TTFIELD therapy with optimization of electrode positions on the head based on MRI-based conductivity measurements

Abstract

When electrodes are used to apply an electric field in target tissue within an anatomical volume (e.g., applying TTFields to treat a tumor), the position of the electrodes can be optimized by obtaining conductivity measurements in the anatomical volume and generating a 3D map of conductivity directly from the obtained conductivity or resistivity measurements, without dividing the anatomical volume into tissue types. The location of the target tissue is identified within the anatomical volume, and the position for the electrode is determined based on the 3D map of conductivity and the location of the target tissue.

Description

[0001] This application is a divisional application of the following application: Application date: October 28, 2016; Application number: 201680062777.8; Invention name “TTFIELD therapy with optimized electrode position on the head based on MRI-based conductivity measurement results”.

[0002] CROSS-REFERENCE TO RELATED APPLICATIONS

[0003] This application claims the benefit of U.S. Provisional Application No. 62 / 247,314 (filed October 28, 2015) and U.S. Provisional Application No. 62 / 294,372 (filed February 12, 2016), each of which is incorporated herein by reference in its entirety. Background Art

[0004] Tumor Treatment Fields or TTFields are low intensity (e.g., 1 V / cm to 3 V / cm) alternating electric fields in the medium frequency range (100 kHz to 300 kHz). This non-invasive treatment targets solid tumors and is described in U.S. Patent 7,565,205, the entire contents of which are incorporated herein by reference. TTFields disrupt cell division by physical interactions with key molecules during mitosis. TTField therapy is an approved monotherapy for recurrent glioblastoma, as well as an approved combination chemotherapy therapy for newly diagnosed patients. These electric fields are non-invasively induced by a transducer array (i.e., an electrode array) placed directly on the patient's scalp. TTFields also appear to be beneficial in treating tumors in other parts of the body.

[0005] TTFields were established as an anti-mitotic cancer treatment modality because they interfere with proper microtubule assembly during metaphase and ultimately destroy the cell during telophase and cytokinesis. Efficacy increases with increasing field strength, and the optimal frequency is cancer cell line dependent, with 200kHz being the frequency with the highest inhibition of glioma cell growth caused by TTFields. For cancer treatment, non-invasive devices have been developed using capacitively coupled transducers placed directly on the skin area close to the tumor. For patients with glioblastoma multiforme (GBM), the most common primary, malignant brain tumor in humans, the device used to deliver TTField therapy is called Optune TM .

[0006] Because the action of TTFields is directional, where cells that divide parallel to the field are more affected than cells that divide in other directions, and because cells divide in all directions, TTFields are typically delivered via two pairs of transducer arrays that generate perpendicular fields within the treated tumor. More specifically, with the Optune system, one pair of electrodes is located on the left and right sides (LR) of the tumor, and another pair of electrodes is located on the anterior and posterior sides (AP) of the tumor. Circulating the field between these two directions (i.e., LR and AP) ensures that the largest range of cell orientations are targeted.

[0007] In vivo and in vitro studies have shown that the efficacy of TTField therapy increases with increasing electric field strength. Therefore, standard practice for the Optune system is to optimize the array placement on the patient's scalp to increase intensity in the area of ​​brain lesions. To date, this has been achieved by rules of thumb (e.g., placing the array on the scalp as close to the tumor as possible) or by using NovoTal TM The system completes array placement optimization. NovoTal TM A limited set of measurements describing the patient's head geometry, tumor size, and its location are used to find the optimal array layout. TM The input measurements are manually derived from the patient's MRI by the physician. TM The optimization algorithm relies on a general understanding of how the electric field is distributed within the head based on the position of the array and does not take into account variations in the distribution of electrical properties within the heads of different patients. These variations may affect the field distribution within the head and tumor, causing NovoTal to TM Situations where the recommended layout is suboptimal. Summary of the invention

[0008] One aspect of the invention relates to a first method of optimizing the positions of a plurality of electrodes placed on a subject's body, wherein the electrodes are used to apply an electric field in a target tissue within an anatomical volume. The first method includes the steps of obtaining conductivity or resistivity measurements in the anatomical volume and generating a 3D map of the conductivity or resistivity of the anatomical volume directly from the obtained conductivity or resistivity measurements without dividing the anatomical volume into tissue types. The method also includes the steps of identifying the location of the target tissue within the anatomical volume and determining the positions for the electrodes based on the 3D map of conductivity or resistivity generated in the generating step and the location of the target tissue identified in the identifying step.

[0009] Some embodiments of the first method further include the steps of attaching electrodes to the subject's body at the locations determined in the determining step, and applying an electrical signal between the electrodes after the attaching step so as to apply an electric field in the target tissue.

[0010] In some embodiments of the first method, the measurements obtained in the obtaining step represent diffusion of molecules. In some embodiments of the first method, the obtaining step includes acquiring MRI data using diffusion weighted imaging. In some embodiments of the first method, the obtaining step includes acquiring MRI data using a customized multi-echo gradient sequence.

[0011] In some embodiments of the first method, the obtaining step includes acquiring MRI data using diffusion tensor imaging. Optionally, in these embodiments, the step of acquiring MRI data using diffusion tensor imaging includes a direct mapping method that assumes a linear relationship between eigenvalues ​​of diffusion and conductivity tensors, σv=s·dv, where σv and dv are the vth eigenvalues ​​of conductivity and diffusion, respectively. Optionally, in these embodiments, the step of acquiring MRI data using diffusion tensor imaging includes a volume normalization method, in which the geometric mean of the eigenvalues ​​of the conductivity tensor in each volume element in the anatomical volume is locally matched to a specific isotropic conductivity value of the tissue type to which the volume element belongs.

[0012] In some embodiments of the first method, the anatomical volume includes white matter and gray matter of the brain.

[0013] In some embodiments of the first method, the anatomical volume is the brain, and the determination of the location for the electrode is based on a composite model in which a 3D map of the conductivity or resistivity of the brain is surrounded by a model of a first shell having a first constant conductivity. In these embodiments, the model of the first shell can represent the scalp, skull, and CSF taken together. Alternatively, in these embodiments, the model of the first shell can represent the CSF, the composite model further includes a second shell representing the skull, the second shell having a second constant conductivity, and the composite model further includes a third shell representing the scalp, the third shell having a third constant conductivity. In these embodiments, the step of determining the location for the electrode can include adding a dipole to the composite model at a location corresponding to the target tissue and selecting an external location where the potential attributable to the dipole is maximum.

[0014] In some embodiments of the first method, the step of determining a position for the electrode comprises calculating a position for the electrode that will provide a maximum strength of the electric field in the target tissue. In some embodiments of the first method, in the generating step, the 3D map has a resolution greater than 1 mm x 1 mm x 1 mm. In some embodiments of the first method, the step of generating the 3D map comprises generating a simple geometric object representing the anatomical volume.

[0015] In some embodiments of the first method, the step of generating a 3D map includes classifying the tissue type for each volume element based on fractional anisotropy. In some embodiments of the first method, the step of generating a 3D map includes classifying the tissue type for each volume element based on average conductivity. In some embodiments of the first method, the step of generating a 3D map includes matching a geometric mean of eigenvalues ​​of the conductivity tensor to a specific isotropic reference value.

[0016] Another aspect of the invention relates to a second method of creating a model of a mammalian head. The head includes brain tissue, CSF, skull, and scalp. The method includes the steps of: modeling a region of the head corresponding to the brain tissue using a 3D conductivity tensor set; and modeling the CSF, skull, and scalp using at least one shell having a constant conductivity.

[0017] In some embodiments of the second method, the step of modeling the head region corresponding to the brain tissue using the set of 3D conductivity tensors is performed without identifying boundaries between different types of healthy brain tissue.

[0018] In some embodiments of the second method, the set of 3D conductivity tensors is obtained using MRI. In some of these embodiments, the set of 3D conductivity tensors is derived from a diffusion tensor imaging dataset.

[0019] In some embodiments of the second method, the step of modeling the CSF, the skull, and the scalp includes the following steps: modeling the CSF as a first shell disposed outside and in contact with brain tissue, the first shell having a first constant conductivity; modeling the skull as a second shell disposed outside and in contact with the CSF, the second shell having a second constant conductivity; and modeling the scalp as a third shell disposed outside and in contact with the skull, the third shell having a third constant conductivity.

[0020] In some embodiments of the second method, the step of modeling the CSF, skull, and scalp includes the step of modeling the CSF, skull, and scalp together as a single shell disposed outside and in contact with the brain tissue, the single shell having a constant conductivity.

[0021] Some embodiments of the second method further include the steps of: identifying the location of the target tissue within the brain tissue; and determining the locations for the plurality of electrodes based on the location of the target tissue identified in the identifying step, the set of 3D conductivity tensors, and the conductivity of the at least one shell. Optionally, these embodiments further include the steps of: attaching the electrodes to the head of the mammal at the locations determined in the determining step, applying electrical signals between the electrodes after the attaching step so as to apply an electric field in the target tissue. Optionally, in these embodiments, the step of determining the locations for the electrodes includes: modeling a dipole at a location corresponding to the target tissue, and selecting a location at which the potential attributable to the dipole is maximum. Optionally, in these embodiments, the step of determining the locations for the electrodes includes calculating the locations for the electrodes that will provide the best combined treatment specifications in the target tissue.

[0022] In some embodiments of the second method, the step of modeling the region using the set of 3D conductivity tensors includes classifying the tissue type for each volume element based on fractional anisotropy. In some embodiments of the second method, the step of modeling the region using the set of 3D conductivity tensors includes classifying the tissue type for each volume element based on average conductivity. In some embodiments of the second method, the step of modeling the region using the set of 3D conductivity tensors includes matching a geometric mean of eigenvalues ​​of the conductivity tensor to a specific isotropic reference value. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flow chart of one example for creating a head model and using the model to optimize the electric field.

[0024] Figure 2 Depicted are the electric field distributions in various cross sections through a virtual tumor in three different models created using the same MRI dataset.

[0025] Figure 3 The electric field distributions for the three anisotropic models in one axial slice through the tumor are depicted.

[0026] Figure 4 Depicted is a front view of a scalp with a transducer array attached to the scalp.

[0027] Figure 5A and Figure 5B A set of shells for two different models is depicted separately.

[0028] Fig. 6A and Figure 6B The lateral and top views of the ventricles and virtual tumor within the white matter shell are depicted, respectively.

[0029] Figure 7The conductivity maps for the five corresponding models and the resulting electric field distribution in the cortex and tumor tissue in the axial plane are depicted. DETAILED DESCRIPTION

[0030] Overcoming NovoTal TM One approach to the limitations of the system is to optimize the array layout based on the accurate calculation of the electric field distribution within the patient's head according to the array position. The patient can be a human or other type of mammal or other animal. This can be done by building a realistic computational model that describes the conductivity distribution within the patient's head. This can be achieved using MRI data. However, to date, deriving such realistic computational head models is time-consuming and requires a lot of manual intervention. The reason is that the model is obtained by dividing the MR images into various tissue types and assigning representative conductivity values ​​to each tissue type. Although the segmentation of the outer layers of the head, such as the scalp, skull, and cerebrospinal fluid (CSF), can be achieved with standard software without great difficulty, the cortical tissue has a very complex geometric pattern and is much more complicated to handle.

[0031] Although automatic and semi-automatic algorithms for segmenting cortical tissue do exist, their performance is often insufficient to create detailed models. Furthermore, the performance of cortical tissue segmentation algorithms deteriorates further when a patient MRI with extensive distortions due to tumor tissue and edema is present in the brain, and thus requires extensive user intervention to complete this task. Therefore, creating realistic patient computational head models by rigorous segmentation of MR images is extremely labor-intensive and time-consuming.

[0032] This application describes a workflow for creating realistic head models that simulate TTFields with minimal user intervention, as well as detailed information on how these head models can be used to optimize TTField array layouts on patients. In the method presented here, the conductivity values ​​in the head model are determined directly from MRI-based conductivity measurements. Therefore, the need for complex and precise segmentation is eliminated, reducing the time and manpower required to create patient computational head models. Once a realistic model has been constructed, the optimization can be performed in a fully or semi-automatic manner using a series of algorithms also described in this article.

[0033] For convenience, this note is divided into three parts: Part 1 provides a detailed description of the method used to create a realistic head model for TTField simulation from MRI data with minimal user intervention. Part 2 provides a detailed description of how to optimize the TTField array position using the model created in Part 1. And Part 3 describes a proof of concept for creating a realistic head model for TTField simulation with minimal user intervention using a simple convex hull to model the outer layer and using a conductivity map to model the brain.

[0034] Figure 1 is a flowchart of one example for creating a model (in steps S11 to S14 ) and optimizing the electric field using the model (in steps S21 to S24 ).

[0035] Part 1: Creating a realistic computational phantom from MRI data.

[0036] Creating an accurate computational phantom preferably involves accurately mapping the electrical properties (e.g., conductivity, resistivity) at each point within the computational phantom. One conventional method for creating a computational phantom involves dividing the head into different tissue types with different isotropic electrical properties. When building a model using this method, it is important to accurately identify the boundaries of each tissue type in 3D space in order to accurately map the electrical properties for each tissue type into the model.

[0037] The embodiments described herein overcome the need for strict segmentation by using MRI sequences, such as diffusion weighted imaging (DWI), diffusion tensor imaging (DTI), or customized multi-echo gradient sequences (GRE), to directly estimate the electrical properties at each point in 3D space. Directly mapping electrical properties using MRI sequences reduces the need for accurate tissue segmentation because the electrical properties of each point are defined directly from the MRI, rather than from the tissue type to which they are assigned during segmentation. Therefore, the segmentation process can be simplified or even eliminated without affecting the accuracy of the computational phantom. Note that while the embodiments described herein discuss mapping conductivity, alternative embodiments can provide similar results by mapping different electrical properties such as resistivity.

[0038] Figure 1 Steps S11 to S14 in illustrative embodiments depict one example of a set of steps that may be used to generate a computational phantom representing a patient based on MRI conductivity measurements.

[0039] Step S11 is the image acquisition step. In this step, both structural data and data from which the conductivity map can be calculated are acquired. The structural data can be obtained from, for example, standard T1 and T2 MRI sequences. Conductivity can be obtained using a variety of MRI data acquisition modes, such as DWI, DTI or GRE. In order to create a good computational phantom, high-resolution images should be obtained. For structural and conductivity related images, a resolution of at least 1 mm×1 mm×1 mm is preferred. Lower resolution images can be used for one or both of these types of images, but the lower resolution will produce a less accurate phantom.

[0040] The dataset is preferably checked and images affected by large artifacts are preferably removed. Some scanner specific preprocessing is preferably applied. For example, the images may be converted from DICOM format to NIFTI. A different step of preprocessing may be to register all images to a standard space (e.g., Montreal Neurological Institute, MNI, space). This can be done using readily available software packages, including but not limited to FSL FLIRT and SPM.

[0041] Step S12 is a step for processing the structural image. As described above, the workflow proposed here utilizes MRI-based conductivity measurements to create a computational phantom. However, the structural image can still be used to identify the boundaries of the head, and to identify regions belonging to specific tissues within the brain, where it may be advantageous to assign typical conductivity values ​​that are not derived from MRI measurements. For example, in some cases, it may be advantageous to identify (and segment) the skull, scalp, and CSF within the image, and assign typical conductivity values ​​to regions corresponding to these tissues (but still relying on the primary MRI-based measurements for regions corresponding to the brain).

[0042] Available software packages can be used to obtain detailed segmentation of these three tissue types and the ventricles. However, due to the complexity of some of these structures, this may still require a lot of manual intervention. Therefore, a simplified solution for creating a head model may be beneficial. One possibility to reduce the complexity of creating a phantom is to simplify the geometry representing the outer model layers (scalp, skull, and CSF). For example, a shell or convex hull of the outer tissue can be used as a model for these layers. If it is feasible to roughly segment the outer layer, the creation of the corresponding convex hull is trivial and can be performed using standard algorithms and software. Another option is that the user measures the thickness of the three outer layers (scalp, skull, and CSF) at a representative area (the area where the transducer array may be placed) by inspection of the structural image. These measurements can be used to create three concentric shells or layers representing the scalp, skull, and CSF. These layers can be obtained by deforming the default elliptical structure, which may be the default convex hull for scalp segmentation.

[0043] Steps S13 and S14 both deal with the processing of DTI images. Step S13 is a step of image preprocessing and tensor estimation. DTI measurements involve the acquisition of multiple images acquired using different imaging conditions. Each image is characterized by its gradient direction and b value. In order to process DTI images, it is first necessary to extract the gradient direction and b value. This can be performed using standard software. Once the gradient direction and b value have been extracted, the image is preferably corrected for distortion caused by sampling motion (such as head movement) and distortion of MRI caused by eddy currents generated during data acquisition. In addition, the image is preferably aligned to overlap with the structural image discussed in the previous stage. Correction for distortion and alignment can be performed using a standard software package. After preprocessing is completed, the diffusion tensor at each point in the relevant area of ​​the model can be estimated.

[0044] There are many software packages for deriving diffusion tensors from DTI images. For example, "A Hitchhiker's Guide to Diffusion Tensor Imaging" by J.M. Suares et al., Frontiers in Neuroscience, Vol. 7, No. 31, pp. 1-14, doi: 10.3389 / fiiins.2013.00031, 2013 includes a detailed summary of available software for tensor estimation and for DTI preprocessing. Two options for deriving diffusion tensors from DTI images were tested. The first option uses the FSL diffusion toolbox for correction and registration of the images and calculates the principal directions (eigenvectors), principal diffusivities (eigenvalues), and fractional anisotropy. The second option is to use the DIFFPREP module from the Tortoise software to perform motion and eddy current distortion correction with B-matrix reorientation. The DIFFCALC module can then be used for the estimation of the diffusion tensor in each voxel and for the calculation of tensor derivatives. In both software packages, the data set can be reoriented to a standard reference frame with B-matrix reorientation, which is naturally the structural image.

[0045] Step S14 is a step of mapping conductivity within the computational phantom. In this step, conductivity values ​​are mapped to each volume element within the computational phantom. In regions belonging to tissue types where the segmentation is sufficiently accurate (e.g., skull or CSF), representative isotropic conductivity values ​​for each tissue type can be assigned. In other regions, conductivity values ​​are assigned based on MRI-based conductivity measurements such as DTI.

[0046] The derivation of conductivity values ​​from DTI data follows from the proposition that the conductivity tensor shares the same eigenvectors as the effective diffusion tensor. Once the diffusion tensor has been estimated for each volume element imaged, an estimate of the conductivity tensor can be formed using any suitable method, some of which are described in detail in "How brain tissue shapes the electric field induced by transcranial magnetic stimulation", A. Opitz et al., Neuroimaging, Vol. 58, No. 3, pp. 49-59, October 2011. For example, one suitable method is called direct mapping (dM), which assumes a linear relationship between the eigenvalues ​​of the diffusion and conductivity tensors, namely, σ v =s·d v , where σ v and d v are the vth eigenvalues ​​of conductivity and diffusion, respectively. Different assumptions about the scaling factor can be used, and the applicable scaling factor s can also be applied as follows. See, for example, M. Rulmann et al., Neuroimaging, Vol. 44, pp. 399-410 (2009), "EEG source analysis of epileptic activity using a 1 mm anisotropic hexahedral finite element head model". Another suitable method is the volume normalization (vN) method, in which the geometric mean of the eigenvalues ​​of the conductivity tensor in each volume element in the brain is locally matched to the specific isotropic conductivity value of the tissue type to which the element belongs. See, for example, D. Gullmar, Neuroimaging, Vol. 51, pp. 145-163 (2010), "The influence of anisotropic conductivity in white matter tissue on EEG / MEG forward and inverse solutions-a high-resolution whole-head simulation study".

[0047] Both of these methods can be used to map conductivity to relevant regions within a computational phantom (primarily cortical regions). However, the vN method requires greater precision in the segmentation because the conductivity values ​​are mapped to each volume element using information about the tissue type in that region. Therefore, assigning a volume element to the wrong tissue type can lead to errors in the conductivity map within the computational phantom. On the other hand, for the dM method, conductivity values ​​are assigned to all elements using the same linear relationship, regardless of the tissue type at that region. Therefore, dM of DTI data may be more useful than vN mapping of DTI data for simplifying the pipeline used to create computational phantoms for TTField simulations. However, note that the constant scaling factor in dM may only lead to accurate values ​​in healthy tissue and may be lower than the optimal value in tumor tissue.

[0048] Alternative mapping methods may also be applied. For example, to overcome the limitation of the vN method (the need for segmentation to exist in order to be able to assign each volume element to a specific tissue type), the tissue type of the volume elements may also be classified by their fractional anisotropy, mean conductivity, or other related measures. Alternatively, the geometric mean of the eigenvalues ​​of the conductivity tensor may be matched to a specific isotropic reference value. This would be a general way to segment or classify tissue types (perhaps even create a complete model) from DTI data alone. Note that when finding fractional anisotropy (or any other measure that can be derived from conductivity data), it is preferred to check neighboring elements to avoid outliers (e.g., to eliminate GM points identified inside the WM).

[0049] Part 2: Optimizing TTField array positions using a realistic head model

[0050] Optimization of the array layout means finding an array layout that optimizes the electric field within the diseased region (tumor) of the patient's brain. The optimization can be implemented by performing the following four steps: (S21) identifying the volume of the treatment target (target volume) within the realistic head model; (S22) automatically placing the transducer array and setting boundary conditions on the realistic head model; (S23) once the array is placed on the realistic head model and the boundary conditions are applied, the electric field generated within the realistic head model is calculated; and (S24) running an optimization algorithm to find a layout that produces the best electric field distribution within the target volume. A detailed example of implementing these four steps is provided below.

[0051] Step S21 involves positioning a target volume (ie defining a region of interest) within the realistic head model.The first step in finding a layout that produces an optimal electric field distribution within the patient's body is to correctly identify the location and target volume where the electric field should be optimized.

[0052] In some embodiments, the target volume will be the gross tumor volume (GTV) or the clinical target volume (CTV). The GTV is the total visible extent and location of the tumor, while the CTV includes the visible tumor (if present) and any other tissue with the presumed tumor. In many cases, the CTV is found by defining a volume that contains the GTV and adding a margin of predetermined width around the GTV.

[0053] In order to identify the GTV or CTV, it is necessary to identify the volume of the tumor within the MRI image. This can be performed manually by the user, automatically, or by a semi-automatic method using a user-assisted algorithm. When this task is performed manually, the MRI data can be presented to the user, and the user can be asked to outline the volume of the CTV on the data. The data presented to the user can be structural MRI data (e.g., T1 data, T2 data). Different MRI modes can be registered with each other, and the user can be presented with options for viewing any of the data sets and outlining the CTV. The user can be asked to outline the CTV on the 3D volume representation of the MRI, or the user can be given the option of viewing a single 2D layer of the data and marking the CTV boundary on each layer. Once the boundary is marked on each layer, the CTV within the anatomical volume (and therefore within the realistic model) can be found. In this case, the volume marked by the user will correspond to the GTV. In some embodiments, the CTV can then be found by adding a margin of a predefined width to the GTV. Similarly, in other embodiments, the user can be asked to mark the CTV using a similar process.

[0054] An alternative to the manual approach is to use an automated segmentation algorithm to find the CTV. These algorithms implement an automated segmentation algorithm to identify the CTV using structural MRI data or possibly DTI data. Note that for this purpose, DTI data can be used for segmentation, as the diffusion tensor within the tumor (and any edematous areas) will be different from its surroundings.

[0055] However, as mentioned above, current fully automatic segmentation algorithms may not be stable enough. Therefore, semi-automatic segmentation methods for MRI data may be preferred. In examples of these methods, the user iteratively provides input to the algorithm (e.g., the location of the tumor on the image, roughly marking the boundaries of the tumor, demarcating the region of interest where the tumor is located), which is then used by the segmentation algorithm. The user may then be provided with the option to refine the segmentation to obtain a better estimate of the CTV location and volume within the head.

[0056] Whether using an automatic or semi-automatic approach, the identified tumor volume will correspond to the GTV, and the CTV can then be automatically found by expanding the GTV volume by a predetermined amount (eg, defining the CTV as the volume containing a 20 mm wide margin around the tumor).

[0057] Note that in some cases it may be sufficient for the user to define a region of interest where they want to optimize the electric field. The region of interest may be, for example, a box volume, a spherical volume, or a volume of any shape within the anatomical volume containing the tumor. When using this method, complex algorithms for accurately identifying the tumor may not be required.

[0058] Step S22 involves automatically calculating the position and orientation of the array on the realistic head model at a given iteration. TM Each transducer array for TTField delivery in the device consists of a set of ceramic disc electrodes that are coupled to the patient's head through a layer of medical gel. When the array is placed on a real patient, the disc is naturally aligned parallel to the skin, and there is good electrical contact between the array and the skin because the medical gel deforms to match the contours of the body. However, the virtual model consists of rigidly defined geometries. Therefore, placing the array on the model requires a precise method for finding the orientation and contours of the model surface at the location where the array is to be placed, as well as finding the thickness / geometry of the gel necessary to ensure good contact between the model array and the realistic patient model. In order to achieve fully automatic optimization of the field distribution, these calculations must be performed automatically.

[0059] Various algorithms can be used to perform this task. The steps of one such algorithm recently designed for this purpose are set forth below.

[0060] a. Define the location where the center point of the transducer array will be placed on the phantom head. This location may be defined by the user or as one of the steps in the field optimization algorithm discussed in step S24.

[0061] b. Using the input from step (a) combined with knowledge of the geometry of the disks and how they are arranged in the array, calculate the approximate locations of the centers of all the disks in the transducer array within the model.

[0062] c. Calculate the orientation of the realistic model surface at the location where the disk is to be placed. This calculation is performed by finding all points on the computational phantom skin that are within a distance of one disk radius from the specified disk center. The coordinates of these points are arranged in the columns of a matrix and a singular value decomposition is performed on the matrix. The normal of the model skin is then the eigenvector corresponding to the smallest eigenvalue found.

[0063] 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 done by finding the parameters of a cylinder whose height is oriented parallel to the skin surface normal. The radius of the cylinder is defined to be equal to the radius of the disk, and its height is set to extend a predetermined amount beyond the point on the skin used to find the normal (this is a predetermined constant). This causes the cylinder to extend from the phantom surface by at least the predetermined amount.

[0064] e. On the model, create the cylinder described in (d).

[0065] f. The area of ​​the cylinder that protrudes into the realistic model of the patient is removed from the model by binary logic operations (eg, subtracting the head from the cylinder). The resulting "truncated cylinder" represents the medical gel associated with the transducer array.

[0066] g. On the outside of the "truncated cylinder" is placed a disk representing the ceramic disk of the transducer array.

[0067] Step S23 involves calculating the electric field distribution within the head model for a given iteration. Once the head phantom is constructed and the transducer array (i.e., electrode array) for applying the field is placed on the realistic head model, a volume mesh suitable for finite element (FE) method analysis can be created. The next boundary condition can be applied to the model. Examples of boundary conditions that can be used include Dirichlet boundary (constant voltage) conditions 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 so that the integral of the normal component of the current density is equal to a specific amplitude. The model can then be solved with a suitable finite element solver (e.g., a low-frequency quasi-static electromagnetic solver) or alternatively with a finite difference (FD) algorithm. Existing software packages such as Sim4Life, Comsol Multiphysics, Ansys or Matlab can be used to perform meshing, applying boundary conditions and solving the model. Alternatively, a custom computer code that implements the FE (or FD) algorithm can be written. The code can leverage existing open source software resources such as C-Gal (for creating meshes) or FREEFEM++ (software written in C++ for rapid testing and finite element simulations). The final solution to the model will be a data set describing the electric field distribution or related quantities, such as the electric potential within the computational phantom for a given iteration.

[0068] Step 24 is the optimization step. An optimization algorithm is used to find an array layout that optimizes the electric field delivery to the diseased region (tumor) of the patient's brain in both application directions (LR and AP, as described above). The optimization algorithm will utilize methods for automatic array placement and methods for solving the electric field within the head model in a well-defined sequence in order to find the best array layout. Taking into account the two directions of application of the electric field, the best layout will be the layout that maximizes or minimizes some objective function of the electric field in the diseased region of the brain. The objective function can be, for example, the maximum intensity within the diseased region or the average intensity within the diseased region. Other objective functions can also be defined.

[0069] There are many methods that can be used to find the best array layout for the patient, three of which are described below. One optimization method is a thorough search. In this method, the optimizer will include a library with a limited number of array layouts that should be tested. The optimizer performs a simulation of all array layouts in the library (e.g., by repeating steps S22 and S23 for each layout), and picks the array layout that produces the best field strength in the tumor (the best layout is the layout in the library that produces the highest (or lowest) value for optimizing the objective function, such as the electric field strength delivered to the tumor).

[0070] Another optimization method is iterative search. This method covers the use of algorithms such as minimum descent optimization method and simplex search optimization. Using this method, the algorithm iteratively tests different array layouts on the head and calculates the objective function for the electric field in the tumor for each layout. Therefore, the method also involves repeating steps S22 and S23 for each layout. At each iteration, the algorithm automatically picks the configuration to be tested based on the results of the previous iteration. The algorithm is designed to converge so that it maximizes (or minimizes) the defined objective function for the field in the tumor.

[0071] Another optimization method is based on placing a dipole at the center of the tumor in the model. This method differs from the other two methods in that it does not rely on solving for field strengths for different array layouts. Instead, the optimal position for the array is found by placing a dipole aligned with the direction of the desired field at the center of the tumor in the model and solving for the electromagnetic potential. The area on the scalp where the potential (or possibly the electric field) is maximum will be where the array is placed. The logic of the method is that the dipole will generate an electric field that is maximum at the center of the tumor. Through interaction, if we can generate a field / voltage on the scalp generated by the computation, we expect to obtain a field distribution that is maximum at the center of the tumor (where the dipole is placed). The closest our current system can get to this in practice is to place the array in the area where the potential induced by the dipole on the scalp is maximum.

[0072] Note that alternative optimization schemes can be used to find an array layout that optimizes the electric field within the brain lesion region. For example, an algorithm that combines 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., positioning the dipole at the center of the tumor in the model) with the second method (i.e., iterative search). With this combination, an array layout is initially found using a dipole at the center of the tumor method. This array layout is used as input to an iterative search to find the optimal layout.

[0073] Part 3: Proof of concept that a simplified head model can be constructed and produces accurate results.

[0074] The proof of concept is based on a modification of a previously developed realistic human head model that contains anisotropic conductivity values ​​for cortical tissue. The model was derived from a healthy subject, so tumors had to be represented by virtual lesions. The phantom has been used to calculate the electric field distribution after TTField application.

[0075] To test the concept, a convex hull of all tissue types except the ventricle was first created. The cystic tumor in this model is represented by two concentric spheres, a viable shell surrounding a necrotic core. It is placed in the right hemisphere close to the lateral ventricle.

[0076] Figure 2 The electric field distribution in various cross sections through the tumor is shown for three different models created using the same MRI dataset. More specifically, Figure 2 Results for two vertical configurations for TTField therapy are shown: left and right (LR) arrays (Panels 21 to 23) and arrays in the anterior and posterior (AP) parts of the head (Panels 24 to 26). Panels 21 and 24 show the results of the classical modeling approach, i.e., a realistic head model in which the MRI is accurately segmented and representative isotropic dielectric properties of each tissue are assigned to all volume elements belonging to that tissue. Panels 22 and 24 show the results of the simplified modeling approach in which tissue types are segmented into convex hulls and representative isotropic dielectric properties are assigned to each tissue type. Panels 23 and 26 show the results of the simplified model in which conductivity values ​​are assigned to each volume element of cortical tissue (GM, WM, and cerebellum) based on conductivity maps derived from DTI images.

[0077] There is a strong correlation between the various modeling approaches. More specifically, the electric field distribution induced by TTFields in the brain and tumor of the realistic head model is not uniform. This means that, although the field strength is closest to the active transducer array, additional hot spots are induced in the center of the head (in tissues with lower conductivity close to the boundary perpendicular to the electric field), as seen in Panels 21 and 24. In the isotropic simplified model, the field distribution only decays away from the transducer due to the smooth tissue interface. Nevertheless, because inhomogeneous dielectric properties are used, "common" hot spots are seen to be close to the ventricles and also within the active shell of the tumor. Close observation of the field distribution inside the tumor shows very similar patterns in the original model and the simple isotropic model, as seen in Panels 22 and 25. The introduction of anisotropic conductivity tensors in brain tissue causes more similar electric field distributions in the brain, as seen in Panels 23 and 26. It appears that the gyri are visible and some major fiber bundles are visible, and the current flowing through them becomes significant.

[0078] When comparing the average electric field values ​​in the tumor calculated using the realistic model with the simplified model, the percentage difference for the isotropic model was less than 6%. When the realistic anisotropic model was compared with the simplified anisotropic model, the percentage difference between the average field strengths in the tumor shell was less than 5%. In both cases, the simplified model's predictions were slightly lower.

[0079] exist Figure 3 , the electric field distribution is again presented in one axial slice through the virtual tumor. In each of panels 31 to 33, the electric field distribution in this axial slice for the LR and AP arrays appears at the top and bottom of the panel, respectively. The original model (panel 31) corresponds to a realistic representation of all tissues with dM anisotropy for cortical tissue. The simple model (panel 32) also uses a dM anisotropic conductivity tensor for cortical tissue (represented by a convex hull), and it takes the convex hull or shell of all surfaces except the ventricles, all of which have isotropic conductivity values. The sample 2 model (panel 33) is similar to the sample model, but the detailed representation of the ventricles is ignored, and their presence is accounted for by using an anisotropic conductivity tensor for this region derived from the DTI data (for the original model and the simple 1 model, this data is ignored or overwritten by the ventricle segmentation with isotropic conductivity values). Table 1 compiles the corresponding average field strength values ​​in the brain and two tumor tissues. Since the virtual lesion is close to the ventricles, the simplification being carried out affects the fields in the tumor more. The differences were relatively small, however, while the mean field in the tumor induced by the LR array increased to 114% in the original realistic model (compared to sample 2) and decreased to 95% in the AP stimulation.

[0080]

[0081] Table 1

[0082] This shows that using the method described herein leads to sufficiently accurate electric field distribution in the head and correct field strength values, while being more efficient in time and computation. It is worth noting that the simplified model should be accurate enough to be used for optimization of the electrode layout.

[0083] Additional details of the modeling of Section 3 will now be discussed, including models in which a simple convex hull or shell is used to model the outer layers and a conductivity map is used to model the brain. These models are able to account for anisotropic conductivity in cortical tissue by using a tensor representation estimated from diffusion tensor imaging. The induced electric field distributions were compared in a simplified head model and a realistic head model. The average field strength values ​​in the brain and tumor tissues were generally slightly higher in the realistic head model, with the maximum ratio being 114% for the standard simplified model (when ensuring layers of reasonable thickness). Therefore, it provides a fast and efficient way to reduce the complexity between tissue interfaces of a personalized head model, thereby enabling accurate prediction of increased electric field distributions.

[0084] This study presents the first approach to personalize head models that does not require an underlying segmentation of different head tissues. Instead, the approach uses a simple convex hull and an electrical conductivity representation of cortical tissue derived from a diffusion tensor imaging (DTI) dataset to model the outer layers.

[0085] A previously developed realistic human head model was used as a baseline model. The MRI dataset of a healthy young female was segmented into scalp, skull, cerebrospinal fluid (CSF), gray matter (GM) including cerebellum, white matter (WM), and ventricles. The centrally located virtual tumor was modeled as two concentric spheres with an inner necrotic core surrounded by a live tumor shell. For all calculations, Optune with a centrally symmetric layout was used. TM system. Figure 4 is a front view of the scalp 40, wherein the Optune TM Transducer arrays 42 and 44 depict this layout. Note that only three of the four patches are visible in the figure, and that no eyes or ears are represented on the convex hull. The final volume mesh is the same as Mimics ( www.materialise.com ) assembled together.

[0086] Isotropic conductivity and permittivity values ​​for heterogeneous tissues were assumed to be the same as in previous studies, and anisotropic conductivity tensors for cortical tissues were estimated from diffusion tensor imaging (DTI) data. Different methods were assumed for scaling the diffusion tensor. In this example, only the direct mapping (dM) method with the same scaling factor for each voxel was used. Further details are presented in C. Wenger et al., "Electric field distribution in the brain during TTField therapy and its dependence on tissue dielectric properties and anatomy: a computational study," Medical Journal, Vol. 60, No. 18, pp. 7339-7357, 2015, which is incorporated herein by reference.

[0087] One way to simplify the model is to use the convex hull of the surface mesh instead of the complex and irregular geometry. In this study, the convex hull was constructed using MeshLab ( http: / / meshlab.sourceforge.net / ). The GM and cerebellum are approximated as a single envelope, and the WM, scalp, skull, and CSF are each represented by a convex hull. Figure 5A and Figure 5B Depicted are the arrangements of the convex hulls (i.e., shells) of two similar simplified models, referred to as SHM1 (51) and SHM2 (52), respectively. In both models, the convex hulls include the skull 54, cerebrospinal fluid 55, gray matter (GM) 56, and white matter (WM) 58. Note that the CSF 55 in SHM1 51 is very thin compared to the CSF 55 in SHM2 52. Fig. 6A and Figure 6B Depicted are the side and top views, respectively, of the ventricle 64 and tumor 66 inside the WM convex hull 62. The ventricle and tumor tissue (viable shell and necrotic core) remain intact.

[0088] Four different simple head models (SHM1 to SHM4) were developed. The first SHM1 consists of the above-mentioned convex hull, which causes very different tissue volumes compared to the RHM. The WM is the innermost layer of the tissue that is changed, which is highly affected by the application of a convex hull with more than one times the tissue volume. This affects the surrounding tissue. The GM is smaller in volume in SHM1. The encapsulation of the GM gyrus and the entire cerebellum reduces the volume of the CSF in SHM1. The only tissue that is slightly larger in volume in SHM1 compared to the RHM is the skull, which in turn causes a reduction in scalp volume. It should still be noted that in SHM1 and RHM, the thickness of the scalp and skull layers below the transducer are very similar, with an average (among all 36 transducers) ratio between the layer thickness of the RHM and SHM1 of 102% in the scalp and 110% in the skull. Nevertheless, the ratio for the CSF is 270%. The thickness was estimated using the volume of the intersecting cylinders, i.e., a cylinder extending the transducer was created and then the volume of the intersection with the next tissue surface was calculated. Thus, the higher volume of the CSF cylinder for RHM is due to the additional volume created by the sulci rather than the planar GM in SHM1.

[0089] A second simple model, SHM2, was created to reduce these differences, i.e., the altered tissue volumes and minimal CSF thickness in SHM1 (e.g. Figure 5ASHM2 was generated by scaling the mesh in Mimics: WM and GM were scaled simultaneously by a factor of 0.97, and then CSF was scaled by a factor of 0.995. This resulted in a reduced difference in layer thickness for SHM2 compared to 102% for RHM, 100% for skull, and 128% for CSF. The two models were first solved as isotropic and anisotropic models and compared to the RHM results. The conductivity tensor estimated with DTI data remained unchanged. Note that in RHM, all DTI data outside the GM boundary is ignored. Diffusion information is added for all additional voxels that are part of the convex hull of the GM in SHM1 and SHM2.

[0090] SHM3 is a simpler model that uses only one convex hull for the cortical tissue and ignores the border between WM and GM. As a final simplification step, SHM4 further cuts the ventricles and is used only with conductivity data derived from DTI instead of the isotropic CSF-filled chambers in all other models.

[0091] To calculate the electric field distribution, the finite element (FE) software Comsol Multiphysics ( http: / / www.comsol.com )A quasi-static approximation of Maxwell's equations in the frequency domain is solved at a frequency of 200kHz. The properties of isotropic and anisotropic materials have been discussed. The boundary conditions assume continuity of the normal component at the inner boundaries, electrical insulation at the outer boundaries. The TTField activation simulates a floating potential condition with 100mA for each active transducer.

[0092] The results of the study are as follows: Each model setting (type of model, isotropic or dM representation of brain conductivity) was solved for both array field directions LR and AP.

[0093] The first simulations were performed with the SHM1 model and the isotropic and anisotropic solution schemes were compared with the RHM model. This initial simplified model SHM1 with its thin CSF produced high electric field strength values ​​in brain and tumor tissue (Table 2). When adjusting the CSF thickness introduced by SHM2, the average field strength values ​​obtained were very similar and slightly reduced in the tumor compared to the RHM. As shown in Table 2, the highest increase in average field strength reported in the tumor shell was 107% with LR activation and isotropic conductivity.

[0094] Figure 7Contains five panels 71 to 75, each of which depicts a conductivity map and the resulting electric field distribution in the cortex and tumor tissue in the axial slices of each model. In each panel 71 to 75, the trace of the conductivity tensor appears on the top, where the legend of the tensor trace is fixed, ranging from 0.1 S / m to 0.6 S / m. The color of the tumor tissue in this figure is arbitrary. In each panel, the electric field distribution for the LR and AP electrodes appears in the middle and bottom, respectively, and the intensity legend ranges from 0 V / cm to 4 V / cm.

[0095] Panels 71 and 72 show the isotropic RHM and SHM2 models with their isotropic brain and tumor conductivity. Although there are only minor details of the electric field distribution in the brain in SHM2, the field distribution in the tumor is similar for LR and AP settings, and the induced average field strength is similar (Table 2). When anisotropy is introduced for brain tissue, the field distribution in the brain of the RHM model changes only slightly (compare Panels 71 and 73); and the SHM2 anisotropic model (Panel 74) shows more details, and the calculated average field strength values ​​are more consistent with those of the anisotropic RHM model (Panel 73).

[0096] The SHM2 model (Panel 74) was used as a further simplified baseline model for the above SHM3 and SHM4. Considering that GM and WM are represented by only two convex hulls, the removal of the inner shell would not be expected because of the dM method for scaling of the conductivity tensor. In fact, almost no changes were found in the mean field strength values ​​(Table 2).

[0097] The ventricles are complex structures in the center of the brain filled with CSF and are therefore considered to be isotropic. Therefore, information that is usually estimated from DTI data is omitted for electric field calculations with a realistic head model, and a detailed segmentation with isotropic conductivity is used. SHM4 was created to study the impact of ignoring ventricle segmentation and considering its presence by using tensors evaluated from DTI data sets. The final trace of the conductivity tensor is shown at the top of panel 75. The average field strength in the brain in the RHM is only slightly higher than in SHM4 (102% in LR and 101% in AP). In the tumor shell, the highest field strength in the RHM increased to 114% of that in the LR compared to the SHM4 model (Table 2). This shows that despite the additional simplifications introduced in the SHM4 model, the results are still acceptable.

[0098] Table 2 depicts the variation of field strength between various models in both LR and AP directions. Note that SHM3 and SHM4 in Table 2 correspond to the sample 1 and sample 2 models in Table 1 above, respectively.

[0099]

[0100] Table 2 - Average field strength (V / cm) of brain and tumor tissue in different models

[0101] The proposed method can be used to quickly create a head model of a GBM patient for personalized treatment planning with TTFields. The scalp contour can be obtained by segmenting the structural image with known software in a minimal amount of time. Alternatively, the head measurements can be used to predict the overall head shape. The following layers (skull, cerebrospinal fluid, brain) can be created by thickness measurements from the structural images. In summary, the proposed technique should be easily applicable to future modeling because the convex hull of the brain exterior can be generated with measurements of the head as the only input. As for the tumor and the brain itself, the DTI dataset for the patient is used to determine the dielectric properties (e.g. conductivity).

[0102] The acquisition of DTI is not standard, however, diffusion weighted imaging (DWI) with fewer directions is fairly commonly acquired, and the traces of the conductivity tensor can be estimated from only three directions. In an alternative embodiment, the induced field distribution can be determined using only the trace values ​​in each voxel rather than the full tensor. This will provide additional simplification of the model, but may sacrifice accuracy.

[0103] DTI is still a relatively new technique and the image resolution is low (i.e., >1 mm 3 isotropic). Therefore, careful selection of image correction and tensor estimation methods is important, and appropriate interpolation methods are recommended. In order to scale the diffusion tensor to the conductivity tensor, two methods are introduced. In addition to the dM method, in the volume normalization (vN) method, the geometric mean of the eigenvalues ​​is matched to the isotropic reference value for each voxel. To achieve this, a basic segmentation of tissue types can be implemented. In some embodiments, the estimated traces of the tensors in each voxel can be used to classify tissue types and used as a proxy for segmentation.

[0104] As already noted, automated segmentation algorithms already exist for detailed GBM segmentation. An example of a publicly available algorithm is the recent Brain Tumor Image Analysis (BraTum1A) software, which distinguishes between necrotic core, edema, non-enhancing tumors, and enhancing tumors while requiring four different imaging modalities (T1, T1-contrast, T2-contrast, and FLAIR). Techniques that only require T1 as input also exist. Nevertheless, the heterogeneous environment of the GBM and surrounding edema can be depicted in more detail even with a voxel-wise tensor representation. Therefore, while the complexity of the simplified model is reduced, it can still be used to describe the electric field distribution of the TTField in more detail.

[0105] This section (i.e., Section 3) presents a first attempt to create a simple head model that provides accurate results for calculating the electric field distribution for TTField application. There is no significant change in the electric field strength in one central tumor when using the simple model compared to a realistic human head model derived from structural images. The method described in this article can be extended to create personalized models without the need for time-consuming tissue segmentation. In the future, this method can be used to quickly develop individual patient head models with detailed representation of lesions, although this requires the availability of an available DTI dataset.

[0106] Once a layout has been determined that optimizes the electric field within the diseased region of the patient's brain (e.g., using any of the methods explained herein), the electrodes are positioned at the determined locations. An AC voltage is then applied to the electrodes (e.g., as described in U.S. Pat. No. 7,565,205, which is incorporated herein by reference) to treat the disease.

[0107] Note that the concepts described herein are not limited to the use of DTI to derive electrical properties of the brain. Rather, it extends to other methods that can be used for the same purpose, including but not limited to DWI, conductivity imaging, electrical impedance tomography (EIT), and multi-echo GRE.

[0108] Note that the concepts described herein are not limited to the representation of the outer layers (scalp, skull, CSF) as a convex hull, and other methods can be used to roughly approximate the MRI data. Examples include simple geometric forms such as ellipsoids, spheres, elliptical structures or other methods for creating an envelope of tissue. In addition, the concepts described herein are not limited to the approximation of the outer layers, i.e., the scalp, skull and CSF layers can also be obtained by conventional segmentation of MRI.

[0109] Optionally, post-processing of the conductivity map may be performed to improve the results (e.g., smoothing or outlier removal / replacement, adaptive interpolation techniques, etc.). Furthermore, other mapping methods from the diffusion to conductivity approach may be used, as well as combinations of the two mentioned methods (e.g., dM and vN methods). Thus, it is advantageous to use dM for cortical tissue and vN for ventricles and tumor tissue including edematous areas, which may have been identified as regions of interest (ROIs) by a clinician or radiologist.

[0110] Some of the above-described embodiments use a hybrid approach in which some volume elements are assigned representative electrical properties of the tissue type to which they belong, while other volume elements are assigned electrical properties based on specific MRI sequence data (DTI in this case). For example, the skull, scalp, and CSF are assigned representative isotropic dielectric properties, while the conductivity of the white and gray matter (and the ventricles in some embodiments) is derived from the DTI data. Note that in the presented case, tumor tissue is also assigned isotropic dielectric properties at virtual locations because the images are derived from healthy subjects. However, in alternative embodiments, the total amount of volume elements within the entire head can be assigned isotropic or anisotropic dielectric properties that are derived only from a specific imaging technique.

[0111] Note that in some embodiments, the boundary surface of the head is identified, for example, by conventional segmentation of the scalp surface, and conductivity and / or dielectric constant are assigned to all points within the phantom using conductivity measurements derived from MRI conductivity measurements.

[0112] In some embodiments, the brain is identified using an existing whole brain extraction algorithm. Next, the scalp, skull, and CSF are segmented using an automated procedure. Conductivity values ​​are assigned to the brain, tumor tissue (including viable shell and necrotic core), possible edema areas, and ventricles using MRI conductivity measurements. Body conductivity values ​​are assigned to the scalp, skull, and CSF.

[0113] In some embodiments, the brain is identified using existing whole brain extraction algorithms. Next, the scalp, skull, CSF, and ventricles are segmented using an automated procedure. Conductivity values ​​are assigned to the brain, tumor tissue (including viable shell and necrotic core), and possible edema areas using MRI conductivity measurements. Body conductivity values ​​are assigned to the scalp, skull, CSF, and ventricles.

[0114] In some embodiments, the brain is identified using an existing whole brain extraction algorithm. The clinician or radiologist marks the tumor as a ROI. Next, an automated procedure is used to segment the scalp, skull, and CSF. Conductivity values ​​are assigned to the brain and ventricles using MRI conductivity measurements. Body conductivity values ​​are assigned to the scalp, skull, CSF, and tumor tissue (e.g., by assigning a constant conductivity value to each of these regions).

[0115] Also note that instead of using the segmentation of scalp, skull and CSF, approximations of these outer layers can be used. For example, the user can be asked to measure the thickness of the scalp, skull and CSF in a representative area. These tissues are then approximated as concentric geometric entities (similar to the default convex hull of the scalp, sphere, ellipsoid, etc.), where the thickness measured by the user surrounds the brain. This approximation simulates the head as a (almost) elliptical structure, ignoring features such as ears, nose, mouth and chin, etc. However, since the array and treatment are only delivered to the supratentorial region of the head, this approximation seems reasonable. In some embodiments, two or more of the three tissue types can also be combined into a layer and a single conductivity value is assigned to the layer. For example, the scalp and skull can be introduced as a layer with a single conductivity (and optionally with uniform thickness).

[0116] The inventors anticipate that the ability to develop realistic head models for individual patients will allow for optimization of not only the electric field within the tumor, but also treatment plans that mitigate field recurrence outside the tumor. This can be achieved by developing optimization methods that not only consider the electric field strength within the tumor, but also attempt to optimize the field strength in other areas of the brain.

[0117] Optionally, patient-specific computational head models can be used for retrospective patient analysis, which can elucidate the relationship between field strength distribution within a patient and disease progression, ultimately leading to a better understanding of how to deliver TTFields within a patient.

[0118] Computational phantoms constructed in this manner can also be used for other applications where calculation of electric fields and / or current distribution within the head may be useful. These applications include, but are not limited to: direct and alternating current transcranial stimulation; simulation of implanted stimulation electrode field maps; planning placement of implanted stimulation electrodes; and source localization in EEG.

[0119] Finally, while this application describes methods for optimizing array placement on the head, it can potentially be extended to optimize array placement for treating other body regions, such as the chest or abdomen.

[0120] Although the present invention has been disclosed with reference to certain embodiments, various modifications, variations and changes may be made to the described embodiments without departing from the scope and range of the invention as defined by the appended claims. Therefore, it is intended that the present invention not be limited to the described embodiments, but rather has the full scope defined by the language of the appended claims and their equivalents.

Claims

1. A method of optimizing the positions of a plurality of electrodes for placement on a subject's body, the method comprising the steps of: generating a 3D map of conductivity or resistivity of an anatomical volume directly from the conductivity or resistivity measurements without separating the anatomical volume into tissue types; identifying a location of target tissue within the generated 3D map of conductivity or resistivity; as well as Positions for the electrodes are determined based on the generated 3D map of conductivity or resistivity and the identified location of the target tissue, including determining the positions for the electrodes by calculating positions for placing the electrodes that will provide maximum strength of the electric field in the target tissue, or by adding dipoles at locations corresponding to the target tissue and selecting external positions at which the potential attributable to the dipoles is maximum.

2. The method of claim 1, wherein the conductivity or resistivity measurements comprise a diffusion-weighted MRI dataset.

3. The method of claim 1, wherein the conductivity or resistivity measurements comprise a customized multi-echo gradient sequence MRI data set.

4. The method of claim 1, wherein the conductivity or resistivity measurements comprise a diffusion tensor MRI dataset.

5. The method of claim 4, wherein the diffusion tensor MRI data set comprises a direct map having a linear relationship between eigenvalues ​​of the diffusion and conductivity tensors, σ v =s·d v , where σ v and d v are the v-th eigenvalues ​​of the conductivity and diffusion, respectively.

6. The method of claim 4, wherein the diffusion tensor MRI dataset is volume normalized and the geometric mean of the eigenvalues ​​of the conductivity tensor in each volume element in the anatomical volume is locally matched to the isotropic conductivity value of the tissue type to which the volume element belongs.

7. The method of claim 1, wherein the anatomical volume comprises white matter and gray matter of the brain.

8. The method of claim 1, wherein the anatomical volume is the brain, and the locations for placing the electrodes are determined based on a composite model, wherein the generated 3D map of conductivity or resistivity is surrounded by a model of a first shell having a first constant conductivity.

9. The method of claim 8, wherein the model of the first shell represents the scalp, skull, and CSF taken together.

10. The method of claim 8, wherein the model of the first shell represents CSF, and the composite model further comprises a second shell representing a skull and a third shell representing a scalp, the second shell having a second constant conductivity and the third shell having a third constant conductivity.

11. The method of claim 8, wherein the locations for placing the electrodes are determined by adding dipoles to the composite model at locations corresponding to the target tissue and selecting external locations at which the potential attributable to the dipoles is maximum.

12. The method of claim 1, wherein the generated 3D map of conductivity or resistivity has a resolution better than 1 mm x 1 mm x 1 mm.

13. The method of claim 1, wherein the generated 3D map of conductivity or resistivity includes geometric objects representing the anatomical volume.

14. The method of claim 1, wherein the generated 3D map of conductivity or resistivity includes classifying each volume element as a tissue type based on fractional anisotropy.

15. The method of claim 1, wherein the generated 3D map of conductivity or resistivity includes classifying each volume element into a tissue type based on average conductivity.

16. The method of claim 1, wherein the generated 3D map of conductivity or resistivity comprises a geometric mean of eigenvalues ​​of a conductivity tensor matched to an isotropic reference value.

17. The method of claim 1, wherein the anatomical volume comprises a chest of the subject.

18. The method of claim 1, wherein the anatomical volume comprises the abdomen of the subject.

Images