Method, system, and apparatus for fast approximation of electric field distribution

Abstract

A method, system, and apparatus for fast approximation of electric field distribution is described.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Application No. 62 / 955,678, filed December 31, 2019, which is incorporated herein by reference in its entirety. [Background technology]

[0002] Tumor Treating Fields, or TTFields, are low-intensity (e.g., 1–3 V / cm) alternating electric fields in the mid-frequency range (100–300 kHz). TTFields therapy is a U.S. Food and Drug Administration (FDA)-approved treatment for glioblastoma multiforme (GBM) and malignant pleural mesothelioma (MPM). This noninvasive treatment targets solid tumors and is described in U.S. Patent No. 7,565,205, which is incorporated herein by reference in its entirety. Clinical trials have shown that adding TTFields to standard treatment significantly extends overall survival in glioblastoma patients. Similar improvements have also been observed in MPM patients. TTFields are applied noninvasively using a pair of transducer arrays placed on the skin near the tumor. These arrays are connected to a field generator that, when activated, generates an alternating electric field in the 100–300 kHz range that propagates into cancerous tissue. TTFields inhibit cell division through physical interactions with key molecules during mitosis. TTFields therapy is an approved monotherapy for recurrent glioblastoma and an approved combination therapy with chemotherapy for newly diagnosed patients. These electric fields are induced noninvasively by a transducer array (i.e., an array of electrodes) placed directly on the patient's scalp. TTFields also appears to be beneficial for treating tumors in other parts of the body. The distribution of the electric field generated by the transducer array maximizes the benefits of TTFields therapy, but optimal positioning of the transducer array is not easily determined. Current methods for estimating the intensity distribution of TTFields rely on finite element methods, which are time-consuming and can require many hours to calculate the electric field generated by a single pair of transducer arrays. For example, estimating the electric field distribution involves complex and time-consuming calculations, requiring at least 3–4 hours on a dedicated high-performance computer. Therefore, in a practical optimization scheme for TTFields treatment planning, only a limited number of transducer array arrangements can be evaluated, and the optimization results may be suboptimal. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] U.S. Patent No. 7,565,205 [Patent Document 2] US Patent Application Publication No. 2019 / 0117956 [Non-patent literature]

[0004] [Non-Patent Document 1] Ballo et al., “Correlation of Tumor treating Fields Dosimetry to Survival Outcomes in Newly Diagnosed Glioblastoma: A Large-Scale Numerical Simulation-based Analysis of Data from the Phase 3 EF-14 randomized Trial,” 2019. Summary of the Invention [Means for solving the problem]

[0005] The described method includes determining a plurality of sets of image data associated with a plurality of patients, each patient associated with a set of image data derived from imaging a portion of the patient, each set of image data including a plurality of voxels, each voxel of the plurality of voxels labeled with a tissue type and each voxel of the plurality of voxels labeled with an electric field strength distribution value (Vcm-1) derived from simulated application of an alternating electric field from a pair of transducer arrays to the portion of the patient; determining a plurality of features for a predictive model based on a first portion of the plurality of sets of image data; training the predictive model based on the plurality of features and the first portion of the plurality of sets of image data, the predictive model being configured to estimate the electric field strength distribution value; testing the predictive model based on a second portion of the plurality of sets of image data; and outputting the predictive model based on the test results.

[0006] The described method also includes determining a set of image data for a patient, the image data set including a plurality of voxels; submitting the image dataset to a predictive model, the predictive model configured to estimate electric field intensity distribution values ​​based on one or more simulated alternating electric fields from pairs of transducer arrays at a plurality of locations; estimating, for each voxel of the plurality of voxels, one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the plurality of locations using the predictive model; and determining a transducer array map including one or more of the plurality of locations based on the estimated one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the plurality of locations, the region of interest, and anatomical constraints associated with the patient.

[0007] Additional advantages will be set forth in part in the description which follows or may be learned by practice. The advantages will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive.

[0008] To facilitate identification of a particular element or description of an activity, one or more of the most significant digits of a reference number may refer to the figure number in which that element is first introduced. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 illustrates an exemplary device for electrotherapy treatment. [Figure 2] FIG. 1 illustrates an exemplary transducer array. [Figure 3A] FIG. 1 illustrates an application of the device for electrotherapy treatment. [Figure 3B] FIG. 1 illustrates an application of the device for electrotherapy treatment. [Figure 4A] FIG. 1 illustrates a transducer array placed on a patient's head. [Figure 4B] FIG. 1 illustrates a transducer array placed on the abdomen of a patient. [Figure 5A] FIG. 1 shows a transducer array placed on the patient's torso. [Figure 5B] FIG. 1 illustrates a transducer array placed on the patient's pelvis. [Figure 6] FIG. 1 is a block diagram illustrating an electric field generator and a patient support system. [Figure 7A] FIG. 1 shows electric field magnitude and distribution (units: V / cm) shown in coronal images from a finite element method simulation model. [Figure 7B] FIG. 10 illustrates exemplary simulation results. [Figure 7C] FIG. 10 illustrates exemplary simulation results. [Figure 7D]FIG. 10 illustrates exemplary simulation results. [Figure 8A] FIG. 8 shows a three-dimensional array layout map 800. [Figure 8B] FIG. 1 illustrates placement of a transducer array on a patient's scalp. [Figure 9A] FIG. 1 shows an axial T1 sequence slice containing the most apical image, including the orbits, used to measure head size. [Figure 9B] FIG. 1 shows a coronal T1 sequence slice select image at the level of the ear canal used to measure head size. [Figure 9C] FIG. 1 shows a post-contrast T1 axial image showing the maximal enhancing tumor diameter used to measure tumor location. [Figure 9D] FIG. 1 shows a post-contrast T1 coronal image showing the maximum enhancing tumor diameter used to measure tumor location. [Figure 10] 10A-10C show the magnitude of the electric field distributions approximated using different methods. [Figure 11] FIG. 1 illustrates an exemplary machine learning system. [Figure 12] FIG. 1 illustrates an exemplary machine learning method. [Figure 13] FIG. 1 illustrates an exemplary decision tree. [Figure 14A] FIG. 1 illustrates an exemplary TTFields estimation. [Figure 14B] FIG. 1 illustrates an exemplary TTFields estimation. [Figure 14C] FIG. 1 illustrates an exemplary TTFields estimation. [Figure 15A] FIG. 1 illustrates an exemplary TTFields estimation. [Figure 15B] FIG. 1 illustrates an exemplary TTFields estimation. [Figure 16A] FIG. 1 shows an exemplary cross-sectional view based on the center point of a tumor in a patient's head. [Figure 16B] FIG. 1 shows an exemplary cross-sectional plan view based on the center point of a tumor in a patient's chest. [Figure 17] FIG. 1 is a block diagram illustrating an exemplary operating environment. [Figure 18] FIG. 1 illustrates an exemplary method. [Figure 19] FIG. 1 illustrates an exemplary method. DETAILED DESCRIPTION OF THE INVENTION

[0010] Before the present methods and systems are disclosed and described, it is to be understood that they are not limited to particular methods, components, or implementations. It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only, and is not intended to be limiting.

[0011] As used in this specification and the appended claims, the singular forms "a," "an," and "the" (corresponding to the articles "a," "an," and "the") include the plural forms unless the context clearly dictates otherwise. Ranges may be expressed herein as from "about" (approximately) one particular value and / or to "about" (approximately) the other particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values ​​are expressed as approximations, it will be understood that the use of the antecedent "about" causes the particular value to form another embodiment. Further, it will be understood that the endpoints of each of the ranges are significant both in relation to the other endpoint, and independently of the other endpoint.

[0012] "Optional" or "optionally" means that the subsequently described event or circumstance may or may not occur, and that the description includes instances in which said event or circumstance occurs and instances in which said event or circumstance does not occur.

[0013] Throughout the description and claims, the words "include," "comprise," and conjugations of words such as "including," "comprising," and "including," "comprises," mean "including but not limited to," and are not intended to exclude, for example, other components, integers, or steps. "Exemplary" means "an example of," and is not intended to convey indication of a preferred or ideal embodiment. "Etc." is not used in a limiting sense, but rather for illustrative purposes.

[0014] Disclosed are components that can be used to implement the disclosed methods and systems. These and other components are disclosed herein, and when combinations, subsets, interactions, groups, etc. of these components are disclosed, specific reference to each of these various individual and collective combinations and permutations may not be explicitly disclosed, but it is understood that each is specifically contemplated and described herein for each and every method and system. This applies to all aspects of this application, including, but not limited to, steps in the disclosed methods. Thus, where there are various additional steps that may be performed, it is understood that each of these additional steps may be performed in any specific embodiment or combination of embodiments of the disclosed methods.

[0015] The method and system of the present invention may be more readily understood by reference to the following detailed description of the preferred embodiments and examples contained therein, as well as the figures and their accompanying description.

[0016] As will be appreciated by those skilled in the art, the methods and systems may take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Further, the methods and systems may take the form of a computer program product on a computer-readable storage medium having computer-readable program instructions (e.g., computer software) embodied in the storage medium. More particularly, the methods and systems may take the form of web-implemented computer software. Any suitable computer-readable storage medium may be utilized, including a hard disk, a CD-ROM, an optical storage device, or a magnetic storage device.

[0017] Embodiments of methods and systems are described below with reference to block diagrams and flowchart illustrations of methods, systems, apparatuses, and computer program products. It will be understood that each block of the block diagrams and flowchart illustrations, and combinations of blocks in the block diagrams and flowchart illustrations, respectively, can be implemented by computer program instructions. These computer program instructions can be loaded onto a general-purpose computer, special-purpose computer, or other programmable data processing apparatus to produce machine-readable means for implementing the functions specified in one or more flowchart blocks.

[0018] These computer program instructions, which cause a computer (or other programmable data processing apparatus) to function in a particular manner, may further be stored in a computer-readable memory, thereby producing an article of manufacture containing computer-readable instructions for implementing the functions specified in one or more of the flowchart blocks, with the instructions stored in the computer-readable memory. The computer program instructions may also be loaded into a computer or other programmable data processing apparatus, causing a series of operational steps to be executed on the computer or other programmable apparatus to create a computer-implemented process, such that the instructions executing on the computer or other programmable apparatus provide the steps for implementing the functions specified in one or more of the flowchart blocks.

[0019] Thus, the blocks of the block diagrams and flowchart illustrations support combinations of means for performing the specified functions, combinations of steps for performing the specified functions, and program instruction means for performing the specified functions. It will also be understood that each block of the block diagrams and flowchart illustrations, and combinations of blocks in the block diagrams and flowchart illustrations, can be implemented by a special-purpose hardware-based computer system that performs the specified functions or steps, or a combination of special-purpose hardware and computer instructions.

[0020] TTFields, also referred to herein as alternating electric fields, have been established as an anti-mitotic cancer treatment because they disrupt proper microtubule polymerization during metaphase, ultimately disrupting cells during telophase and cytokinesis. Their effectiveness increases with increasing field strength, and the optimal frequency depends on the cancer cell line; the highest inhibition of glioma cell growth induced by TTFields was observed at 200 kHz. For cancer treatment, non-invasive devices have been developed, for example, for patients with glioblastoma multiforme (GBM), the most common primary malignant brain tumor in humans, using capacitively coupled transducers placed directly on the skin near the tumor.

[0021] Clinical trials have shown that adding TTFields to standard treatment significantly extends overall survival in patients with glioblastoma. Similar improvements have also been observed in patients with malignant pleural mesothelioma (MPM). Post-hoc analysis of clinical data has shown that application of higher field strengths to the tumor is associated with increased patient survival. Therefore, positioning the transducer array to maximize TTFields strength in cancerous tissue may potentially further extend patient lifespan.

[0022] Because the effects of TTFields are directional, with cells dividing parallel to the field being more affected than cells dividing in other directions, and cells divide in all directions, TTFields are typically applied through two pairs of transducer arrays that generate perpendicular electric fields within the tumor being treated. More specifically, one pair of transducer arrays may be positioned on the left-right (LR) side of the tumor, and the other pair of transducer arrays may be positioned on the anterior-posterior (AP) side of the tumor. Cycling the electric field between these two directions (i.e., LR and AP) ensures that the widest range of cell orientations is targeted. Other positions of the transducer arrays beyond the perpendicular electric field are contemplated. In one embodiment, asymmetric positioning of three transducer arrays is contemplated, where one pair of three transducer arrays may apply an alternating electric field, then another pair of three transducer arrays may apply an alternating electric field, and then the remaining pair of three transducer arrays may apply an alternating electric field.

[0023] In vivo and in vitro studies have shown that increasing the intensity of the electric field increases the effectiveness of TTFields therapy. Therefore, optimizing the array placement on the patient's scalp to increase intensity in affected areas of the brain is standard practice for the Optune system. Optimizing the array placement may be performed by "rules of thumb" (e.g., placing the array on the scalp as close as possible to the tumor), measurements describing the patient's head geometry, tumor dimensions, and / or tumor placement. The measurements used as input may be derived from image data. Image data is intended to include any type of visual data, such as single-photon emission computed tomography (SPECT) image data, X-ray computed tomography (X-ray CT) data, magnetic resonance imaging (MRI) data, positron emission tomography (PET) data, data that can be captured by optical instruments (e.g., photographic cameras, charge-coupled device (CCD) cameras, infrared cameras, etc.), and the like. In some implementations, the image data may include 3D data (e.g., point cloud data) obtained from or generated by a 3D scanner. The optimization can rely on an understanding of how the electric field is distributed within the head as a function of array position, and in some embodiments, may take into account variations in the distribution of electrical properties within different patients' heads. The novel methods described herein incorporate random forest regression (and / or other machine learning) for fast estimation of TTFields. As described herein, key parameters that influence TTFields intensity can be identified, and methods for extracting these parameters are described. The use of random forest regression for fast estimation of TTFields has been validated in GBM patients (at least 10 patients).

[0024] FIG. 1 illustrates an exemplary apparatus 100 for electrotherapy treatment. Generally, apparatus 100 may be a portable, battery- or mains-powered device that generates alternating electric fields within the body through a non-invasive surface transducer array. Apparatus 100 may include an electric field generator 102 and one or more transducer arrays 104. Apparatus 100 may be configured to generate Tumor Treating Electric Fields (TTFields) (e.g., 150 kHz) via field generator 102 and apply the TTFields to a region of the body through one or more transducer arrays 104. Field generator 102 may be a battery- and / or mains-powered device. In one embodiment, one or more transducer arrays 104 are uniformly shaped. In one embodiment, one or more transducer arrays 104 are not uniformly shaped.

[0025] The electric field generator 102 may include a processor 106 in communication with a signal generator 108. The electric field generator 102 may include control software 110 configured to control the execution of the processor 106 and the signal generator 108.

[0026] The signal generator 108 may generate one or more electrical signals in the form of a waveform or pulse train. The signal generator 108 may be configured to generate alternating voltage waveforms (e.g., TTFields) at frequencies within a range of about 50 kHz to about 500 kHz (preferably about 100 kHz to about 300 kHz). These voltages are such that the electric field strength within the tissue to be treated is within a range of about 0.1 V / cm to about 10 V / cm.

[0027] One or more outputs 114 of the electric field generator 102 may be coupled to one or more conductive leads 112, one end of which is attached to the signal generator 108. The opposite ends of the conductive leads 112 are connected to one or more transducer arrays 104 that are activated by an electrical signal (e.g., a waveform). The conductive leads 112 may comprise standard standoff conductors with flexible metal shielding and may be grounded to prevent the electric field generated in the conductive leads 112 from spreading. The one or more outputs 114 may be operated sequentially. Output parameters of the signal generator 108 may include, for example, the strength of the electric field, the wave frequency (e.g., the treatment frequency), and the maximum allowable temperature of the one or more transducer arrays 104. The output parameters may be set and / or determined by the control software 110 in conjunction with the processor 106. After determining the desired (e.g., optimal) treatment frequency, the control software 110 causes the processor 106 to send control signals to the signal generator 108, causing the signal generator 108 to output the desired treatment frequency to the one or more transducer arrays 104.

[0028] The one or more transducer arrays 104 may be configured in various shapes and positions to generate electric fields of desired configurations, directions, and strengths at the target volume where treatment is to be focused. The one or more transducer arrays 104 may be configured to provide two perpendicular electric field directions through the volume of interest.

[0029] The array(s) of the one or more transducer arrays 104 may include one or more electrodes 116. The one or more electrodes 116 may be made from any material having a high dielectric constant. The one or more electrodes 116 may include, for example, one or more insulating ceramic disks. The electrodes 116 may be biocompatible and coupled to a flexible circuit board 118. The electrodes 116 may be configured so that they do not come into direct contact with the skin because they are separated from the skin by a layer of conductive hydrogel (not shown) (similar to that found in electrocardiogram pads).

[0030] The electrodes 116, hydrogel, and flexible circuit board 118 may be attached to a hypoallergenic medical bandage 120 to hold the one or more transducer arrays 104 in place on the body and in continuous direct contact with the skin. Each transducer array 104 may include one or more thermistors (not shown), e.g., eight thermistors (accuracy ±1°C), to measure the skin temperature beneath the transducer array 104. The thermistors may be configured to measure the skin temperature periodically, e.g., every second. The thermistors may be read by the control software 110 while TTFields are not applied to avoid interference with the temperature measurement.

[0031] If the measured temperature falls below a preset maximum temperature (Tmax), e.g., 38.5-40.0°C ± 0.3°C, between two subsequent measurements, the control software 110 can increase the current until it reaches the maximum therapeutic current (e.g., 4 amps peak-to-peak). If the temperature reaches Tmax + 0.3°C and continues to rise, the control software 110 can decrease the current. If the temperature rises to 41°C, the control software 110 can stop TTFields therapy and an overheating alarm can be triggered.

[0032] The one or more transducer arrays 104 may be of different sizes and may include various numbers of electrodes 116 based on the size of the patient's body and / or different treatment modalities. For example, in the context of a patient's chest, the small transducer arrays may each include 13 electrodes, and the large transducer arrays may each include 20 electrodes, with the electrodes interconnected in series within each array. For example, in the context of a patient's head, as shown in FIG. 2, the transducer arrays may each include 9 electrodes, with the electrodes interconnected in series within each array.

[0033] Alternative structures for the transducer array(s) 104 are contemplated and may be used, including, for example, transducer arrays using non-disc-shaped ceramic elements and transducer arrays using non-ceramic dielectric materials positioned on a plurality of flat conductors. Examples of the latter include polymer films disposed on pads of a printed circuit board or on flat metal pieces. Transducer arrays using electrode elements that are not capacitively coupled may also be used. In this situation, each element of the transducer array is implemented using an area of ​​conductive material configured to rest against the subject / patient's body, with no insulating dielectric layer disposed between the conductive elements and the body. Other alternative structures for implementing the transducer array may also be used. Any transducer array (or similar device / component) configuration, arrangement, type, and / or the like may be used in the methods and systems described herein so long as the transducer array (or similar device / component) configuration, arrangement, type, and / or the like (a) is capable of applying TTFields to the body of a subject / patient, and (b) is capable of being positioned, arranged, and / or resting on a portion of the patient / subject's body as described herein.

[0034] The status and monitored parameters of the device 100 may be stored in a memory (not shown) and transferred to a computing device via a wired or wireless connection. The device 100 may include a display (not shown) for displaying visual indicators such as power on, therapy on, alarms, and low battery.

[0035] 3A and 3B illustrate an example application of the device 100. Transducer arrays 104a and 104b are shown incorporated into hypoallergenic medical bandages 120a and 120b, respectively. The hypoallergenic medical bandages 120a and 120b are applied to a skin surface 302. A tumor 304 is located beneath the skin surface 302 and bone tissue 306 and is positioned within brain tissue 308. The electric field generator 102 causes the transducer arrays 104a and 104b to generate an alternating electric field 310 within the brain tissue 308 that disrupts the rapid cell division exhibited by cancer cells in the tumor 304. The alternating electric field 310 has been shown to halt tumor cell growth and / or destroy tumor cells in non-clinical experiments. The use of alternating electric fields 310 takes advantage of the special properties, geometry, and division rate of cancer cells that make them susceptible to the effects of the alternating electric field 310. The alternating electric field 310 changes its polarity at intermediate frequencies (on the order of 100-300 kHz). The frequency used for a particular treatment can be specific to the cell type being treated (e.g., 150 kHz for MPM). The alternating electric field 310 has been shown to disrupt spindle microtubule polymers and result in dielectrophoretic rearrangement of intracellular macromolecules and organelles during cytokinesis. These processes trigger physical disruption of the cell membrane and programmed cell death (apoptosis).

[0036] Because the effect of the alternating electric field 310 is directional, with cells dividing parallel to the field being more affected than cells dividing in other directions, and cells divide in all directions, the alternating electric field 310 can be applied through two pairs of transducer arrays 104 to generate perpendicular electric fields within the tumor being treated. More specifically, one pair of transducer arrays 104 can be positioned on the left-right (LR) side of the tumor, and the other pair of transducer arrays 104 can be positioned on the anterior-posterior (AP) side of the tumor. Cycling the alternating electric field between these two directions (e.g., LR and AP) ensures that the maximum range of cell orientations is targeted. In one embodiment, the alternating electric field 310 can be applied according to a symmetrical setup of the transducer arrays 104 (e.g., four total transducer arrays 104, two matched pairs). In another embodiment, the alternating electric field 310 can be applied according to an asymmetrical setup of the transducer arrays 104 (e.g., three total transducer arrays 104). An asymmetric setup of the transducer arrays 104 may engage two of the three transducer arrays 104 to apply the alternating electric field 310, then switch to another two of the three transducer arrays 104 to apply the alternating electric field 310, and so on.

[0037] In vivo and in vitro studies have shown that increasing the strength of the electric field increases the effectiveness of TTFields therapy. The described methods, systems, and devices are configured to optimize array placement on a patient's scalp to increase intensity in affected regions of the brain.

[0038] As shown in Figure 4A, the transducer array 104 may be mounted on the patient's head. As shown in Figure 4B, the transducer array 104 may be mounted on the patient's abdomen. As shown in Figure 5A, the transducer array 104 may be mounted on the patient's torso. As shown in Figure 5B, the transducer array 104 may be mounted on the patient's pelvic region. Mounting of the transducer array 104 on other parts of the patient's body (e.g., arms, legs, etc.) is specifically contemplated.

[0039] 6 is a block diagram illustrating a non-limiting example of a system 600 including a patient support system 602. The patient support system 602 may comprise one or more computers configured to operate and / or store an electric field generator (EFG) configuration application 606, a patient modeling application 608, and / or image data 610. The patient support system 602 may comprise, for example, a computing device. The patient support system 602 may include, for example, a laptop computer, a desktop computer, a mobile phone (e.g., a smartphone), a tablet, and the like.

[0040] The patient modeling application 608 may be configured to generate a three-dimensional model (e.g., a patient model) of a portion of a patient's body according to the image data 610. The image data 610 may include any type of visual data, such as single-photon emission computed tomography (SPECT) image data, X-ray computed tomography (X-ray CT) data, magnetic resonance imaging (MRI) data, positron emission tomography (PET) data, data that may be captured by an optical instrument (e.g., a photographic camera, a charge-coupled device (CCD) camera, an infrared camera, etc.), and the like. In some implementations, the image data may include 3D data (e.g., point cloud data) obtained from or generated by a 3D scanner. The patient modeling application 608 may also be configured to generate a three-dimensional array layout map based on the patient model and one or more electric field simulations.

[0041] To properly optimize array placement on a portion of a patient's body, image data 610, such as MRI image data, can be analyzed by a patient modeling application 608 to identify regions of interest, including tumors. In the context of a patient's head, a modeling framework based on an anatomical head model using finite element method (FEM) simulations can be used to characterize how electric fields behave and distribute within the human head. These simulations generate realistic head models based on magnetic resonance imaging (MRI) measurements and distinguish tissue types within the head, such as the skull, white matter, gray matter, and cerebrospinal fluid (CSF). Each tissue type can be assigned dielectric properties in terms of specific conductivity and permittivity, and simulations can be performed in which different transducer array configurations are applied to the surface of the model to understand how an externally applied electric field at a preset frequency distributes throughout any portion of the patient's body, such as the brain. The results of these simulations, employing a paired array configuration, constant current, and a preset frequency of 200 kHz, demonstrated that the electric field distribution is relatively nonuniform throughout the brain, with field strengths exceeding 1 V / cm occurring within most tissue compartments, except for the CSF. These results were obtained assuming a total current of 1800 milliamperes (mA) peak-to-peak at the transducer array-scalp interface, a threshold of electric field strength sufficient to arrest cell proliferation in glioblastoma cell lines.

[0042] Additionally, by manipulating the paired transducer array configuration, it is possible to achieve nearly three times the electric field strength to specific regions of the brain, as shown in Figure 7A, which illustrates the electric field magnitude and distribution (in V / cm) shown in a coronal image from a finite element method simulation model. The simulation employs a left-right paired transducer array configuration.

[0043] Based on Ohm's law, Maxwell's equations in materials, and Coulomb's law, the electric field for TTFields depends on the conductivity (σ), permittivity (ε), and distance from the source (d e ) is inversely proportional to each of the TTFields. Inspection of the simulation results, as shown in Figure 7B-7D, suggests that TTFields are larger when the tissue is in close proximity to cerebrospinal fluid (CSF).

[0044] Figure 7B shows an example of a head MRI T1 with gadolinium from a GBM patient undergoing TTFields treatment. Figure 7C shows an example segmentation of the patient's MRI into tissues with different electrical properties. Figure 7D shows the TTFields spatial distribution calculated using finite element methods as described herein. Note that TTFields are enhanced near the cerebrospinal fluid (arrows). Additionally, TTFields are larger in tissues closer to the dashed line between the centers of the TA pairs.

[0045] A possible explanation for this observation is that CSF is highly conductive, so electrons accumulate at the boundaries of the CSF, thus increasing the potential within these zones. The shortest distance of a voxel from a CSF voxel is d c Another observation is that TTFields are larger in tissues closer to an imaginary line between the centers of the TA pairs (Figure 7B). This observation is consistent with a generalization of Coulomb's law to finite parallel plates in homogeneous materials. The distance between a voxel and a line along the TA center is denoted as dl. The conductivity and permittivity are expected to have a linear relationship with the electric field, and the distance is polynomial to the electric field.

[0046] Given a patient's head MRI, the above key parameters were extracted as follows. First, the head was segmented into eight tissues (Figure 7C): 1) skin and muscle (as a single tissue), 2) skull, 3) CSF, 4) white matter, 5) gray matter, 6) tumor-enhancing, 7) tumor-necrotic, and 8) tumor resection cavity. Tumor segmentation was performed semi-automatically using region growing and active contouring. Segmentation of head tissues (1–5) was performed automatically with a custom atlas-based method. The electrical conductivity and permittivity of different tissues were determined. The distances to each voxel from a line along the power, CSF, and TA centers were efficiently calculated.

[0047] In one embodiment, the patient modeling application 608 can be configured to determine a desired (e.g., optimal) transducer array layout for a patient based on tumor location and extent. For example, initial morphometric head size measurements can be determined from a T1 sequence of a brain MRI using axial and coronal images. Post-contrast axial and coronal MRI slices can be selected to show the maximum diameter of the enhancing lesion. Employing measurements of head size and distance from predetermined fiducial markers to tumor margins, various permutations and combinations of paired array layouts can be evaluated to generate a configuration that applies the greatest field strength to the tumor site. As shown in FIG. 8A, the output can be a three-dimensional array layout map 800. The three-dimensional array layout map 800 can be used by the patient and / or caregiver in positioning and configuring arrays on the scalp during the normal course of TTFields therapy, as shown in FIG. 8B.

[0048] In one embodiment, the patient modeling application 608 can be configured to determine a three-dimensional array layout map for the patient. MRI measurements of a portion of the patient that is to receive the transducer array can be determined. For example, the MRI measurements can be received via a standard Digital Imaging and Communications in Medicine (DICOM) viewer. The MRI measurement determination can be performed automatically, e.g., using artificial intelligence techniques, or manually, e.g., by a physician.

[0049] Manual MRI measurement determination may include receiving and / or providing MRI data via a DICOM viewer. The MRI data may include a scan of a portion of a patient containing a tumor. For example, in the context of a patient's head, the MRI data may include a head scan containing one or more of a right frontotemporal tumor, a right parietotemporal tumor, a left frontotemporal tumor, a left parieto-occipital tumor, and / or a multifocal midline tumor. Figures 9A, 9B, 9C, and 9D illustrate example MRI data showing a scan of a patient's head. Figure 9A shows an axial T1 sequence slice including an apical-most image, including the orbit, used to measure head size. Figure 9B shows a coronal T1 sequence slice selecting an image at the level of the ear canal, used to measure head size. Figure 9C shows a post-contrast T1 axial image showing the maximum enhancing tumor diameter, used to measure tumor location. Figure 9D shows a post-contrast T1 coronal image showing the maximum enhancing tumor diameter, used to measure tumor location. MRI measurements may begin from fiducial markers at the lateral edge of the scalp and extend tangentially from a right, anterior, superior origin. Morphometric head size may be estimated from an axial T1 MRI sequence selecting the most apical image that still included the orbits (or the image directly above the superior edge of the orbits).

[0050] In one embodiment, the MRI measurements may include, for example, one or more of head size measurements and / or tumor measurements. In one embodiment, one or more MRI measurements may be rounded to the nearest millimeter and provided to a transducer array mounting module (e.g., software) for analysis. The MRI measurements may then be used to generate a three-dimensional array layout map (e.g., three-dimensional array layout map 800).

[0051] The MRI measurements may include one or more head size measurements, such as maximum anterior-posterior (AP) head size starting from the outer edge of the scalp, maximum head width perpendicular to the AP measurement, lateral distance from right to left, and / or distance from the right-most edge of the scalp to the anatomical midline.

[0052] The MRI measurements may include one or more head size measurements, such as coronal head size measurements. Coronal head size measurements may be acquired with a T1 MRI sequence that selects an image at the level of the ear canal (FIG. 9B). The coronal head size measurements may include one or more of a vertical measurement from the vertex of the scalp to an orthogonal line delineating the inferior border of the temporal lobes, maximum left and right temporal width, and / or the distance from the right edge of the scalp to the anatomical midline.

[0053] The MRI measurements may include one or more tumor measurements, such as tumor location measurements. Tumor location measurements may be performed first on an axial image showing the maximum enhanced tumor diameter (FIG. 9C) using a T1 post-contrast MRI sequence. The tumor location measurements may include one or more of: maximum AP head size excluding the nose; maximum left-right lateral diameter measured perpendicular to the AP distance; distance from the right edge of the scalp to the anatomical midline; distance from the right edge of the scalp to the nearest tumor edge measured parallel to the left-right lateral distance and perpendicular to the AP measurement; distance from the right edge of the scalp to the farthest tumor edge measured parallel to the left-right lateral distance and perpendicular to the AP measurement; distance from the forehead to the nearest tumor edge measured parallel to the AP measurement; and / or distance from the forehead to the farthest tumor edge measured parallel to the AP measurement.

[0054] The one or more tumor measurements may include coronal tumor measurements. Coronal tumor measurements may include identifying a post-contrast T1 MRI slice characterized by the maximum diameter of tumor enhancement (FIG. 9D). Coronal tumor measurements may include one or more of the following: maximum distance from the vertex of the scalp to the inferior border of the cerebrum. In anterior slices, this is bounded by a horizontal line drawn at the inferior border of the frontal or temporal lobe; posteriorly, it extends to the lowest level of the visible tentorium, maximum left-right lateral head width, the distance from the right edge of the scalp to the anatomical midline, the distance from the right edge of the scalp to the nearest tumor edge measured parallel to the right-left lateral distance, the distance from the right edge of the scalp to the farthest tumor edge measured parallel to the right-left lateral distance, the distance from the vertex to the nearest tumor edge measured parallel to the superior vertex to the inferior cerebral line, and / or the distance from the vertex to the farthest tumor edge measured parallel to the inferior cerebral line.

[0055] Other MRI measurements may be used, especially when the tumor is located in another part of the patient's body.

[0056] The MRI measurements may be used by the patient modeling application 608 to generate a patient model. The patient model may then be used to determine a three-dimensional array layout map (e.g., three-dimensional array layout map 800). Continuing with the example of a tumor in a patient's head, a healthy head model may be generated to serve as a deformable template from which a patient model can be created. When creating the patient model, the tumor may be segmented from the patient's MRI data (e.g., one or more MRI measurements). Segmenting the MRI data identifies tissue types within each voxel, and electrical properties may be assigned to each tissue type based on empirical data. Table 1 shows standard electrical properties of tissues that may be used in simulations. The region of the tumor in the patient MRI data may be masked, and a non-rigid registration algorithm may be used to register the remaining regions of the patient's head onto the 3D discrete image representing the deformable template of the healthy head model. This process produces a non-rigid transformation that maps the healthy portion of the patient's head into template space, as well as an inverse transformation that maps the template into patient space. This inverse transformation is applied to the 3D deformable template to generate an approximation of the patient's head in the absence of the tumor. Finally, the tumor (referred to as the region of interest (ROI)) is implanted back into the deformed template to generate a complete patient model, which may be a digital representation in three-dimensional space of a patient's body part, including internal structures such as tissues, organs, and tumors.

[0057] [Table 1]

[0058] In some cases, the patient modeling application 608 may use artificial intelligence and / or machine learning to determine a three-dimensional transducer array layout map for a patient. For example, features related to electric field distribution may be extracted from a patient MRI. The extracted features may be incorporated into a compact model for quickly estimating, predicting, and / or determining the electric field and / or electric field distribution for the transducer array by providing a pre-computed mapping from MRI features to electric field values. The modeling may be based on linear equations or involve complex machine learning models. For example, the model may be tested based on a dataset determined from a patient who underwent TTFields therapy for the management of glioblastoma multiforme (GBM). The dataset may be based on data from three-dimensional (3D) MRI images of the patient's head (or any other region of the patient's body), with voxels labeled to indicate segmentation of different tissue types, such as tumor and combined skin and muscle tissue, skull / bone, cerebrospinal fluid, gray matter, and white matter. Unlabeled voxels in the 3D image may be considered air. If the electric field strength is inversely proportional to the conductivity, the permittivity, the distance from the power source, and the distance from the conductive material, the following modeling equation may be used:

[0059]

number

[0060] where c represents the electrical conductivity, p represents the dielectric constant, and d TA represents the distance to the nearest transducer array, and d conductor represents the distance to the nearest conductor. irepresents the model and is calculated only once. Equation 1 represents a simple model that can be solved quickly, for example, using a differential equation solver, using linear approximation, and / or the like. A differential equation solver can be used to solve Equation 1. For example, using a differential equation solver, the approximate calculation time is 3 hours. A threshold can be applied to the resulting electric field determined by the differential equation solver to create a binary image representing the placement of the transducer array. Note that in some cases, the data set used for the model can include data from each transducer array. To approximate the electric field using Equation 1, each voxel in the 3D MRI image can be associated with conductivity and permittivity values ​​depending on the tissue type associated with that voxel. The distance between the nearest transducer and the nearest cerebrospinal fluid can be calculated at each voxel. The coefficient a in Equation 1 i can be calculated to minimize the least-squares error. The calculation of Equation 1 can be applied to each voxel of a 3D MRI image, thereby creating an approximate electric field for a given volume. Figure 10 illustrates calculated electric field magnitudes. 1010 illustrates the electric field magnitude calculated with a differential equation solver, with an approximate calculation time of 3 hours. 1020 illustrates the electric field magnitude calculated using a fast linear approximation, with an approximate calculation time of 4 minutes. Comparing the results of quickly approximating the electric field using modeling with the fully calculated electric field, the mean difference between the estimated and fully calculated electric field was 0.56 V / cm² (SD = 2.01 V / cm²), and the Pearson correlation coefficient between the estimate and the full calculation was 0.74 (p < 0.05), indicating that the electric field can be approximated with fast calculations. Machine learning can be used to determine more complex models that yield improved estimates in less time.

[0061] The machine learning model may be trained on multiple datasets (or large datasets) created from features extracted from 3D MRI images from various patients, such as patients who underwent TTFields therapy for the management of glioblastoma multiforme (GBM). The features extracted from each image may include the patient's head, with voxels labeled to indicate segmentation of different tissue types, such as tumor and combined skin and muscle tissue, skull / bone, cerebrospinal fluid, gray matter, and white matter. Unlabeled voxels in the 3D images may be considered air. Note that the datasets (or large datasets) may be altered, modified, and / or similarly manipulated to include any data relevant to calculating the electric field for the transducer array layout.

[0062] 11 , the system 1100 may use machine learning techniques to train at least one machine learning-based classifier 1130 configured to classify features extracted from 3D MRI images comprising a plurality of voxels based on analysis of one or more training data sets 1110A-1110N by a training module 1120. The machine learning-based classifier 1130 may classify features extracted from the 3D MRI images to enable fast approximation of the electric field distribution based on a predictive model.

[0063] One or more training datasets 1110A-1110N may include labeled baseline tissue types, electric field distributions, transducer array locations inducing therapeutic tumor treatment fields, and / or the like from a large number of patient 3D MRI images. The labeled baseline data may be associated with patients treated for glioblastoma (GBM) or any other cancer. In some cases, the labeled datasets may include features extracted from 3D MRI images of other parts of the patient's anatomy, from different tumor tissues, from transducer arrays composed of various materials, and / or the like. In some cases, the labeled baseline data may include labeled baseline data from patients before, during, and after administration of TTFields treatment. The labeled baseline data may include any number of feature sets (labeled data identifying extracted features). The feature set may be based on tumor segmentation, different tissue types (e.g., combined skin and muscle tissue, skull / bone, cerebrospinal fluid, gray matter, and white matter, etc.), or any other feature related to (e.g., that may affect, etc.) the electric field and / or electric field distribution relative to the transducer array. For example, each voxel in a 3D MRI image may be labeled with a tissue type and a field strength distribution value (Vcm) derived from a simulated application of an alternating electric field from a pair of transducer arrays to a portion of a patient.

[0064] The labeled baseline data may be stored in one or more databases. Data for each patient may be randomly assigned to a training data set or a test data set. In some implementations, the assignment of data to a training data set or a test data set may not be completely random. In this case, one or more criteria may be used in the assignment, such as ensuring that similar numbers of patients with various grades of cancer (e.g., benign, malignant, etc.), different transducer array layouts, different transducer construction materials, and / or the like are used in each of the training and test data sets. In general, any suitable method may be used to assign data to a training data set or a test data set.

[0065] The training module 1120 may train the machine learning-based classifier 1130 by extracting feature sets from the labeled baseline data according to one or more feature selection techniques. In some cases, the learning module 1120 may further define feature sets obtained from the labeled baseline data by applying one or more feature selection techniques to the labeled baseline data in one or more training datasets 1110A-1110N. The training module 1120 may extract feature sets from the training datasets 1110A-1110N in various ways. The training module 1120 may perform feature extraction multiple times, each time using a different feature extraction technique. In some cases, feature sets generated using different techniques may each be used to generate a different machine learning-based classification model 1140. In one embodiment, the feature set with the highest quality metric may be selected for use in training. The training module 1120 may use the feature sets to build one or more machine learning-based classification models 1140A-1140N configured to perform fast approximation of the electric field distribution.

[0066] In some cases, the training data sets 1110A-1110N and / or the labeled baseline data may be analyzed to determine any dependencies, associations, and / or correlations between transducer array positions, tissue types, and electric field distributions in the training data sets 1110A-1110N and / or the labeled baseline data. As used herein, the term "feature" may refer to any characteristic of an item of data that may be used to determine whether the item of data falls within one or more particular categories. For example, the features described herein may include transducer array characteristics (e.g., material properties, positioning rules and characteristics, tissue types, and related electrical / dielectric properties (e.g., conductivity, permittivity, etc.)), electric field distribution, and / or the like.

[0067] In some cases, the feature selection technique may include one or more feature selection rules. The one or more feature selection rules may include a field distribution rule. The field distribution rule may include determining which features in the labeled baseline data occur more than a threshold number of times in the labeled baseline data and identifying features that meet the threshold as candidate features. For example, any feature that occurs more than twice in the labeled baseline data may be considered as a candidate feature. Any feature that occurs less than twice may be excluded from consideration as a feature. In some cases, a single feature selection rule may be applied to select features, or multiple feature selection rules may be applied to select features. In some cases, feature selection rules may be applied in a cascading manner, where feature selection rules are applied in a specific order and apply to the results of previous rules. For example, a field distribution rule may be applied to the labeled baseline data to generate a field distribution based on the transducer array arrangement. The final list of candidate genes may be analyzed according to additional features.

[0068] In some cases, the fast approximation of the electric field distribution may be based on a wrapper method. The wrapper method may be configured to use a subset of features and train a machine learning model using the subset of features. Features may be added to and / or removed from the subset based on inferences drawn from previous models. Wrapper methods include, for example, forward feature selection, backward feature elimination, recursive feature elimination, combinations thereof, and the like. In some cases, forward feature selection may be used to identify one or more candidate electric field distributions. Forward feature selection is an iterative method that starts with no features in the machine learning model. In each iteration, features that best improve the model are added until adding new variables no longer improves the performance of the machine learning model. In one embodiment, backward feature elimination may be used to identify one or more candidate electric field distributions. Backward feature elimination is an iterative method that starts with all features in the machine learning model. In each iteration, the least important features are removed until no improvement is observed with feature removal. In one embodiment, recursive feature elimination may be used to identify one or more candidate electric field distributions. Recursive feature elimination is a greedy optimization algorithm that aims to find the best-performing feature subset. Recursive feature elimination iteratively builds models, setting aside the best or worst-performing features at each iteration. Recursive feature elimination builds the next model with the remaining features until all features are exhausted. Recursive feature elimination then ranks the features based on the order of their elimination.

[0069] In some cases, one or more candidate field distributions (approximations) may be selected according to an embedding method. The embedding method combines the attributes of filter and wrapper methods. The embedding method includes, for example, least absolute shrinkage and selection operator (LASSO) and ridge regression, which implement a penalty function to reduce overfitting. For example, LASSO regression performs L1 regularization, which applies a penalty equal to the absolute value of the coefficient magnitude, while ridge regression performs L2 regularization, which applies a penalty equal to the square of the coefficient magnitude.

[0070] After the training module 1120 generates the feature set, the training module 1120 may generate a machine learning-based predictive model 1140 based on the feature set. A machine learning-based predictive model may refer to a complex mathematical model for data classification generated using machine learning techniques. In one example, the machine learning-based classifier may include a map of support vectors representing boundary features. As one example, the boundary features may be selected from the feature set and / or represent the highest-ranking features in the feature set.

[0071] In one embodiment, the training module 1120 may use feature sets extracted from the training datasets 1110A-1110N and / or labeled baseline data to construct machine learning-based classification models 1140A-1140N to approximate electric field distributions for various transducer array layouts. In some examples, the machine learning-based classification models 1140A-1140N may be combined into a single machine learning-based classification model 1140. Similarly, the machine learning-based classifier 1130 may represent a single classifier including single or multiple machine learning-based classification models 1140 and / or multiple classifiers including single or multiple machine learning-based classification models 1140. Also, in some embodiments, the machine learning-based classifier 1130 may include each of the training datasets 1110A-1110N and / or respective feature sets extracted from the training datasets 1110A-1110N and / or extracted from the labeled baseline data.

[0072] Features extracted from 3D MRI images may be combined in a classification model trained using machine learning approaches such as discriminant analysis, decision trees, nearest neighbor (NN) algorithms (e.g., k-NN models, replicator NN models, etc.), statistical algorithms (e.g., Bayesian networks, etc.), clustering algorithms (e.g., k-means, mean shift, etc.), neural networks (e.g., reservoir networks, artificial neural networks, etc.), support vector machines (SVMs), logistic regression algorithms, linear regression algorithms, Markov models or chains, principal component analysis (PCA) (e.g., for linear models), multilayer perceptron (MLP) ANNs (e.g., for nonlinear models), replicated reservoir networks (e.g., for nonlinear models, typically for time series), random forest classification, combinations thereof, and / or the like. The resulting machine learning-based classifier 1130 may include a decision rule or mapping using the transducer array layout to the candidate electric field distributions.

[0073] The transducer array layout and machine learning-based classifier 1130 may be used to approximate the relevant electric field distributions of the test samples in the test dataset. In one example, the results for each test sample include a confidence level corresponding to the likelihood or probability that the corresponding test sample belongs to the approximated electric field distribution. The confidence level may be a value between 0 and 1 indicating the likelihood that the approximated electric field matches the calculated value. Multiple confidence levels may be provided for each test sample and each candidate (approximated) electric field distribution. The best-performing candidate electric field distribution may be determined by comparing the results obtained for each test sample with the calculated electric field distribution for each test sample. Generally, the best-performing candidate electric field distribution has results that closely match the calculated electric field distribution. The best-performing candidate electric field distribution may be used for fast approximation of the electric field distribution.

[0074] 12 is a flowchart illustrating an example training method 1200 for fast approximation of electric field distribution using a training module 1120. The training module 1120 can implement supervised, unsupervised, and / or semi-supervised (e.g., reinforcement-based) machine learning-based classification models 1140. The method 1200 illustrated in FIG. 12 is one example of a supervised learning method, and variations of this example training method are described below, although other training methods can similarly be implemented to train unsupervised and / or semi-supervised machine learning (prediction) models.

[0075] At 1210, the training method 1200 may determine (e.g., access, receive, retrieve, etc.) 3D MRI data for one or more populations of patients. The electric field distribution data may include one or more datasets, each dataset associated with a tumor segmentation, different tissue types (e.g., combined skin and muscle tissue, skull / bone, cerebrospinal fluid, gray matter, and white matter, etc.), or any other features related to (e.g., that may influence, etc.) the electric field and / or field distribution relative to the transducer array. Each dataset may include labeled baseline data. Each dataset may further include labeled tumor, tissue types (e.g., combined skin and muscle tissue, skull / bone, cerebrospinal fluid, gray matter, and white matter, etc.), or any other data related to (e.g., that may influence, etc.) the electric field and / or field distribution relative to the transducer array.

[0076] The training method 1200 may generate a training data set and a test data set at 1220. The training data set and the test data set may be generated by calculating and / or computing the electric field distribution of the transducer array at different locations on different regions of the patient. In some cases, the training data set and the test data set may be generated by randomly assigning electric field distribution data to either the training data set or the test data set. In some cases, the assignment of electric field distribution data as training or test samples may not be completely random. In some cases, only labeled baseline data for a particular feature extracted from the 3D MRI may be used to generate the training data set and the test data set. In some cases, a majority of the labeled baseline data extracted from the 3D MRI may be used to generate the training data set. For example, 75% of the labeled baseline data extracted from the 3D MRI may be used to generate the training data set, and 25% may be used to generate the test data set. Any method or technique may be used to create the training data set and the test data set.

[0077] The training method 1200 may, for example, determine (e.g., extract, select, etc.) one or more features that can be used by the classifier label features extracted from various 3D MRI images at 1230. The one or more features may include tumor identifiers / data (e.g., cell type, cell orientation, etc.), tissue types (e.g., combined skin and muscle tissue, skull / bone, cerebrospinal fluid, gray matter, and white matter, etc.), or any other data related to (e.g., that may affect, etc.) the electric field and / or electric field distribution relative to the transducer array. In some cases, the training method 1200 may determine a set of training baseline features from a training dataset. In some cases, the training method 1200 may determine a set of training in-treatment transducer array layouts and associated electric field distributions from a training dataset. The features of the 3D MRI images may be determined by any method.

[0078] The training method 1200 may train one or more machine learning models using one or more features at 1240. In some cases, the machine learning models may be trained using supervised learning. In other embodiments, other machine learning techniques, including unsupervised learning and semi-supervised learning, may be employed. The machine learning models trained at 1240 may be selected based on different criteria (e.g., the material of the transducer array, the applied signal strength and frequency, etc.) and / or the data available in the training dataset. For example, machine learning classifiers may suffer from different degrees of bias. Thus, multiple machine learning models may be trained at 1240 and optimized, improved, and cross-validated at 1250.

[0079] The training method 1200 may select one or more machine learning models (e.g., machine learning classifiers, predictive models, etc.) to build a predictive model at 1260. The predictive engine may be evaluated using a test dataset. The predictive engine may analyze the test dataset and generate classified values ​​and / or predicted values ​​at 1270. The classified values ​​and / or predicted values ​​may be evaluated at 1280 to determine whether such values ​​achieved a desired level of accuracy. The performance of the predictive engine may be evaluated in a number of ways based on the number of true positive, false positive, true negative, and / or false negative classifications of a plurality of data points indicated by the predictive engine. For example, a false positive of a predictive engine may refer to the number of times the predictive engine inaccurately approximated the electric field distribution. Conversely, a false negative of a predictive engine may refer to the number of times the machine learning model incorrectly approximated the electric field distribution for a particular transducer array layout when, in fact, the approximated electric field distribution matches the calculated electric field distribution. True negatives and true positives may refer to the number of times the predictive engine correctly approximated the electric field distribution. Related to these measurements are the concepts of recall and precision. Generally, recall refers to the ratio of true positives to the sum of true positives and false negatives, which quantifies the sensitivity of a prediction engine. Similarly, precision refers to the ratio of true positives to the sum of true positives and false positives.

[0080] If such a desired level of accuracy is reached, the training phase ends and the prediction engine is output at 1290; however, if the desired level of accuracy is not reached, a subsequent iteration of the training method 1200 may be performed starting at 1210 with modifications, such as by considering larger collected transducer array layout-related electric field distribution data determined from more 3D MRI images.

[0081] A trained predictive model (such as system 1100) may use random forest regression to estimate TTFields. TTFields estimates can be used to determine optimal positions for a transducer array. Random forests are an ensemble of decision tree predictors, with each tree constrained by a random vector that governs the tree's sensitivity to input features. Regression trees have been demonstrated to facilitate effective modeling of FEM-based nonlinear maps for mechanical force fields. Furthermore, random forests partition datasets into groups of similar features, facilitating local group fitting, suggesting that good models can be generated even when data is heterogeneous, irregular, and size-limited.

[0082] The methods, systems, and apparatus described herein for fast approximation of electric field distribution may use a random forest regressor. For example, a 30-tree setup using bootstrap and out-of-bag samples, mean squared error, and segmentation quality measures may be used to estimate regression quality for unseen samples. The number of trees may be selected by a trial-and-error process to balance the trade-off between accuracy and prediction time. The inputs per voxel to the regression tree may be: 1) conductivity (σ), 2) permittivity (ε), 3) distance from the nearest source (d e ), 4) distance from the nearest CSF (d c ), and 5) distance from the TA midline (dl).

[0083] The test is performed by measuring the characteristics (e.g., conductivity (σ), permittivity (ε), distance from the nearest power source (d e ), distance from the nearest CSF (d cThis study demonstrates that the relevance of features such as the distance from the TA midline (dl) and the distance from the TA midline (dl) can be used for prediction in an experimental setup (as described later in this specification) using the mean decreasing impurity method, which results in feature importance scores ranging from 0 to 1. Distance from the nearest power source was by far the most important feature (0.65). Distance from the transducer array (TA) midline and CSF were of secondary importance (0.15 and 0.1, respectively). The importance scores for conductivity and permittivity were both 0.05. Figure 13 shows an example of a random forest decision tree. The mean squared error (MSE) decreased, particularly for locations farther from the TA. The particular tree in this example splits the data at a distance of 16.6 mm to the TA. In fact, larger errors were observed for 15% of the data within this range.

[0084] Additionally, tests were used to compare random forests with multiple regression. Specifically, a linear equation (Equation 2) was incorporated to estimate TTFields.

[0085]

number

[0086] The coefficient a is used to find the best fit between the finite element output and the linear regression model. i was calculated.

[0087] Experimental setup Testing was performed to validate the methods described herein using a dataset of 10 patients who underwent TTFields therapy. First, the patient's MRI was segmented, and the outer surface of the head was extracted using a marching cubes algorithm. Two transducer array (TA) pairs were then virtually placed on the surface of the head. The first pair was placed with one TA on the forehead and the other on the back of the head. In this case, the TTFields were oriented approximately parallel to the anterior-posterior (AP) axis. The second pair was placed with the TA on the opposite side of the head. In this case, the TTFields were oriented approximately parallel to the left-right (LR) axis of the head. For each of the 20 pairs, the absolute field strength spatial distribution was determined using the finite element method. Figure 14 shows exemplary results. The field association for each voxel in the patient's MRI was determined, as shown in Figures 14A-14C. This dataset was validated with imaginary lines and marked as the reference standard because it correlates with patient survival. Figure 14A shows the results of TTFields estimation calculated by standard finite element method, Figure 14B shows the results of TTFields estimation calculated by random forest regression, and Figure 14C shows the results of TTFields estimation calculated by linear regression.

[0088] A leave-one-out approach was used for training: one test patient was left out at a time, and the 18 datasets of the remaining nine patients were incorporated to train a random forest and multiple regression model (Equation 2).

[0089] TTFields were predicted using random forest and multiple regression models on test patient data. A large portion of the image is associated with non-conductive air. This can bias the model results. To address this situation, only a small fraction of voxels with air were considered in training by ensuring that their number was similar to that in other segmented tissues. Training and prediction were performed for each voxel, independent of its neighbors.

[0090] The method may be implemented, for example, in Python 3.6 using the scipy, numpy, scikit-learn, and SimpleITK packages. A 3D slicer was used, for example, for visual inspection of the results. Standard baseline simulations may be calculated using sim4life (Zurich Med Tech, Zurich, Switzerland) on a dedicated simulation computer (e.g., Intel i7 CPU, NVidia 1080 Ti GPU, 128 GB RAM, etc.). The patient's MRI was T1-weighted with gadolinium, with a voxel spacing of 1 × 1 × 1 mm3, and incorporated the entire head.

[0091] result The mean absolute difference between random forest predictions and the standard reference was 0.14 V / cm (patient SD = 0.035, range 0.08–0.23 V / cm, N = 20). Table 2 shows a patient-by-patient summary of our results.

[0092] [Table 2]

[0093] As shown in Table 2, random forest regression results yielded better accuracy compared to multiple regression. Compare the random forest mean absolute difference above with the linear regression result of 0.29 V / cm (SD = 0.04, range 0.23–0.37 V / cm, N = 20 patients). The random forest mean prediction time was 15 seconds (SD = 1.5 seconds). Note that this measure excludes the preprocessing required to extract distance measures, which typically took 30 seconds for each TA, CSF, and midline.

[0094] Figures 14A-14C show typical electric field spatial distributions calculated by standard baseline, random forest, and multiple regression. Figures 15A-15B show typical absolute differences between standard baseline and random forest predictions. Figures 15A-15B show that larger errors were observed near the transducer array (Figure 15A) and near the ventricle along the main transducer array axis (Figure 15B).

[0095] Values ​​were very similar (<0.4 V / cm) in most configurations, however, large errors (>2 V / cm) were observed near the TA (Fig. 4A) and just outside the ventricle along the TA major axis (Fig. 15B).

[0096] As shown and explained, the methods, systems, and devices described herein can be used for fast estimation of TTFields spatial distribution. Results show that an average accuracy of 0.14 V / cm can be achieved in a short time. Compare the computation time of ~1.5 minutes using the proposed random forest method with the computation time of 3-4 hours using standard baseline methods. Note that the computation time can be further reduced by a factor of 3 by parallelizing the data preparation.

[0097] Selection of optimal TA placement involves calculating the average TTFields across the tumor region. Averaging can further improve accuracy. Random forest methods can also be used to estimate TTFields for TA placement optimization.

[0098] The methods described herein may incorporate neighboring voxels and consider convolutional and recurrent neural networks. Rapid manipulation of these maps based on alternating one-time distance map calculations and TA placement can be used to reduce overall computation time to a few seconds. The methods described herein can be used for any part of a patient's body undergoing TTFields therapy.

[0099] In some cases, the fast approximation of the electric field distribution (e.g., the output of the prediction engine in 1290, etc.) may be used to provide one or more candidate locations for transducer array placement on a region of the patient's anatomy (e.g., the head, the torso, etc.). For example, the fast approximation of the electric field distribution may be used to determine one or more electric field distribution strength values ​​for a pair of transducer arrays at each of a plurality of locations on the patient's anatomy. The one or more candidate locations for transducer array placement may be used to determine an optimal placement for the transducer arrays.

[0100] The optimal placement for the transducer array may be determined based on a patient model, a fast approximation of the distributed field strength values ​​for the pair of transducer arrays at each of a plurality of locations, a region of interest (ROI) (e.g., tumor location, etc.), and anatomical constraints associated with the patient's anatomy. The anatomical constraints may indicate one or more locations of a cross-section of the ROI that should be excluded from use in determining the field distribution map. A plane that crosses a portion of the patient's body may be determined based on the center of the ROI.

[0101] FIG. 16A illustrates the definition of a transverse plane 1602 at a patient's head, and FIG. 16B illustrates the transverse plane 1602 as defined at a patient's chest. The transverse plane 1602 may be defined at any part of the patient's body. The transverse plane 1602 may be initially defined by the center and tilt angle of the region of interest 1604. The tilt angle may be defined by one skilled in the art. For example, the tilt angle relative to the head may be 150 to 20 degrees off an axial plane (e.g., horizontal). The transverse plane 1602 may include a contour created by the boundary (e.g., outline) of an anatomical model (e.g., head, chest, torso, abdomen, legs, arms, and the like). For example, the contour may resemble an oval, a circle, an irregular shape, and the like. Determining the transverse plane 1602 may include determining multiple locations 1606 along the contour of the transverse plane 1602. The plurality of locations 1606 may represent one or more locations determined from a fast approximation (e.g., the output of the prediction engine in 1290) of the electric field distribution of a potential transducer array positioned along the cross-section 1602. The fast approximation may be used to determine that the electric field generated by the transducer array at each of the plurality of locations 1606 will pass through the ROI 1604. The plurality of locations 1606 may be determined based on the fast approximation of the electric field distribution (e.g., the output of the prediction engine in 1290). Any number of locations 1606 is contemplated. In some cases, the locations 1606 may be divided into pairs such that the alternating electric fields generated by pairs of transducer arrays (each transducer array disposed at one of the pair of locations) pass through the ROI 1604. The locations 1606 may be spaced apart by, for example, 15 degrees, corresponding to a translation of about 2 cm, giving a total of 12 different locations within a 180-degree range. Other spacings are also contemplated.

[0102] The transverse plane may include a plurality of pairs of positions for the pair of transducer arrays along the contour of the plane. One or more positions of the plurality of pairs of positions may be adjusted based on anatomical constraint parameters to generate the corrected plane. For each combination of the plurality of combinations of two pairs of transducer arrays, a plurality of dose metrics may be determined within the ROI based on the electric field distribution map. The plurality of dose metrics may be based on simulated electric fields generated for each combination of the plurality of combinations of two pairs of transducer arrays.

[0103] In some cases, one or more candidate transducer array layout floorplans may be determined based on the angle constraint parameter and the plurality of dose metrics within the ROI. In some cases, the angle constraint parameter may indicate an orthogonal angle between the plurality of pairs of transducer arrays. In some cases, the angle constraint parameter may indicate an angle range between the plurality of pairs of transducer arrays. For optimization, the candidate transducer array layout floorplans may be compared, correlated, and / or matched with one or more estimated electric field distribution intensity values ​​for the pair of transducer arrays at each of the plurality of locations.

[0104] For each of the one or more candidate transducer array layout floorplans specifically compared, associated, and / or matched with the estimated one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the multiple locations, one or more adjusted candidate transducer array layout floorplans may be determined by adjusting the position or orientation of the one or more transducer arrays of the pair of transducer arrays. For each adjusted candidate transducer array layout floorplan, there is an adjusted dose metric within the ROI. Based on the adjusted dose metric within the ROI, a final transducer array layout floorplan may be determined from the adjusted candidate transducer array layout floorplan. In some cases, the accuracy of the predictive model may be readjusted (e.g., optimized, etc.) through further training if the final transducer array layout floorplan does not correlate with at least a portion of the one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the multiple locations.

[0105] After a three-dimensional array layout map (e.g., a final transducer array layout floor plan) has been determined for the patient, the application of TTFields may then be simulated by a patient modeling application 608 using the patient model. The simulated field distributions, dosimetry, and simulation-based analysis are described in U.S. Patent Application Publication No. 20190117956A1 and in the publication by Ballo et al., "Correlation of Tumor Treating Fields Dosimetry to Survival Outcomes in Newly Diagnosed Glioblastoma: A Large-Scale Numerical Simulation-based Analysis of Data from the Phase 3 EF-14 Randomized Trial" (2019), which are incorporated herein by reference in their entireties.

[0106] To ensure systematic positioning of the transducer array relative to the tumor location, a reference coordinate system can be defined. For example, the transverse plane can be first defined by conventional LR and AP positioning of the transducer array. The left-right direction can be defined as the x-axis, the AP direction as the y-axis, and the craniocaudal direction, which is normal to the XY plane, as the z-axis.

[0107] After defining the coordinate system, the transducer arrays can be virtually placed on a patient model with its center and longitudinal axis on the XY plane. The pair of transducer arrays can be systematically rotated from 0 to 180 degrees around the z-axis of the head model, i.e., in the XY plane, thereby covering the entire circumference of the head (due to symmetry). The rotation interval can be, for example, 15 degrees, corresponding to a translation of about 2 cm, giving a total of 12 different positions within the 180-degree range. Other rotation intervals are also contemplated. Electric field distribution calculations can be performed for the position of each transducer array relative to the tumor coordinates.

[0108] The electric field distribution in the patient model can be determined by the patient modeling application 608 using a finite element (FE) approximation of the electric potential. Generally, quantities defining a time-varying electromagnetic field are given by complex Maxwell's equations. However, in biological tissue, at the low to mid-frequency range of TTFields (f = 200 kHz), the wavelength of the electromagnetic wave is much larger than the head size, and the permittivity ε is negligibly small compared to the real-valued electrical conductivity σ, i.e., ω = 2πf is the angular frequency. This means that electromagnetic wave propagation effects and capacitive effects in tissue are negligible, and therefore the scalar electric potential can be well approximated by the static Laplace equation ∇·(σ∇φ) = 0 under appropriate boundary conditions at the electrodes and skin. Therefore, complex impedances are treated as resistive (i.e., reactance is negligible), and therefore, the current flowing in volume conductors is primarily free (ohmic) current. The FE approximation of the Laplace equation was calculated using SimNIBS software (simnibs.org). Calculations were based on the Galerkin method, and the residuals of the conjugate gradient solver were required to be <1E-9. Dirichlet boundary conditions were used, and the potential was set to a fixed (arbitrarily chosen) value for each set of electrode arrays. The electric field (vector field) was calculated as the numerical gradient of the potential, and the current density (vector field) was calculated from the electric field using Ohm's law. The electric field values, potential difference and current density, were linearly rescaled to a peak-to-peak total amplitude of 1.8 A for each array pair and calculated as the (numerical) surface integral of the normal current density component over all triangular surface elements on the active electrode disk. This corresponds to the current level used for clinical TTFields therapy with the Optune® device. TTFields "dose" was calculated as the strength (L2 norm) of the electric field vector. The modeled current is assumed to be supplied by two separate, sequentially activated sources, each connected to a pair of 3 × 3 transducer arrays. In the simulations, the left and rear arrays were defined as sources, and the right and front arrays were defined as corresponding sinks.However, since TTFields employs an alternating electric field, this choice is arbitrary and does not affect the results.

[0109] The average strength of the electric field generated by the transducer array placed in multiple locations on the patient may be determined by the patient modeling application 608 for one or more tissue types. In one embodiment, the transducer array location corresponding to the highest average electric field strength in the tumor tissue type may be selected as the desired (e.g., optimal) transducer array location for the patient. In another embodiment, one or more candidate locations for the transducer array may be eliminated as a result of the patient's health condition. For example, one or more candidate locations may be eliminated based on areas of skin irritation, scarring, surgery sites, discomfort, etc. Thus, after eliminating one or more candidate locations, the transducer array location corresponding to the highest average electric field strength in the tumor tissue type may be selected as the desired (e.g., optimal) transducer array layout location for the patient. As such, a transducer array location that results in less than the maximum possible average electric field strength may be selected.

[0110] The patient model may be modified to include indications of desired transducer array locations. The resulting patient model, including indications of desired transducer array locations, may be referred to as a three-dimensional array layout map (e.g., three-dimensional array layout map 600). Thus, a three-dimensional array layout map may include a digital representation in three-dimensional space of a portion of a patient's body, indications of tumor placement, location instructions for placement of one or more transducer arrays, combinations thereof, and the like.

[0111] The three-dimensional array layout map may be provided to the patient in digital and / or physical form, and the patient and / or their caregiver may use the three-dimensional array layout map to apply one or more transducer arrays to relevant parts of the patient's body (e.g., the head).

[0112] 17 is a block diagram illustrating an environment 1700 including a non-limiting example of a patient support system 1702. In one aspect, some or all steps of any described method may be performed on a computing device as described herein. The patient support system 1702 may include one or more computers configured to store one or more of the EFG configuration application 606, the patient modeling application 608, the image data 610, and the like.

[0113] In terms of hardware architecture, the patient support system 1702 may be a digital computer generally including a processor 1709, a memory system 1720, an input / output (I / O) interface 1712, and a network interface 1714. These components (606, 608, 610, 1718, and 1722) are communicatively coupled via a local interface 1716. The local interface 1716 may be, for example, but not limited to, one or more buses or other wired or wireless connections as known in the art. The local interface 1716 may have additional elements, omitted for simplicity, such as controllers, buffers (caches), drivers, repeaters, and receivers to enable communication. Additionally, the local interface may include address, control, and / or data connections to enable appropriate communication between the aforementioned components.

[0114] The processor 1709 can be a hardware device for executing software, particularly software stored in the memory system 1720. The processor 1709 can be any custom or commercially available processor, a central processing unit (CPU), a coprocessor among multiple processors associated with the patient support system 1702, a semiconductor-based microprocessor (taking the form of a microchip or chipset), or generally any device for executing software instructions. When the patient support system 1702 is operating, the processor 1709 can be configured to execute software stored in the memory system 1720, communicate data to and from the memory system 1720, and generally control the operation of the patient support system 1702 in accordance with the software.

[0115] The I / O interface 1712 may be used to receive user input and / or provide system output from one or more devices or components. User input may be provided, for example, via a keyboard and / or a mouse. System output may be provided via a display device and a printer (not shown). The I / O interface 1712 may include, for example, a serial port, a parallel port, a Small Computer System Interface (SCSI), an IR interface, an RF interface, and / or a Universal Serial Bus (USB) interface.

[0116] The network interface 1714 may be used to transmit and receive from the patient support system 1702. The network interface 1714 may include, for example, a 10BaseT Ethernet adapter, a 100BaseT Ethernet adapter, a LAN PHY Ethernet adapter, a token ring adapter, a wireless network adapter (e.g., WiFi), or any other suitable network interface device. The network interface 1714 may include address, control, and / or data connections to enable appropriate communications.

[0117] The memory system 1720 can include any one or combination of volatile memory elements (e.g., random access memory (RAM, such as DRAM, SRAM, SDRAM)) and non-volatile memory elements (e.g., ROM, hard drives, tape, CD-ROM, DVD-ROM, etc.). Additionally, the memory system 1720 can incorporate electronic, magnetic, optical, and / or other types of storage media. It should be noted that the memory system 1720 can have a distributed architecture in which various components are located remotely from each other but can be accessed by the processor 1709.

[0118] The software in the memory system 1720 may include one or more software programs, each of which includes an ordered listing of executable instructions for implementing logical functions. In the example of Figure 17, the software in the memory system 1720 of the patient support system 1702 may include an EFG configuration application 606, a patient modeling application 608, image data 610, and a suitable operating system (O / S) 1718. The operating system 1718 essentially controls the execution of other computer programs and provides scheduling, input / output control, file and data management, memory management, and communication control and related services.

[0119] For purposes of explanation, application programs and other executable program components, such as the operating system 1718, are illustrated herein as separate blocks, with the understanding that such programs and components may reside at various times in different storage components of the patient support system 1702. An implementation of the EFG configuration application 1706, the patient modeling application 608, the image data 610, and / or the control software 1720 may be stored on or transmitted via some form of computer-readable media. Any of the disclosed methods may be performed by computer-readable instructions embodied on a computer-readable medium. A computer-readable medium may be any available medium that can be accessed by a computer. For example, but not limited to, computer-readable media may include “computer storage media” and “communications media.” “Computer storage media” may include volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information, such as computer-readable instructions, data structures, program modules, or other data. Exemplary computer storage media may include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage devices, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by a computer.

[0120] In the embodiment illustrated in FIG. 18 , one or more of the apparatus 100, the patient support system 602, the patient modeling application 608, the patient support system 1702, and / or any other devices / components described herein can be configured to perform a method 1800 that includes determining 1810 a plurality of sets of image data associated with a plurality of patients, each patient associated with a set of image data derived from imaging a portion of the patient, each set of image data including a plurality of voxels, each voxel of the plurality of voxels labeled with a tissue type, and each voxel of the plurality of voxels labeled with an electric field strength distribution value (Vcm) derived from simulated application of alternating electric fields from a pair of transducer arrays to the portion of the patient. For example, determining the multiple sets of image data may include determining RAW image data for each patient, the RAW image data including a plurality of voxels; assigning a tissue type to each of the plurality of voxels; simulating the application of alternating electric fields from pairs of transducer arrays at a plurality of locations within the RAW image data based on the tissue type of each voxel; labeling each of the plurality of voxels of the RAW image data with a tissue type, a simulated electric field, and a location associated with the simulated electric field; and generating the multiple sets of image data based on the labeled RAW data.

[0121] At 1820, a plurality of features for the predictive model are determined based on the first portion of the plurality of sets of image data. For example, determining a plurality of features for the predictive model based on the first portion of the plurality of sets of image data may include feature selection techniques including one or more of a filter method, a wrapper method, or an embedding method. The plurality of features may include two or more of a conductivity value, a permittivity value, a distance to the nearest transducer array, a distance to the nearest conductive material, and / or the like.

[0122] At 1830, a predictive model is trained based on the first portion of the plurality of features and the plurality of sets of image data, where the predictive model is configured to estimate field strength distribution values. For example, training the predictive model based on the first portion of the plurality of features and the plurality of sets of image data may include machine learning techniques including one or more of discriminant analysis, decision trees, nearest neighbor (NN) algorithms (e.g., k-NN models, replicator NN models, etc.), statistical algorithms (e.g., Bayesian networks, etc.), clustering algorithms (e.g., k-means, mean shift, etc.), neural networks (e.g., reservoir networks, artificial neural networks, etc.), support vector machines (SVMs), logistic regression algorithms, linear regression algorithms, Markov models or Markov chains, principal component analysis (PCA) (e.g., for linear models), multilayer perceptron (MLP) ANNs (e.g., for nonlinear models), replicated reservoir networks (e.g., for nonlinear models, typically for time series), random forest classification, combinations thereof, and / or the like.

[0123] At 1840, the predictive model is tested based on a second portion of the sets of image data.

[0124] Based on the test results, a predictive model is output at 1850. For example, the predictive model may be used for fast approximation of the electric field distribution.

[0125] In some cases, method 1800 may also include determining a new set of image data for the new patient, the new set of image data including a plurality of voxels; submitting the new image dataset to a predictive model; and estimating, for each voxel of the plurality of voxels, one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the plurality of locations using the predictive model. In some cases, method 1800 may also include selecting one of the plurality of locations based on the estimated one or more electric field distribution intensity values. In some cases, method 1800 may also include determining a tissue type for each voxel of the plurality of voxels of the new image dataset, and determining a conductivity value, a permittivity value, a distance to the nearest transducer array, and a distance to the nearest conductive material for each voxel of the plurality of voxels of the new image dataset.

[0126] In one embodiment illustrated in FIG. 19 , one or more of the apparatus 100, the patient support system 602, the patient modeling application 608, the patient support system 1702, and / or any other devices / components described herein can be configured to perform a method 1910 that includes determining, at 1910, a set of image data for a patient, the set including a plurality of voxels.

[0127] At 1920, the image data set is presented to a predictive model configured to estimate field strength distribution values ​​based on one or more simulated alternating electric fields from pairs of transducer arrays at multiple locations.

[0128] At 1930, the predictive model estimates, for each voxel of the plurality of voxels, one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the plurality of locations.

[0129] In some cases, method 1900 may also include selecting one of the plurality of locations based on the estimated one or more electric field distribution intensity values. In some cases, method 1900 may also include determining a tissue type for each voxel of the plurality of voxels of the image dataset, and determining a conductivity value, a permittivity value, a distance to the nearest transducer array, and a distance to the nearest conductive material for each voxel of the plurality of voxels of the image dataset.

[0130] At 1940, a transducer array map may be determined. For example, an optimal placement for the transducer arrays may be determined based on the set of image data, one or more estimated field distribution intensity values ​​for the pair of transducer arrays at each of a plurality of locations, a region of interest (ROI) (e.g., tumor location, etc.), and anatomical parameters of the patient. The image dataset may be used to determine a three-dimensional (3D) model of a portion of the patient's body. The electric field distribution map may be determined for each of a plurality of locations for the pair of transducer arrays based on the 3D model, the ROI, and anatomical constraints associated with the patient, and based on the one or more estimated field distribution intensity values ​​for the pair of transducer arrays at each of the plurality of locations.

[0131] The anatomical constraint may indicate one or more locations of a cross section of the ROI that should be excluded from use in determining the electric field distribution map. In some cases, the method may include determining a plane that crosses a portion of the patient's body based on a center of the ROI, the plane including multiple pairs of positions for pairs of transducer arrays along the contour of the plane, and adjusting one or more positions of the multiple pairs of positions based on the anatomical constraint parameters to generate a corrected plane. For each combination of the multiple combinations of two pairs of transducer arrays, multiple dose metrics within the ROI may be determined based on the electric field distribution map. The multiple dose metrics may be based on simulated electric fields generated for each combination of the multiple combinations of two pairs of transducer arrays.

[0132] In some cases, the method may include determining one or more candidate transducer array layout floorplans based on an angular constraint parameter and a plurality of dose metrics within the ROI. In some cases, the angular constraint parameter may indicate an orthogonal angle between the plurality of pairs of transducer arrays. In some cases, the angular constraint parameter may indicate an angular range between the plurality of pairs of transducer arrays. For optimization, the candidate transducer array layout floorplans may be compared, correlated, and / or matched with one or more estimated electric field distribution intensity values ​​for the pairs of transducer arrays at each of the plurality of locations.

[0133] For each of the one or more candidate transducer array layout floorplans specifically compared, associated, and / or matched with the estimated one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the multiple locations, one or more adjusted candidate transducer array layout floorplans may be determined by adjusting the position or orientation of the one or more transducer arrays of the pair of transducer arrays. For each adjusted candidate transducer array layout floorplan, an adjusted dose metric within the ROI. Based on the adjusted dose metric within the ROI, a final transducer array layout floorplan may be determined from the adjusted candidate transducer array layout floorplan. In some cases, the accuracy of the predictive model may be readjusted (e.g., optimized, etc.) through further training if the final transducer array layout floorplan does not correlate with at least a portion of the one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the multiple locations.

[0134] In view of the described apparatus, systems, and methods, and variations thereof, set forth herein below are some more particularly described embodiments of the present invention. However, these particularly described embodiments should not be construed as having any restrictive effect on any different claims that include different or more general teachings set forth herein, or as limiting the "particular" embodiments in any way other than to the inherent meaning of the word as literally used therein.

[0135] Embodiment 1: A method comprising: determining a plurality of sets of image data associated with a plurality of patients, each patient associated with a set of image data derived from imaging a portion of the patient, each set of image data comprising a plurality of voxels, each voxel of the plurality of voxels labeled with a tissue type and each voxel of the plurality of voxels labeled with an electric field strength distribution value (Vcm) derived from simulated application of an alternating electric field from a pair of transducer arrays to the portion of the patient; determining a plurality of features for a predictive model based on a first portion of the plurality of sets of image data; training the predictive model based on the plurality of features and the first portion of the plurality of sets of image data, the predictive model being configured to estimate the electric field strength distribution value; testing the predictive model based on a second portion of the plurality of sets of image data; and outputting the predictive model based on test results.

[0136] Embodiment 2: An embodiment of any one of the preceding embodiments, wherein determining the multiple sets of image data includes determining RAW image data for each patient, the RAW image data including a plurality of voxels; assigning a tissue type to each voxel of the multiple voxels; simulating application of alternating electric fields from pairs of transducer arrays at a plurality of locations in the RAW image data based on the tissue type of each voxel; labeling each voxel of the multiple voxels of the RAW image data with a tissue type, a simulated electric field, and a location associated with the simulated electric field; and generating the multiple sets of image data based on the labeled RAW data.

[0137] Embodiment 3: An embodiment described in any one of the preceding embodiments, wherein determining a plurality of features for the predictive model based on a first portion of the plurality of sets of image data includes feature selection techniques including one or more of a filter method, a wrapper method, or an embedding method.

[0138] Embodiment 4: An embodiment of any one of the preceding embodiments, wherein the plurality of features includes two or more of a conductivity value, a permittivity value, a distance to the nearest transducer array, or a distance to the nearest conductive material.

[0139] Embodiment 5: An embodiment of any one of the preceding embodiments, wherein training the predictive model based on the first portion of the plurality of sets of features and image data comprises machine learning techniques including one or more of discriminant analysis, decision trees, nearest neighbor (NN) algorithms (e.g., k-NN models, replicator NN models, etc.), statistical algorithms (e.g., Bayesian networks, etc.), clustering algorithms (e.g., k-means, mean shift, etc.), neural networks (e.g., reservoir networks, artificial neural networks, etc.), support vector machines (SVMs), logistic regression algorithms, linear regression algorithms, Markov models or chains, principal component analysis (PCA) (e.g., for linear models), multilayer perceptron (MLP) ANNs (e.g., for nonlinear models), replicated reservoir networks (e.g., for nonlinear models, typically for time series), or random forest classification.

[0140] Embodiment 6: An embodiment of any one of the preceding embodiments, further comprising: determining a new set of image data for a new patient, the new set of image data including a plurality of voxels; submitting the new image dataset to a predictive model; and estimating, by the predictive model, for each voxel of the plurality of voxels, one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of a plurality of locations.

[0141] Embodiment 7: An embodiment as described in any one of the preceding embodiments, further comprising selecting one location from the plurality of locations based on the estimated one or more electric field distribution strength values.

[0142] Embodiment 8: An embodiment of any one of the preceding embodiments, further comprising determining a tissue type for each voxel of the plurality of voxels of the new image dataset, and determining a conductivity value, a permittivity value, a distance to the nearest transducer array, and a distance to the nearest conductive material for each voxel of the plurality of voxels of the new image dataset.

[0143] Embodiment 9: The embodiment of any one of the preceding embodiments, further comprising using a predictive model.

[0144] Embodiment 10: A method comprising: determining a set of image data for a patient, the set of image data including a plurality of voxels; submitting the image dataset to a predictive model, the predictive model configured to estimate electric field intensity distribution values ​​based on one or more simulated alternating electric fields from pairs of transducer arrays at a plurality of locations; estimating, for each voxel of the plurality of voxels, one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the plurality of locations using the predictive model; and determining a transducer array map including one or more of the plurality of locations based on the estimated one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of the plurality of locations, a region of interest, and anatomical constraints associated with the patient.

[0145] Embodiment 11: The embodiment of embodiment 10, further comprising selecting one location from the plurality of locations based on the estimated one or more electric field distribution strength values.

[0146] Embodiment 12: An embodiment described in embodiments 10 to 11, further comprising determining a tissue type for each voxel of the plurality of voxels of the image dataset, and determining a conductivity value, a permittivity value, a distance to the nearest transducer array, and a distance to the nearest conductive material for each voxel of the plurality of voxels of the image dataset.

[0147] Unless otherwise specified, it is in no way intended that any method set forth herein be construed as requiring that its steps be performed in a particular order. Thus, if a method claim does not actually recite the order in which its steps should be followed, or is not otherwise specifically stated in the claim or description that the steps are to be limited to a particular order, no order is intended to be inferred in any way. This applies to any possible implicit basis for interpretation, including questions of logic regarding the sequence of steps or flow of operations, the apparent meaning derived from grammatical construction or punctuation, and the number or type of embodiments described in the specification.

[0148] While the methods and systems have been described in connection with preferred embodiments and specific examples, the scope is not intended to be limited to the particular embodiments described, as the embodiments herein are intended in all respects to be illustrative rather than restrictive.

[0149] It will be apparent to those skilled in the art that various modifications and variations can be made without departing from the scope or spirit. Other embodiments will be apparent to those skilled in the art from consideration of the specification and practice disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit being indicated by the appended claims. [Explanation of symbols]

[0150] 100 devices 102 Electric Field Generator 104, 104a, 104b Transducer array 106 processors 108 Signal Generator 110 Control Software 112 Conductive Lead 114 Output 116 Electrode 118 Flexible circuit board 120 Hypoallergenic Medical Bandages 120a, 120b Hypoallergenic medical adhesive bandages 302 Skin surface 304 Tumor 306 Bone tissue 308 Brain Tissue 310 Alternating Electric Field 600 System 602 Patient Support System 606 Electric Field Generator (EFG) Configuration Applications 608 Patient Modeling Applications 610 Image Data 800 Three-dimensional Array Layout Map 1100 System 1110A, 1110N training dataset 1120 Training Module 1130 Machine Learning Based Classifiers 1140 Machine Learning-Based Classification Models 1140A, 1140N Machine learning based classification model 1200 Training method 1602 cross section 1604 Region of Interest (ROI) 1606 position 1700 Environment 1702 Patient Support System 1706 EFG Configuration Application 1709 processor 1712 Input / Output (I / O) Interface 1714 Network Interface 1716 Local Interface 1718 Operating System (O / S) 1720 Memory System 1722 Control Software

Claims

1. Determining a plurality of sets of image data associated with a plurality of patients, each patient associated with a set of image data derived from imaging a portion of the patient, each set of image data including a plurality of voxels, each voxel of the plurality of voxels labeled with a tissue type, each voxel of the plurality of voxels representing a field strength distribution value (Vcm) derived from simulated application of alternating electric fields from a pair of transducer arrays to the portion of the patient. -1 ), and determining a plurality of features for a predictive model based on a first portion of the plurality of sets of image data; training the predictive model based on the plurality of features and the first portion of the plurality of sets of image data, the predictive model being configured to estimate a field strength distribution value; testing the predictive model based on a second portion of the plurality of sets of image data; outputting the predictive model based on the test results; A method comprising: The method, wherein the plurality of features includes a distance to a nearest transducer array and a distance to a nearest conductive material.

2. The step of determining a plurality of sets of image data comprises: determining raw image data for each patient, the raw image data including a plurality of voxels; assigning a tissue type to each voxel of the plurality of voxels; simulating the application of alternating electric fields from pairs of transducer arrays at a plurality of locations within the raw image data based on the tissue type of each voxel; labeling each voxel of the plurality of voxels of the raw image data with the tissue type, the simulated electric field, and the location associated with the simulated electric field; generating the plurality of sets of image data based on the labeled raw image data; 2. The method of claim 1, comprising:

3. 2. The method of claim 1, wherein determining a plurality of features for a predictive model based on a first portion of the plurality of sets of image data comprises feature selection techniques including one or more of a filter method, a wrapper method, or an embedding method.

4. The method of claim 1 , wherein the plurality of features further comprises one or more of a conductivity value, a permittivity value.

5. 10. The method of claim 1, wherein training the predictive model based on the plurality of features and the first portion of the plurality of sets of image data comprises machine learning techniques including one or more of discriminant analysis, decision trees, statistical algorithms, or neural networks.

6. determining a new set of image data for a new patient, the new set of image data including a plurality of voxels; submitting the new set of image data to the predictive model; estimating, for each voxel of the plurality of voxels, one or more electric field distribution intensity values ​​for the pair of transducer arrays at each of a plurality of locations using the predictive model; The method of claim 1 further comprising:

7. selecting one of the plurality of locations based on the estimated one or more electric field distribution strength values; 7. The method of claim 6, further comprising:

8. 1. An apparatus comprising: one or more processors; a memory for storing processor-executable instructions; Equipped with 8. An apparatus, wherein the instructions, when executed by the one or more processors, cause the apparatus to perform the method of any one of claims 1 to 7.

9. One or more non-transitory computer-readable media having stored thereon processor-executable instructions that, when executed by a processor, cause the processor to perform the method of any one of claims 1 to 7.

Images