely disturbed due to the presence of a tumor or other defect.

[0045] One way to achieve this masking step is to use supervised segmentation to correctly label the anomalous head regions. During this step of supervised segmentation, multiple types of anomalous structures are labeled to achieve the desired level of detail in the final model, as described below. Supervised segmentation can be performed semi-automatically using tools such as ITK-SNAP (see, for example, Yushkevich et al., Neuroimaging, 311116-1128 (2006), which is incorporated herein by reference).

[0046] Alternatively, automated segmentation algorithms can be used to perform masking. For example, Porz et al., “Multi-modal glioblastoma segmentation: man versus machine,” PLOS ONE, 9.5 (2014), taught a method for automatically segmenting preoperative MRI images. In some cases, manual correction of the results of the automated segmentation process may be necessary to ensure accurate tumor masking.

[0047] In some embodiments, the area to be masked is determined manually. One way to achieve this is to present MRI data to a user and ask the user to draw the outline of the tumor on the data. The data presented to the user may include structural MRI data (e.g., T1, T2 data). Different MRI modalities can be registered with each other, and the user may be presented with the option to view any of the datasets and describe the tumor. The user may be asked to describe the outline of the tumor on a 3D volumetric representation of the MRI data, or the user may be given the option to view individual 2D slices of the data and mark the tumor boundaries on each slice. Once the boundaries are marked on each slice, the tumor within the anatomical volume can be located. In this case, the volume marked by the user corresponds to the tumor. In some cases, a predefined width (e.g., 20 mm) of edge is added to the tumor, and the resulting volume is used as the area to be masked.

[0048] Note that step S3 is omitted when the patient's MRI images do not show any tumors or other abnormalities (e.g., when the patient is healthy).

[0049] For some patients, the segmentation results will reveal that the tumor is non-uniform. In such cases, the tumor can also be segmented into several sub-regions so that this segmentation information can be used to more accurately implant the tumor back into the real head model after the deformation step, as will be described in further detail in this paper. Examples of such sub-regions are active / enhanced tumors, necrotic areas, resection cavities, etc. Conventional automatic segmentation algorithms can be used for detailed GBM segmentation. One example of a publicly available algorithm is the recent brain tumor image analysis software (BraTumlA), which distinguishes between necrotic cores, edema, non-enhanced tumors, and enhanced tumors, while requiring four different imaging modalities (T1, T1-contrast, T2-contrast, and FLAIR). Techniques also exist that only require T1 as input. However, regardless of any changes within the tumor, all regions of the tumor are masked outside the original patient image. If skull defects are present in the image, these regions are also segmented and masked.

[0050] Note that while various methods for identifying anomalous regions in images have been described above, a wide range of alternative methods will be readily apparent to those skilled in the art.

[0051] Process 100 continues in step S4, which is spatial normalization / registration. In step S4, a mapping is identified that warps the current set of MRI images of a given patient into a standard space of the template model. Figure 4 A normalization / registration process 400 is described, which generates a mapping and inverse mapping between a patient MRI image 402 (with masked anomalies) and a deformable model template 404 of a healthy individual. The inversion of this mapping is also identified (used in step S5 below to map from the standard space to the space of the patient MRI group).

[0052] For example, one method for generating this mapping is to register the patient's MRI images to a standard coordinate space, such as MNI space or Talairach space. Image registration refers to the spatial transformation of an image so that certain features of the image are aligned with corresponding features in another image / space. This can be accomplished by any known method that is obvious to those skilled in the art, for example, by using readily available software packages, including but not limited to FSL, FLIRT, and SPM.

[0053] It is worth noting that the anomalous regions masked in step S3 are omitted during the registration process. Ignoring the masked regions during registration ensures that only healthy regions of the head are used for registration, which can be effectively mapped to the TPM model, which describes the probability that a particular voxel in the standard space belongs to a particular tissue type. Advantageously, omitting anomalous regions improves the robustness of the registration process. In some embodiments, the TPM is constructed in the model template space.

[0054] Alternatively, non-rigid registration algorithms (e.g., described in Zhuang et al., IEEE Transactions on Medical Imaging, 300278-0062 (2011), which is incorporated herein by reference and teaches algorithms for image registration using mutual information) can be used to register patient MRI images to a standard coordinate space (e.g., a real-world model template of a healthy individual) or a voxelized version of the corresponding segmentation model template. Note that various algorithms for mapping patient MRI images to a standard space are well known to those skilled in the art. Moving in the opposite direction (i.e., from standard space to the patient MRI image, as described below) will use the inversion of those same mappings.

[0055] For points in the patient's head that fall outside the masked region, the aforementioned mapping is found. For example, the transformation in the masked region(s) prior to registration can be estimated by interpolating a deformable map found in the rest of the head into these regions, or by using any of a variety of alternative methods that are obvious to those skilled in the art. In some embodiments, it may not be necessary to find the transformation in the masked region(s) prior to registration. This is because the areas of the deformable model template corresponding to the masked regions contain information related to certain "natural" structures (e.g., healthy tissue). Therefore, after applying the aforementioned mapping to the deformable model template for points falling outside the masked region, the deformable model template already includes some model data in these regions because the "natural" structure is preserved in these regions. For example, if a sphere is masked from the left hemisphere in a patient image, and the mapping is applied only to the deformable model template for points falling outside the sphere, the spherical content of the left hemisphere of the deformable model template will resemble certain natural structures.

[0056] In some embodiments, the model TPM is used to find a mapping from a standard space to a patient space. In some embodiments, the model TPM may be derived from an MRI dataset from which deformable templates are obtained. Using a TPM derived from this MRI dataset provides a more accurate representation of the patient in the final model compared to using other TPMs. The reason is as follows. The TPM describes the probability of voxels belonging to each tissue type in a standard space. Typically, a TPM is derived from multiple MRIs of different subjects. Therefore, the TPM represents the probability of voxels belonging to each tissue type across the entire population of individuals. This means that when registration is performed using TPMs derived from multiple individuals, the output mapping represents a mapping to some representative space that, by definition, smooths out anatomical variations between individuals from which the TPM was derived. However, when creating a patient model by deforming a head model of a healthy individual, it may be desirable that the mapping computed when registering the patient's MRI onto the TPM captures the anatomical features of the healthy head model as accurately as possible. This accuracy ensures that when the deformable template is later deformed into the patient space in step S5 below, the resulting model is as accurately similar to the patient as possible. Therefore, the TPM on which the registration in step S4 is expected to be performed represents the individual from which the healthy head model is derived, rather than the group of individuals from which the TPM is usually derived.

[0057] One approach to creating a TPM representing a healthy individual from which a deformable model template is derived is to simultaneously register and segment MRI images of the healthy individual using an existing set of generic TPMs (e.g., TPMs constructed in standard space using data from multiple individuals). An example algorithm for achieving this is Ashburner and Friston's Unified Segmentation algorithm (“Unified Segmentation,” Neuroimaging, 26.3, 2005), implemented in the aforementioned SPM 8.0 toolkit. The output of this process includes probability maps that describe the probability that a voxel (registered to the standard space MRI image) belongs to a specific tissue type. The number of probability maps generated in this process equals the number of tissue types in the model (typically 6), and each voxel in the map is assigned a value from 0 to 1, indicating the probability that the voxel belongs to a specific tissue type. By definition, these probability maps are TPMs representing a healthy individual from which a healthy head model (deformable template) is derived.

[0058] In some cases, the TPM is manually corrected to obtain a better representation of the deformable template. For example, the probability maps of the skull and scalp can be modified to enhance the boundaries of the skull or scalp. This can be done, for example, by manually assigning probability values ​​to specific voxels such that the probability of a voxel belonging to one tissue type is close to 1, and its probability of belonging to other tissue types is close to 0. The final step in creating the TPM from these probability maps is to apply a smoothing filter to the individual map. Smoothing is important for allowing adjustment of MRI of any individual. Smoothing can be performed, for example, using a Gaussian filter with a smoothing kernel of 4 mm × 4 mm × 4 mm FWHM (full width at half maximum).

[0059] Process 100 continues in step S5, which involves deforming / warping the template to the desired space. In step S5, the inverse mapping found in step S4 is applied to the deformable model template to map the deformable model template into the coordinate system of the patient's MRI image. Figure 5 A deformation / warping process 500 is depicted, which applies an inverse mapping to a deformable model template 502 to obtain a deformable model 504. In some embodiments, the inverse mapping applies a three-dimensional transformation to the deformable model template 502, thereby warping the deformable model template 502 to conform to the patient's specific anatomical properties.

[0060] It should be noted that before warping, model template 502 is a model of the brain of a healthy reference individual; after warping, the warped model 504 will represent an approximation of what the patient's brain would look like in a healthy state. In other words, this step results in a model of a healthy individual being warped to fit the head shown in the patient's MRI images, but without tumors. It is worth noting that although this warped model comes from a model template (rather than from the head of each individual patient), it can still be used to analyze the electric fields that can be induced within the head of each individual patient.

[0061] The deformation in step S5 can be applied to either a voxelized version or a meshed version of the model. In the voxelized version, each voxel indicates the tissue type (or tissue type probability) at the coordinate location of that voxel. In the meshed version, each mesh defines the boundaries between different tissue types, and the deformation is applied to these meshes in the deformable model template. In some embodiments, binary images of each tissue type are created, and each resulting binary image is deformed separately.

[0062] Optionally, any holes appearing in the deformed image of a tissue type can be assigned to one of the tissue types appearing in the image. An example of a procedure designed to assign tissue types to holes appearing between binary masks can be found in Timmons et al., “End-to-end workflow for finite element analysis of tumor treating fields inglioblastomas,” Physics in Medicine and Biology, 62.21 (2017), where the software ScanIP, with its Gaussian filter function, smooths the boundaries between masks to avoid convergence problems. Cavities in the masks are filled, and islands above a threshold (which can vary with tissue type) are removed. The current mask is copied and then expanded (by one to three voxels, depending on the tissue mask) and its Boolean value is added to the next mask on all slices. Any of a variety of alternative methods for filling holes appearing in the deformed image can also be used.

[0063] After forming an image for each individual tissue type, all binary images are combined into a single image representing a segmented image of the deformed head model.

[0064] When voxels in a composite model are assigned to more than one tissue type, heuristic logic can be used to determine the tissue type in the final image. For example, the logic could specify that all voxels in a composite model where gray and white matter overlap are assigned only to white matter, and vice versa.

[0065] In embodiments where the model template includes a TPM (i.e., each tissue in the model template is represented by a 3D matrix describing the probability that each voxel belongs to a specific tissue type), the TPM is deformed, and the deformed TPMs are combined into a final model such that each voxel in the combined model is assigned a tissue type based on some heuristic logic. For example, each voxel is assigned to the tissue type with the highest probability of occupying that voxel.

[0066] In some embodiments, the probabilities assigned to each voxel by different TPMs are used to determine the combination of conductive properties in the created voxelized model. In other words, it is assumed that voxels do not necessarily contain a certain tissue type, and the final conductivity is assigned to voxels as a weighted sum of the conductivity of all tissue types, where the weights come from the probability values ​​assigned to each tissue type in that voxel.

[0067] In some embodiments, conductivity values ​​are assigned to tissue maps by additionally incorporating information obtained from MRI imaging techniques such as DTI or Wept, which are known to provide information about tissue conductivity, for example, as disclosed in U.S. Application No. 15 / 336,660 (published as US2017 / 0120041), the entire contents of which are incorporated herein by reference. This information can be incorporated into the model, for example, by assigning conductivity to each voxel based on a weighted average of the conductivity derived from the model and the conductivity derived from Wept / DTI.

[0068] Process 100 continues in step S6, which implants the abnormality back into the deformable template. In step S6, the deformable template is edited such that each voxel of the template corresponding to the masking region found in step S3 is assigned to the abnormal tissue type (e.g., tumor or surrounding area). Figure 6 The process 600 is depicted, in which abnormalities identified in patient image 602 are implanted into deformed model template 604. In some embodiments, implantation is performed by assigning a tissue type to each abnormal region based on the segmentation performed in step S3. More specifically, the tissue type assigned to each point in the deformed abnormal region is based on the tissue type identified for the corresponding point in the segmentation in step S3 prior to deformation. Therefore, if the segmentation in step S3 identifies more than one tissue type in the abnormal region, more than one tissue type may be assigned to the abnormal region after deformation. In an alternative embodiment, implantation can be performed by assigning a default abnormal tissue type to the abnormal region after deformation. In other alternative embodiments, implantation can be performed by allowing the user to manually assign tissue types to points in the abnormal region.

[0069] Process 100 continues in step S7, which is for model creation. In the modeling step (S7), electrical properties such as conductivity and dielectric constant are assigned to various tissue types. Note that tissue types are typically obtained from a deformable template. However, the tissue type corresponding to tumor tissue will be assigned to each voxel corresponding to the implantation abnormality. The model of the electrodes (or transducer array) is placed on the model skin, and appropriate boundary conditions are applied. In some embodiments, modeling step S7 assumes that each tissue type is homogeneous, and therefore assigns a single value of the electrical property to each tissue type (e.g., as described by Miranda et al. in Physics in Medicine and Biology, 594137-4147 (2014), Wenger et al. in Physics in Medicine and Biology, 607339-7357 (2015), and Wenger et al. in the International Journal of Radiation Oncology, Biology and Physics, 941137-43 (2016). In other models, the conductivity in each voxel is specified by a DTI or DWI image acquired during the image acquisition step. DTI assigns anisotropic electrical properties (3×3 tensors) to each voxel, while DWI assigns isotropic conductivity (scalars) to each voxel. Finally, the model is divided into volumetric elements, for example, by voxelization or alternatively by volumetric meshing.

[0070] Process 100 continues in step S8. After the head model is created and the model electrodes have been added to the head model, a simulation is run in step S8. This simulation finds the optimal electrode array layout by solving for the corresponding induced electric field using appropriate numerical techniques (including but not limited to the finite element method or the finite difference method).

[0071] Optimizing the electrode array layout means finding an array layout that optimizes the electric field within the affected area (tumor) of the patient's brain. This optimization can be achieved on the target volume (the treatment target volume) within a realistic head model through the following steps: automatically placing the transducer array on the realistic head model and setting boundary conditions; once the array has been placed on the realistic head model and the boundary conditions applied, calculating the electric field generated within the realistic head model; and running the optimization algorithm to find the layout that produces the optimal electric field distribution within the target volume. Although various alternative methods can be used, an example for implementing these four steps is provided below.

[0072] It can automatically calculate the position and orientation of an array on a real head model for a given iteration. Used in Optune. TMEach transducer array delivering the TTField in the device comprises a set of ceramic disc electrodes coupled to the patient's head via a layer of medical gel. When the array is placed on a real patient, the discs naturally align parallel to the skin, and good electrical contact occurs between the array and the skin as the medical gel deforms to match the body's contours. However, the virtual model consists of a strictly defined geometry. Therefore, placing the array on the model requires an accurate method for finding the orientation and contour of the model surface at the desired placement location, as well as for determining the gel thickness / geometry required to ensure good contact between the model array and the real patient model. These calculations must be performed automatically to achieve fully automated optimization of the field distribution.

[0073] Various algorithms can be used to perform this task. The steps of one such algorithm designed for this purpose are described below.

[0074] a. Define the location where the transducer array center point will be placed on the head of the model. This location can be defined by the user or as one of the steps in a field optimization algorithm.

[0075] b. Using the input from step (a) and knowledge about the disk geometry and how the disks are arranged in the array, calculate the approximate location of the center of all disks in the transducer array within the model.

[0076] c. Calculate the orientation of the real model surface at the location where the disk will be placed. This is performed by finding all points on the phantom skin within a disk radius of a specified center. The coordinates of these points are arranged in columns of a matrix, and singular value decomposition is performed on the matrix. The normal to the model skin is then the eigenvector corresponding to the found minimum eigenvalue.

[0077] d. For each disk in the transducer array: Calculate the thickness of the medical gel required to ensure good contact between the disk and the patient's body. This is accomplished by finding parameters for a cylinder whose height is parallel to the skin surface normal. The cylinder is defined with a radius equal to the radius of the disk, and its height is set to exceed a predetermined amount (a predetermined constant) beyond the point on the skin used to find the normal. This results in the cylinder extending from the phantom surface by at least a predetermined amount.

[0078] e. On the model, create the cylinder described in (d).

[0079] f. Remove the cylindrical region protruding into the patient's real-world model from the model using binary logic operations (e.g., subtracting the head from the cylinder). The resulting "truncated cylinder" represents the medical gel associated with the transducer array.

[0080] g. A ceramic disk representing a transducer array is placed on the outside of the truncated cylinder.

[0081] Then, the electric field distribution is calculated within the head model for a given iteration. Once the head model is built and the transducer array (i.e., electrode array) used to apply the field is placed on the real head model, a volumetric mesh suitable for finite element analysis can be created. Next, boundary conditions can be applied to the model. Examples of boundary conditions that can be used include Dirichlet boundary conditions (constant voltage) on the transducer array, Neumann boundary conditions (constant current) on the transducer array, or floating potential boundary conditions that set the potential at the boundary such that the integral of the normal component of the current density equals the specified amplitude. The model can then be solved using a suitable finite element solver (e.g., a low-frequency quasi-static electromagnetic solver) or alternatively, a finite difference algorithm. Existing software packages such as Sim4Life, Comsol Multiphysics, Ansys, or Matlab can be used to perform mesh generation, boundary condition application, and model solving. Alternatively, custom computer code implementing the finite element (or finite difference) algorithm can be written. This code can leverage existing software resources such as C-Gal (for mesh creation) or FREEFEM++ (software written in C++ for rapid testing and finite element simulations). The final solution to the model will be a dataset describing the electric field distribution or related quantities, such as the potential within the computational model for a given iteration. In some embodiments, the model is voxel-based (i.e., it includes box-shaped volume elements). In these embodiments, finite-difference time-domain (FDTD) algorithms can be used to solve the model, for example, using a quasi-electrostatic solver associated with the "Sim4Life" software package from ZMT Zurich MedTech AG.

[0082] Then, an optimization algorithm is used to find an array layout that optimizes the electric field delivery to the affected region of the patient's brain (tumor) in two application directions (LR and AP). The optimization algorithm will utilize methods for automatic array placement and methods for solving for the electric field within the head model in a well-defined order to find the optimal array layout. Considering the two directions in which the electric field is applied, the optimal layout will be one that maximizes or minimizes some objective function of the electric field in the affected region of the brain. This objective function could be, for example, the maximum intensity or the average intensity within the affected region. Other objective functions can also be defined.

[0083] There are many methods available for finding the optimal array layout for a patient, three of which are described below. One optimization method is exhaustive search. In this method, the optimizer includes a library of array layouts with a finite number of layouts that should be tested. The optimizer performs simulations on all array layouts in the library and selects the layout that produces the optimal field strength in the tumor (the optimal layout is the layout in the library that produces the highest (or lowest) value of the optimization objective function, e.g., the electric field strength delivered to the tumor).

[0084] Another optimization approach is iterative search. This method encompasses the use of algorithms such as minimum descent optimization and simplex search optimization. Using this method, the algorithm iteratively tests different array layouts on the head and computes an objective function for the electric field in the tumor for each layout. At each iteration, the algorithm automatically selects the configuration to be tested based on the results of the previous iteration. The algorithm is designed to converge to maximize (or minimize) the objective function defined for the field in the tumor.

[0085] Another optimization method is based on placing the dipole at the center of the tumor in the model. This method differs from the other two because it does not rely on solving for the field strength of different array layouts. Instead, the optimal location of the array is found by aligning the dipole with the expected field direction at the center of the tumor in the model and solving for the electromagnetic potential. The region on the scalp with the highest potential (or possibly electric field) will be the location where the array is placed. The logic of this method is that the dipole will generate the highest electric field at the center of the tumor. Through reciprocity, if we can generate the calculated field / voltage on the scalp, then we can expect to obtain the maximum field distribution at the center of the tumor (the location where the dipole is placed). The closest our current system can actually achieve is placing the array on the scalp in the region of highest potential caused by the dipole.

[0086] Note that alternative optimization schemes can be used to find the optimal array layout of the electric field in the diseased brain region. For example, algorithms combining the various methods described above. As an example of how these methods can be combined, consider an algorithm that combines the third method described above (i.e., locating dipoles at the tumor center in the model) with the second method (i.e., iterative search). With this combination, the array layout is first found using the dipoles located at the tumor center. This array layout is used as input for the iterative search to find the optimal layout.

[0087] Once the optimal layout of the electric field in the affected area of ​​the patient's brain has been determined (e.g., using any of the methods described herein, or appropriate alternative methods), electrodes are positioned in the determined locations. An AC voltage is then applied to the electrodes (e.g., as described in U.S. Patent 7,565,205, which is incorporated herein by reference) to treat the disease.

[0088] Figure 7 An example system 700 for electrotherapy is depicted, which can be used after the electrode placement has been optimized as described herein. System 700 includes a controller 702 that applies a TTField to the patient by applying voltage to a capacitively coupled array of transducers 42, 44 attached to the patient's scalp 40 at a defined location. Note the front view of the scalp 40. Figure 7As shown in the figure, only three of the four electrode plates are visible, and the eyes and ears are not shown.

[0089] Optionally, the system can be designed to work with multiple model templates. In this case, an additional step S3.5 is performed after step S3 and before step S4. In step S3.5, the similarity between the patient's MRI image and each of the multiple templates is first measured (using a metric such as correlation or mutual information). The deformable template most similar to the patient's MRI image is selected and used in all subsequent steps. Alternatively, in some embodiments, the selection of the deformable template most closely similar to the patient's MRI image can be performed after registering the patient image to a standard space in step S4 and before step S5. In these embodiments, the deformable template most closely similar to the patient's MRI image is used in all steps after S4.

[0090] Optionally, the system can be configured as a learning system, wherein each real head model created using the above process serves as a deformable template for future models. Both the deformable healthy model created in step S5 and the resulting model including defects (created in step S6) can be added to the database. If the patient's MRI images in the original image stack resemble a stored template of a brain with a tumor to a sufficiently close degree, a model representing the patient's MRI images can be created by measuring deformations on the previously stored template.

[0091] Finally, although the concepts presented in this paper are discussed in the context of MRI images of the patient's head, the same principles can be applied to other parts of the patient's body and / or imaging modalities other than MRI.

[0092] Figure 8 This is a flowchart 800 of a method for optimizing electrode placement to be used subsequently for electrotherapy by creating a realistic head model of the patient using a deformable template. The electrotherapy may be TDCS, TMS, or TTField.

[0093] In S10, one or more 3D images of a patient's body region are received. The 3D images may be MRI images, CT images, or images of any other modality known in the art. The body region may be the patient's head or any other body region. Optionally, the images may be preprocessed using any of the methods described herein (e.g., as referenced herein). Figure 1 (as described in step S2).

[0094] In S20, the image portion corresponding to the abnormal tissue is identified. For example, when the body region is the patient's head, such a portion may correspond to a tumor or skull abnormality. The abnormality may be identified manually, automatically, or semi-automatically according to any of the methods described herein or any other suitable method obvious to those skilled in the art. In some embodiments, the portion of the image corresponding to the abnormal tissue is identified by image segmentation.

[0095] In S30, a dataset is generated corresponding to the image with the anomalous tissue masked. This can be achieved, for example, by masking the anomalous tissue, including ignoring the anomalous region during the registration process described in S50 below. In some embodiments, the anomalous region is masked by marking data points in the region and excluding all marked data points during the registration process described in S50 below.

[0096] In S40, model templates are retrieved for the locations of multiple tissue types in a healthy version of a specified patient's body region. For example, when the body region is the patient's head and the abnormal tissue corresponds to a tumor on the patient's head, the model template corresponds to the head of a healthy individual without any tumors. In some embodiments, a model template may be selected from multiple existing model templates based on the similarity between an image and each of the multiple model templates. For example, a similarity metric such as mutual information or distance may be determined between the patient dataset (derived by masking anomalies in patient images) and each of the several model templates, and the model template most similar to the patient dataset (e.g., having the minimum distance or the maximum mutual information) may be selected accordingly. In some embodiments, the model template may include a TPM, and the TPM may correspond to the same healthy individual or multiple individuals from which a model template has been derived (and derived from images of healthy individuals).

[0097] In S50, the model template is deformed in space such that features in the deformed model template are aligned with corresponding features in the dataset. In some embodiments, the model template is deformed by determining a mapping that maps the dataset to the coordinate space of the model template; and the inverse of the mapping is applied to the model template. In some embodiments, the mapping can be determined by registering the dataset to the coordinate space of the model template. That is, the mapping warps the dataset to the model template. Therefore, the inverse of the mapping warps the model template to the dataset, thereby providing a realistic model for the patient if there are no abnormalities. In some embodiments, the mapping from the dataset to the model template is determined for points in the dataset that fall outside the masked portion. In embodiments where the model template includes TPMS, the mapping maps the dataset to TPM, and the inverse of the mapping is applied to each of the TPMs, and the inversely mapped TPMs are combined into a segmented image including the deformed model template.

[0098] In S60, a portion of the deformable model template corresponding to the masked portion of the dataset is modified such that the modified portion represents anomalous tissue. This modification can be performed based on information obtained during the identification of anomalous portions in S20. For example, one or more anomalous tissue types identified in S20 can be assigned to corresponding portions in the deformable model template. Alternatively, a predetermined general tissue type can be assigned to the masked portion.

[0099] In S70, an electrical property model of tissues in the body region is generated based on (a) the locations of multiple tissue types in a modified and altered model template and (b) the locations of anomalous tissues in the modified and altered model template. The electrical properties may be conductivity, resistivity, or any other electrical property related to electrotherapy treatment of the body region. In some embodiments, for example, different electrical property values ​​may be assigned to each tissue type according to a previously populated lookup table.

[0100] In S80, an electrode placement layout that maximizes the field strength in at least a portion of the anomalous tissue is determined using an electrical property model to simulate the electromagnetic field distribution in the body region caused by simulated electrodes placed at multiple different candidate locations corresponding to the body region, and one of the groups is selected. In some embodiments, the electrode placement layout is determined by applying boundary conditions to the simulated electrodes in each of at least two electrode placement layouts; solving for the field distribution in the body region of each of the at least two electrode placement layouts; and selecting the electrode placement layout that produces the strongest field in the anomalous region. The boundary conditions may correspond to, for example, a voltage applied to the simulated electrodes. In some embodiments, numerical techniques such as the finite element method or the finite difference method are used to solve for the field distribution.

[0101] In S90, the determined electrode placement layout is output for subsequent use as a guide to place the electrodes in accordance with the patient's body area before using the electrodes for electrotherapy treatment (e.g., TTField).

[0102] Models constructed in this way can also be used for other applications where calculating the distribution of electric fields and / or currents within the head may be useful. These applications include, but are not limited to: transcranial stimulation with direct current and alternating current; simulating field mapping of implanted stimulation electrodes; planning the location of implanted stimulation electrodes; and source localization in electroencephalography (EEG).

[0103] Finally, although this application describes a method for optimizing the array layout on the head, the same steps can be used to optimize the array layout in other body regions (including but not limited to the chest or abdomen).

[0104] While the invention has been disclosed with reference to certain embodiments, various modifications, alterations, and variations can be made to the described embodiments without departing from the scope and spirit of the invention as defined by the appended claims. Therefore, the invention is not limited to the described embodiments, but has the full scope defined by the language of the appended claims and their equivalents.

Claims

1. An electrotherapy device, comprising a processor configured to execute instructions stored in one or more memory devices to perform electrotherapy, the electrotherapy comprising: The processor receives three-dimensional images of the patient's body regions; Identify the portion of an image that corresponds to abnormal tissue; Generate a dataset corresponding to the image, wherein the anomalous tissue is masked; A model template is retrieved from the computer system's memory device. The model template includes a tissue probability map that specifies the location of multiple tissue types in a healthy version of the patient's body region. The model template is deformed in space so that the features in the deformed model template are aligned with the corresponding features in the dataset. Modify the portion of the deformable model template corresponding to the masked portion of the dataset, such that the modified portion represents the anomalous organization; Based on the locations of the multiple tissue types in the deformed and modified model template (a) and the locations of the anomalous tissues in the deformed and modified model template (b), an electrical property model of the tissues in the body region is generated; The electrode placement layout that maximizes the field strength in at least a portion of the anomalous tissue is determined by using the electrical property model to simulate the electromagnetic field distribution in the body region caused by simulated electrodes placed at multiple different candidate locations corresponding to the body region, and one of the groups is selected.

2. The electrotherapy device according to claim 1, wherein the deformation of the model template includes: Determine the mapping that maps the dataset to the coordinate space of the model template; as well as The inversion of the mapping is applied to the model template.

3. The electrotherapy device of claim 2, wherein the mapping is determined for points in the dataset that fall outside the masking portion.

4. The electrotherapy device according to claim 2, wherein the model template includes a tissue probability map, wherein the mapping maps the dataset to the tissue probability map.

5. The electrotherapy device of claim 4, wherein the tissue probability map is derived from an image of a healthy individual from which the model template has been derived.

6. The electrotherapy device of claim 5, wherein the tissue probability map is derived by simultaneously registering and segmenting the image of the healthy individual using an existing tissue probability map, and wherein the existing tissue probability map is derived from images of multiple individuals.

7. The electrotherapy device according to claim 4, wherein the tissue probability map is an existing tissue probability map derived from images of multiple individuals.

8. The electrotherapy device of claim 4, wherein the inversion of the mapping is applied to each of the tissue probability maps, wherein the inverse mapped tissue probability maps are combined into a segmented image including the deformable model template.

9. The electrotherapy device of claim 8, wherein combining the inverse mapping tissue probability map includes assigning each voxel the tissue type with the highest probability of occupying that voxel on the inverse mapping tissue probability map.

10. The electrotherapy device of claim 8, wherein combining the inverse mapping tissue probability map includes using a lookup table to assign tissue types to each voxel on which more than one tissue type has been assigned on the inverse mapping tissue probability map.

11. The electrotherapy device of claim 1, wherein the identification of the portion of the image corresponding to the abnormal tissue comprises performing segmentation of the image.

12. The electrotherapy device according to claim 1, wherein the electrical property model of the tissue includes a model of conductivity or resistivity.

13. The electrotherapy device according to claim 1, wherein the image includes an MRI image or a CT image.

14. The electrotherapy device of claim 1, wherein the body area includes the patient's head.

15. The electrotherapy device of claim 1, wherein the portion of the image corresponding to the abnormal tissue corresponds to a tumor.

16. The electrotherapy device according to claim 1, wherein the electrotherapy includes a tumor treatment field (TTField).

17. The electrotherapy device of claim 1, wherein determining the electrode placement layout includes: Boundary conditions are applied to the simulated electrodes in each of at least two electrode placement layouts; Solve for the field distribution in the body region of each of the at least two electrode placement layouts; as well as The electrode placement layout is selected to generate the strongest field in the anomalous region.

18. The electrotherapy device of claim 17, wherein the boundary condition corresponds to the voltage or current applied to the simulated electrode.

19. The electrotherapy device of claim 1, wherein the model template is selected from a plurality of model templates based on the similarity between the image and each of the model templates.

20. The electrotherapy device according to claim 1, further comprising: The electrodes are placed in relation to the patient's body area based on the determined electrode placement layout.

Images