Reconstruction of truncated medical images

Abstract

A method for reconstructing a truncated medical image includes accessing a medical image of a subject from a memory, the medical image including voxels; determining that the medical image has a truncated portion of the subject, the voxels of the truncated portion having non-tissue image values; generating a reconstructed medical image using a trained machine learning model and the medical image, the trained machine learning model trained to generate a reconstructed medical image without truncation, the voxels in the truncated portion of the reconstructed medical image having tissue image values.

Description

[0001] Cross-reference to related applications This application claims priority to U.S. Provisional Patent Application No. 63 / 613,796, filed December 22, 2023, the entire contents of which are incorporated herein by reference. Background Technology

[0002] A tumor therapeutic electric field (TTField) is a low-intensity alternating electric field in the mid-frequency range (e.g., 50 kHz to 1 MHz), as described in U.S. Patent No. 7,565,205, which can be used to treat tumors. The TTField is non-invasively sensed into a region of interest by placing transducers on the patient's body and applying an alternating current (AC) voltage between the transducers. Traditionally, a first pair of transducers and a second pair of transducers are placed on the subject's body. An AC voltage is applied between the first pair of transducers during a first time interval to generate an electric field with field lines extending generally in a front-back direction. Then, during a second time interval, an AC voltage is applied between the second pair of transducers at the same frequency to generate an electric field with field lines extending generally in a left-right direction. The system then repeats this two-step sequence throughout the treatment process. Attached Figure Description

[0003] Figure 1 This is a flowchart depicting an example method for training a machine learning model to reconstruct truncated medical images.

[0004] Figure 2A and Figure 2B (Collectively referred to as Figure 2) is a flowchart depicting an example method for reconstructing truncated medical images and using the reconstructed medical images to select a transducer for delivering an alternating electric field to a subject.

[0005] Figure 3 This is a flowchart depicting an example method for determining whether a medical image has a truncated portion of a subject.

[0006] Figure 4A and Figure 4B These are flowcharts illustrating example methods for determining whether a medical image contains a truncated portion of a subject.

[0007] Figure 5 This is an example medical image with a truncated portion.

[0008] Figure 6A , Figure 6B and Figure 6C An example of a medical image processed according to an exemplary embodiment is depicted.

[0009] Figure 6D , Figure 6E and Figure 6F An example of a medical image processed according to an exemplary embodiment is depicted.

[0010] Figure 7 An example system for applying an alternating electric field to a subject is described.

[0011] Figure 8 An example placement of a transducer located on the subject's head is depicted.

[0012] Figure 9 An example computer apparatus for use with the embodiments described herein is depicted.

[0013] Various embodiments are described in detail below with reference to the accompanying drawings, wherein the same reference numerals denote the same elements. Detailed Implementation

[0014] This application describes exemplary techniques for computationally reconstructing truncated medical images and provides one or more recommended transducer layouts for applying TTFields to a subject based on the reconstructed medical images. As an example, the truncated medical image may be generated by generating a medical image of the subject's torso, and the subject's left and right sides may be truncated in the medical image.

[0015] Previously, truncated medical images might have been acceptable when considering radiation therapy plans for treating a subject's tumor. However, for TTField treatment plans, truncated medical images may be unacceptable, as these plans may require knowledge of the conductivity of certain sites on the subject to determine the energy flow used in the TTField treatment plan. If a portion of the subject is missing due to truncation in the medical image (i.e., a truncated medical image), the conductivity of that missing portion cannot be known, and therefore the energy flow in that missing portion cannot be determined, resulting in an inability to provide an accurate TTField treatment plan. If a subject has only one medical image, and that image is truncated, the subject will need to reschedule for an untruncated image, thus delaying TTField treatment and increasing its cost. If a subject has multiple medical images, including a truncated image with particularly relevant information, that truncated image may not be used in the TTField treatment plan, resulting in the loss of that particularly relevant information.

[0016] To alleviate these problems, the inventors have discovered a computational technique for reconstructing truncated medical images based on a trained machine learning model to obtain reconstructed medical images, which can then be used for TTField treatment planning. Therefore, medical images with truncated portions that might previously be unusable for TTField treatment planning can now be used by employing the technique of this invention. The technique of this invention is particularly integrated into practical applications that provide reconstructed medical images that can be used for further computational processing techniques, such as generating one or more recommended transducer layouts for applying TTField to a subject based on the reconstructed medical images. Utilizing the technique of this invention, the process of generating reconstructed medical images can produce faster, cheaper, and / or more accurate TTField treatment plans.

[0017] In some embodiments, the technique of the present invention includes training a machine learning model using multiple existing medical images of multiple subjects to obtain a trained machine learning model that generates reconstructed medical images from truncated medical images. Using the reconstructed medical images generated by the technique of the present invention, multiple transducer layouts for applying a TTField to a subject can be generated, and one or more of these transducer layouts can be recommended for applying a TTField treatment to the subject.

[0018] Figure 1 An example method 100 for training a machine learning model to reconstruct truncated medical images is described. Certain steps of method 100 are described as computer-implemented steps. The computer may include one or more processors and memory accessible by the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the computer to perform the relevant steps of method 100. Method 100 may be modified, added to, or omitted. Although for illustrative purposes... Figure 1 The instructions specify the order of operations, but the timing and order of such operations can be changed where appropriate without negating the purpose and advantages of the examples detailed herein.

[0019] refer to Figure 1At step 102, method 100 may include accessing multiple medical images of multiple subjects. In some embodiments, the medical images may be stored locally or via a network in computer memory and accessible from that computer memory. The multiple medical images may include at least one of: magnetic resonance imaging (MRI) medical images, computed tomography (CT) medical images, or positron emission tomography (PET) medical images. The medical images may include multiple voxels and multiple slices, wherein each slice has multiple voxels. These medical images may belong to the same part of each of the subjects. In some embodiments, the same part of each of the subjects in the multiple medical images may be the torso and / or head of the corresponding subject. As an example, the multiple medical images may belong to the torso of each of the multiple subjects. In some embodiments, step 102 may be computer-implemented.

[0020] At step 104, method 100 may include processing the medical images by truncating a portion of each medical image to obtain truncated medical images. As an example, the truncated portion of each medical image may correspond to the right and left sides of each subject. Figure 5 An example medical image with a truncated portion is shown, the truncated portion including a right side 504 and a left side 502. Other portions of each subject may be truncated. The truncated portion of the subject may include voxels with non-tissue image values. In some embodiments, the non-tissue image values ​​of the truncated portion may be the same image values. In some embodiments, the non-tissue image values ​​of the truncated portion may correspond to background image values. In some embodiments, step 104 may be computer-implemented. In some embodiments, step 104 may be implemented based on user input, such as specifying a portion of each medical image to be truncated, or such as specifying a portion of one or several medical images to be truncated, wherein the computer performs the truncation of the medical image based on the user input.

[0021] At step 106, method 100 may include designating a set of truncated medical images as training truncated medical images. In some embodiments, step 106 may be computer-implemented. In some embodiments, step 106 may be implemented based on user input, such as identifying which truncated medical images and / or how many truncated medical images should be designated as training truncated medical images.

[0022] At step 108, method 100 may include training a machine learning model to obtain a trained machine learning model, wherein the machine learning model is trained using training truncated medical images, and wherein the machine learning model is trained to generate untruncated reconstructed medical images. In some embodiments, step 108 may be computer-implemented. Before feeding the training truncated medical images to the machine learning model, a quality test may be performed on the training truncated medical images, for example by the human eye or by calculating the loss value of the network itself, to help provide better training quality.

[0023] A machine learning model may include one or more algorithms and / or architectures. In some embodiments, the machine learning model may be an unsupervised machine learning model. In some embodiments, the trained machine learning model may be a deep learning neural network. In some embodiments, the machine learning model may include at least one of the following: generative adversarial networks (GANs), MedGAN, super-resolution GAN, pix2pix GAN, cycleGAN, discoGAN, fila-sGAN, projection adversarial networks (PANs), variational autoencoders (VAEs), or unsupervised neural networks. As an example, the machine learning model may be a GAN comprising a deconvolutional neural network as a generator and a convolutional neural network as a discriminator.

[0024] At step 110, method 100 may include designating a set of truncated medical images as test truncated medical images. The test truncated medical images may differ from the training truncated medical images. In some embodiments, each truncated medical image may be identified as a training truncated medical image in step 106 or as a test truncated medical image in step 110. In some embodiments, step 110 may be computer-implemented. In some embodiments, step 110 may be implemented based on user input, such as identifying which truncated medical images and / or how many truncated medical images should be designated as test truncated medical images. Quality testing may also be applied to the test truncated medical images.

[0025] At step 112, method 100 may include generating a reconstructed medical image using a trained machine learning model and a test truncated medical image. In some embodiments, step 112 may be computer-implemented.

[0026] At step 114, method 100 may include comparing the reconstructed medical image with a medical image corresponding to a truncated training medical image to obtain a comparison result. In some embodiments, step 114 may be computer-implemented.

[0027] At step 116, method 100 may include determining whether the comparison result is satisfactory or unsatisfactory. In some embodiments, the comparison can be determined to be satisfactory if the reconstructed medical image and the medical image corresponding to the truncated medical image used for training have the same TTField treatment dose and / or the same ranking of the transducer layout for applying the TTField. After determining that the comparison result is satisfactory, the process proceeds to step 118.

[0028] If the comparison result is deemed unsatisfactory, the process returns to repeat steps 108 through 114 to further train the machine learning model and obtain a retrained machine learning model. Specifically, this retraining may involve using updated training data with truncated medical images. In some embodiments, step 116 may be computer-implemented.

[0029] At step 118, method 100 may include obtaining a trained machine learning model. The trained machine learning model is obtained once the comparison result is determined to be satisfactory. In some embodiments, step 118 may be computer-implemented.

[0030] Figure 2A and Figure 2B An example method 200 is depicted for reconstructing truncated medical images and using the reconstructed medical images to select a transducer for delivering an alternating electric field to a subject. Certain steps of method 200 are described as computer-implemented steps. The computer may include one or more processors and a memory accessible by the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the computer to perform the relevant steps of method 200. Method 100 may be modified, added to, or omitted. Although the sequence of operations is indicated in Figure 2 for illustrative purposes, the timing and sequence of such operations may be changed where appropriate without negating the purpose and advantages of the examples detailed herein.

[0031] refer to Figure 2A At step 202, method 200 may include: accessing a medical image of the subject, wherein the medical image includes voxels. In some embodiments, the medical image may be stored locally or via a network in computer memory and accessible from the computer memory. In some embodiments, the medical image of the subject may include an image of the subject's torso and / or head. As an example, the medical image may include the subject's lungs. In some embodiments, step 202 may be computer-implemented.

[0032] At step 204, method 200 may include determining whether the medical image has a truncated portion of the subject. Figure 5An example medical image with a truncated portion is shown. The truncated portion of the subject may include voxels with non-tissue image values. In some embodiments, the non-tissue image values ​​of the truncated portion may be the same image values. In some embodiments, the non-tissue image values ​​of the truncated portion may correspond to background image values.

[0033] In some embodiments, step 204 may be computer-implemented. In some embodiments, the medical image may be identified as having a truncated portion of the subject after determining that the subject depicted in the medical image has at least two flat surfaces. In some embodiments, the medical image may be identified as having a truncated portion of the subject after determining that the subject depicted in the medical image has at least one flat surface. In some embodiments, the determination may be based on a determined perimeter of the subject in the medical image. In some embodiments, step 204 may be performed manually by a user or based on input from a user. As an example, the determination may include presenting the medical image on a display and receiving user input identifying the medical image as having a truncated portion of the subject. Figure 3 and Figure 4A and Figure 4B Additional details regarding step 204 are illustrated in the accompanying description. After determining that the medical image contains a truncated portion of the subject, the process proceeds to step 206. Otherwise, the process proceeds to step 208.

[0034] At step 206, method 200 may include generating a reconstructed medical image using a trained machine learning model and a medical image. The voxels of the reconstructed medical image in the truncated portion have tissue image values. The trained machine learning model is trained to generate an untruncated reconstructed medical image. This can be based on... Figure 1 Method 100 obtains a trained machine learning model.

[0035] At step 208, method 200 may include determining whether another medical image of the subject is available. If yes, the process returns to repeat steps 202 through 206. If no, the process continues to step 210.

[0036] At step 210, method 200 may include defining a region of interest (ROI) in the medical image (in one or more reconstructed medical images and / or one or more medical images that do not require reconstruction) for applying the TTField to the subject. The ROI may define the location where the TTField will converge. In some embodiments, the ROI may be defined based on user input, or may be defined partially or entirely by a computer.

[0037] At step 212, method 200 may include creating a three-dimensional (3D) model of the subject based on reconstructed medical images and medical images that do not require reconstruction, wherein the 3D model of the subject may include a region of interest (ROI). In some embodiments, the 3D model may include a 3D conductivity map depicting the conductivity of the subject's tissues. In some embodiments, the 3D model may be computer-generated and may be based on user input.

[0038] In some embodiments, creating a 3D model may include performing calculations based on reconstructed medical images and any other medical images, as well as the tissue types therein, to determine the electrical conductivity of a subject's tissues. As an example, creating a 3D model may include assigning tissue types and associated electrical conductivity to voxels in the subject's 3D model. In some embodiments, creating a 3D model of a subject may include automatically segmenting normal tissue in medical images. In some embodiments, the 3D model may be created based on user input, such as user approval of a 3D conductivity map associated with the 3D model.

[0039] At step 214, method 200 may include generating a plurality of transducer layouts for applying the TTField to the subject based on a 3D model of the subject, which was created at step 212 based on reconstructed medical images. The transducer layouts may define one or more locations relative to the subject for placing one or more transducers. In some embodiments, the transducer layouts may include four locations on the subject (such as on the subject's head or torso) for placing four corresponding transducers. In some embodiments, the plurality of transducer layouts may include two locations on the subject (such as on the subject's head or torso) for placing two corresponding transducers. In some embodiments, the transducers may include one or more electrode elements. The electrode element may be of any suitable type or material. For example, at least one electrode element may include a ceramic dielectric layer, a polymer film, etc., including combinations and / or multiple composites thereof. In some embodiments, step 214 may be computer-implemented.

[0040] At step 216, method 200 may include selecting one or more transducer layouts from the transducer layouts as recommended transducer layouts. In some embodiments, one, two, three, four, or more transducer layouts may be selected as recommended transducer layouts. In some embodiments, step 216 may be computer-implemented. In some embodiments, the selection may be based on user input.

[0041] At step 218, method 200 may include presenting a recommended transducer layout on a display. In some embodiments, the recommended transducer layout may be presented on a display via one or more output devices for user consideration.

[0042] At step 220, method 200 may include receiving a user selection of at least one recommended transducer layout. The user selection may indicate a preferred transducer layout for applying the TTField in the subject's ROI.

[0043] At step 222, method 200 may include providing a report for at least one selected recommended transducer layout, wherein the report indicates the selection result from step 122.

[0044] Figure 3 An example method 300 for determining whether a medical image contains a truncated portion of a subject is described. Method 300 can achieve... Figure 2A Step 204. Method 300 can be implemented by a computer including one or more processors and a memory accessible by the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the computer to perform the relevant steps of method 300. Method 300 can be modified, added to, or omitted. Although for illustrative purposes... Figure 3 The instructions specify the order of operations, but the timing and order of such operations can be changed where appropriate without negating the purpose and advantages of the examples detailed herein.

[0045] refer to Figure 3 At step 302, method 300 may include determining the perimeter of the subject in the medical image. The perimeter may be determined using automatic edge detection by identifying voxels with background image values ​​compared to voxels with non-background image values.

[0046] At step 304, method 300 may include determining whether the subject's circumference has a straight segment with a length greater than or equal to a straight segment threshold. In some embodiments, the straight segment threshold may be the maximum length of a straight segment that is considered not to be truncated in a medical image. In some embodiments, the straight segment threshold may be determined by the user. As an example, the straight segment threshold may be between 3 cm and 5 cm. When it is determined in step 304 that the subject's circumference has a straight segment with a length greater than or equal to the straight segment threshold, the process proceeds to step 306. When it is determined in step 304 that the subject's circumference does not have a straight segment with a length greater than or equal to the straight segment threshold, the process proceeds to step 308.

[0047] In step 306, the medical image is designated as having a truncated portion of the subject, and is therefore designated as a truncated medical image.

[0048] In step 308, the medical image is designated as not having the subject's truncated portion.

[0049] As Figure 3For example, if in step 304 the subject's circumference has a straight line segment equal to or greater than 5 cm, then in step 306 the medical image is designated to have the subject's truncated portion. Otherwise, in step 308, the medical image is designated not to have the subject's truncated portion.

[0050] Figure 4A and Figure 4B Example methods 400A and 400B for determining whether a medical image contains a truncated portion of a subject are described. Methods 400A and 400B can achieve... Figure 2A Step 204. Methods 400A and 400B can be implemented by a computer including one or more processors and a memory accessible by the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the computer to perform the relevant steps of methods 400A and 400B. Methods 400A and 400B can be modified, added to, or omitted. Although for illustrative purposes... Figure 4A and Figure 4B The instructions specify the order of operations, but the timing and order of such operations can be changed where appropriate without negating the purpose and advantages of the examples detailed herein.

[0051] refer to Figure 4A At step 402, method 400A may include determining the perimeter of the subject in a slice of a medical image. The perimeter may be determined using automatic edge detection by identifying voxels with background image values ​​compared to voxels with non-background image values.

[0052] At step 404, method 400A may include comparing the subject's circumference with the subject's expected circumference. In some embodiments, the subject's expected circumference may be based on a view of the subject in a slice of a medical image, and at least one of sex, age, body mass index, circumference measurement, height measurement, or a measurement of a portion of the subject.

[0053] At step 406, method 400A may include determining whether the subject's circumference exceeds a tolerance range for the subject's expected circumference. The tolerance range for the subject's expected circumference can be expressed as a percentage. In some embodiments, the tolerance range for the subject's expected circumference may be determined by the user. As an example, the tolerance range for the subject's expected circumference may be between 3% and 5%. After determining that the subject's circumference exceeds the tolerance range for the subject's expected circumference, the process proceeds to step 408. After determining that the subject's circumference does not exceed the tolerance range for the subject's expected circumference, the process proceeds to step 410.

[0054] In step 408, the medical image is designated as having a truncated portion of the subject, and is therefore designated as a truncated medical image.

[0055] In step 410, the medical image is designated as not having the subject's truncated portion.

[0056] As Figure 4A For example, assuming the subject's expected circumference is 100 cm and the tolerance range for the subject's expected circumference is 5%, then if the subject's circumference is less than 95 cm or greater than 105 cm, the medical image can be designated to have the subject's truncated portion. If the subject's circumference is between 95 cm and 105 cm, the medical image can be designated to not have the subject's truncated portion.

[0057] Go to Figure 4B At step 422, method 400B may include determining the perimeter of the subject in a slice of a medical image. The perimeter may be determined using automatic edge detection by identifying voxels with background image values ​​compared to voxels with non-background image values.

[0058] At step 424, method 400B may include determining whether the subject's perimeter touches one or more edges of a medical image. In some embodiments, the subject's perimeter touches one or more edges of a medical image if one or more straight line segments are formed in the subject's region, rather than one or more edges of the medical image being tangent only to the subject's perimeter. After determining that the subject's perimeter touches one or more edges of the medical image, the process proceeds to step 426. After determining that the subject's perimeter does not touch an edge of the medical image, the process proceeds to step 428.

[0059] In step 426, the medical image is designated as having a truncated portion of the subject, and is therefore designated as a truncated medical image.

[0060] In step 428, the medical image is designated as not having the subject's truncated portion.

[0061] Figure 5 An example medical image with a truncated portion is shown. According to... Figure 5 The medical image has truncated portions 502 and 504. Each of the truncated portions 502 and 504 has a straight line segment. In some embodiments, each straight line segment may have a length greater than or equal to a straight line segment threshold, as discussed in method 300. In some embodiments, due to the truncated portions 502 and 504, the subject's perimeter will not perfectly match the perimeter of an untruncated subject and may exceed the tolerance range of the subject's expected perimeter, as discussed in method 400A. In some embodiments, Figure 5The circumference of the subjects shown touches the left and right edges of the medical image at portions 502 and 504, respectively, as discussed in method 400B.

[0062] Figure 6A , Figure 6B and Figure 6C An example of a first medical image processed according to an exemplary embodiment is depicted. Figure 6A , Figure 6B and Figure 6C A cross-sectional view of the subject's torso is depicted. Figure 6A The slices are depicted in the first medical image before processing. Figure 6A The first medical image was truncated, thus producing Figure 6B Truncated medical images. According to an example embodiment, Figure 6B The truncated medical image is reconstructed, thus producing Figure 6C Reconstructed medical images.

[0063] Figure 6D , Figure 6E and Figure 6F An example of a second medical image processed according to an exemplary embodiment is depicted. Figure 6D , Figure 6E and Figure 6F Depicting and Figure 6A , Figure 6B and Figure 6C Cross-sectional views of the torso of different subjects in the study. Figure 6D The slices are depicted in the second medical image before processing. Figure 6D The medical images were truncated, thus producing Figure 6E Truncated medical images of the subjects. According to an example embodiment, Figure 6E The truncated medical image is reconstructed, thus producing Figure 6F Reconstructed medical images.

[0064] In comparison Figures 6A to 6C and comparison Figures 6D to 6F hour, Figure 6C and Figure 6F Reconstructed medical images in China are... Figure 6A and Figure 6D A reasonable representation of medical images in [the context]. Therefore, Figure 6C and Figure 6F The reconstructed medical images in the images are of sufficient quality for TTField treatment planning. However, due to truncation on the left and right sides of each image, therefore... Figure 6B and Figure 6E The quality of the truncated medical images in the images is insufficient for use in TTField treatment planning.

[0065] Exemplary device Figure 7 An example system 700 for applying an alternating electric field (e.g., TTField) to a subject's body is depicted. This system can be used to treat a target area of ​​a subject's body using an alternating electric field (e.g., TTField). As an example, the target area can be in the subject's brain, and the alternating electric field can be applied via two pairs of transducers positioned on the subject's head (e.g., in...). Figure 8 In this configuration, the two pairs of transducers (each having four transducers 800) deliver an alternating electric field to the subject's body. As an example, the target area can be within the subject's torso, and the alternating electric field can be delivered to the subject's body via two pairs of transducers positioned on at least one of the subject's chest, abdomen, or one or both thighs. As an example, a single pair of transducers can be used. Additional transducers may also be placed on the subject's body.

[0066] Example device 700 depicts an example system having four transducers (or “transducer arrays”) 700A to 700D. Each transducer 700A to 700D may include substantially flat electrode elements 702A to 702D positioned on substrates 704A to 704D and electrically and physically connected (e.g., via conductive lines 706A to 706D). Substrates 704A to 704D may include, for example, cloth, foam, flexible plastic, and / or conductive medical gel. Two transducers (e.g., 700A and 700D) may be a first pair of transducers configured to apply an alternating electric field to a target region of a subject's body. Two additional transducers (e.g., 700B and 700C) may be a second pair of transducers configured to similarly apply an alternating electric field to the target region.

[0067] Transducers 700A to 700D may be coupled to AC voltage generator 720, and the system may also include a controller 710 communicatively coupled to AC voltage generator 720. Controller 710 may include a computer having one or more processors 724 and a memory 726 accessible by the one or more processors. Memory 726 may store instructions that, when executed by the one or more processors, control AC voltage generator 720 to sense alternating electric fields between pairs of transducers 700A to 700D according to one or more voltage waveforms and / or cause the computer to perform one or more methods disclosed herein. Controller 710 may monitor operations performed by AC voltage generator 720 (e.g., via processor 724). One or more sensors 728 may be coupled to controller 710 to provide measurements or other information to controller 710.

[0068] In some embodiments, the voltage generating component may supply an electrical signal to the transducers 700A to 700D, the electrical signal having an alternating current waveform with a frequency in the range of about 50 kHz to about 1 MHz, and suitable for delivering TTField therapy to the subject's body.

[0069] Electrode elements 702A to 702D may be capacitively coupled. As an example, electrode elements 702A to 702D may be ceramic electrode elements coupled to each other via conductive lines 706A to 706D. When viewed in a direction perpendicular to their surface, the ceramic electrode elements may be circular or non-circular. In other embodiments, the array of electrode elements may not be capacitively coupled, and there may be no dielectric material (such as a ceramic or high-dielectric polymer layer) associated with the electrode elements.

[0070] The transducers 700A to 700D can take various structural forms. The transducers can be fixed to the subject's body or attached to or integrated into clothing covering the subject's body. The transducers may include suitable materials for attaching the transducers to the subject's body. For example, these suitable materials may include fabric, foam, flexible plastics, and / or conductive medical gel. The transducers may be conductive or non-conductive.

[0071] The transducer may include any desired number of electrode elements (e.g., one electrode element, or more than one electrode element). These electrode elements may use a variety of shapes, sizes, and materials. Any configuration for implementing the transducer (or electric field generating device) used with embodiments of the invention may be used, provided that they are capable of (a) delivering the TTField to the subject's body and (b) being positioned at the location specified herein. In some embodiments, at least one electrode element of the first, second, third, or fourth transducer may include at least one ceramic disk adapted to generate an alternating electric field. In some embodiments, at least one electrode element of the first, second, third, or fourth transducer may include a polymer film adapted to generate an alternating electric field.

[0072] Figure 9 An example computer apparatus for use with one or more embodiments described herein is depicted. As an example, apparatus 900 may be a computer to implement certain techniques of the invention disclosed herein, such as reconstructing truncated medical images and / or training machine learning models to generate reconstructed medical images. For example, Figure 1 Steps 102 to 118 and Figure 2A and Figure 2B Steps 202 to 222 can be performed by a computer, such as device 900. As an example, device 900 can be used as... Figure 7The controller 710, or used as a separate computer device remote from the controller 710.

[0073] The device 900 may include one or more processors 902, memory 903, one or more input devices 905, and one or more output devices 906.

[0074] Inputs to device 900 may be provided by one or more input devices 905, via link 901 (e.g., a wired or wireless link; for example, using a direct connection or via a network) from one or more input devices communicating with device 900, and / or via link 901 from another computer communicating with device 900. As an example, based on input 901, one or more processors 902 may generate control signals for controlling AC voltage generator 720. As an example, input 701 may be user input. As an example, input 701 may come from another computer communicating with device 700.

[0075] The output for device 900 may be provided by one or more output devices 906, provided to one or more output devices communicating with device 900 via link 901, and / or provided from another computer communicating with device 900 via link 901. One or more output devices 906 may provide the operational status of the invention, such as transducer layout selection, generating voltage, and other operational information. According to certain embodiments of the invention, output device 905 may provide visualized data.

[0076] In some embodiments, one or more input devices 905 and one or more output devices 906 may be combined into one or more modular input / output devices (e.g., touch screens).

[0077] In some embodiments, one or more processors 902 may perform the operations described herein based on input from one or more input devices 905 or input from outside the device 900 via link 901. As an example, user input may be received from one or more input devices 905. As an example, input may come from another computer communicating with the device 900 via link 901. As an example, input may come from one or more input devices communicating with the device 900 via link 901.

[0078] In some embodiments, one or more processors 902 may perform the operations described herein and provide the results of the operations as output. As an example, the output may be provided to one or more output devices 906. As an example, the output may be provided to another computer communicating with device 900 via link 901. As an example, the output may be provided to one or more output devices communicating with device 900 via link 901.

[0079] Memory 903 may be accessible to one or more processors 902, such that one or more processors 902 may read information from memory 903 and write information to memory. Memory 903 may store instructions that, when executed by one or more processors 902, implement one or more embodiments described herein. Memory 903 may be a non-transitory computer-readable medium (or a non-transitory processor-readable medium) containing a set of instructions for reconstructing truncated medical images and / or training machine learning models to generate reconstructed medical images, wherein, when executed by a processor (such as one or more processors 902), the instructions cause the processor to perform one or more methods discussed herein.

[0080] The apparatus 900 may be an apparatus for reconstructing truncated medical images and / or training machine learning models to generate reconstructed medical images, the apparatus comprising: one or more processors (such as one or more processors 902); and a memory (such as memory 903) accessible by the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the apparatus to perform one or more methods described herein.

[0081] The memory 903 may be a non-transitory processor-readable medium containing thereon a set of instructions for reconstructing a truncated medical image and / or training a machine learning model to generate a reconstructed medical image, wherein the instructions, when executed by one or more processors (such as one or more processors 902), cause the one or more processors to perform one or more methods described herein.

[0082] Exemplary Examples The present invention includes the following other exemplary embodiments (“Embodiments”).

[0083] Example 1. A computer-implemented method for reconstructing a truncated medical image, the method comprising: accessing a medical image of a subject from a memory, the medical image including voxels; determining that the medical image has a truncated portion of the subject, the voxels of the truncated portion having non-tissue image values; and generating a reconstructed medical image using a trained machine learning model and the medical image, the trained machine learning model being trained to generate an untruncated reconstructed medical image, the reconstructed medical image having tissue image values ​​in the voxels of the truncated portion.

[0084] Example 2: According to the method of Example 1, determining that the medical image has a truncated portion of the subject includes: determining that the subject depicted in the medical image has at least two flat surfaces.

[0085] Example 2A: According to the method of Example 1, determining that the medical image has a truncated portion of the subject includes: determining that the subject depicted in the medical image has at least one flat surface.

[0086] Example 3: According to the method of Example 1, determining that the medical image has a truncated portion of the subject includes: determining the perimeter of the subject in the medical image; determining whether the perimeter of the subject has a line segment with a length greater than or equal to a line segment threshold; if the perimeter of the subject has a line segment with a length greater than or equal to the line segment threshold, then designating the medical image as having a truncated portion of the subject; and if the perimeter of the subject does not have a line segment with a length greater than or equal to the line segment threshold, then designating the medical image as not having a truncated portion of the subject.

[0087] Example 4: According to the method of Example 1, determining that the medical image has a truncated portion of the subject includes: determining the perimeter of the subject in a slice of the medical image; comparing the perimeter of the subject with the expected perimeter of the subject; if the perimeter of the subject exceeds a tolerance range of the expected perimeter of the subject, then designating the medical image as having a truncated portion of the subject; and if the perimeter of the subject does not exceed a tolerance range of the expected perimeter of the subject, then designating the medical image as not having a truncated portion of the subject.

[0088] Example 5: The method according to Example 4, wherein the expected circumference of the subject is based on a view of the subject in the slice of the medical image, and at least one of gender, age, body mass index, circumference measurement, height measurement, or a measurement of a portion of the subject.

[0089] Example 6: According to the method of Example 1, determining that the medical image has a truncated portion of the subject includes: presenting the medical image on a display; and receiving user input, the user input identifying the medical image as having a truncated portion of the subject.

[0090] Example 7: According to the method described in Example 1, the non-tissue image values ​​of the truncated portion are the same image values.

[0091] Example 8: According to the method described in Example 1, the non-tissue image value of the truncated portion corresponds to the background image value.

[0092] Example 9: The method according to Example 1, wherein the medical image includes the torso of the subject.

[0093] Example 10: The method according to Example 1, wherein the medical image includes the head of the subject.

[0094] Example 11: According to the method of Example 1, the medical images include: computed tomography (CT) medical images, magnetic resonance imaging (MRI) medical images, or positron emission tomography (PET) medical images.

[0095] Example 12: The method according to Example 1, further comprising: defining a region of interest (ROI) in at least one of the medical image or the reconstructed medical image for applying a tumor therapeutic electric field to the subject; creating a three-dimensional model of the subject based on the reconstructed medical image, the three-dimensional model of the subject including the region of interest; generating a plurality of transducer layouts for applying a tumor therapeutic electric field to the subject based on the three-dimensional model of the subject; selecting at least one transducer layout as a recommended transducer layout; presenting the recommended transducer layout; receiving a user selection for at least one recommended transducer layout; and providing a report for the at least one selected recommended transducer layout.

[0096] Example 12A: A non-transitory processor-readable medium containing thereon a set of instructions for reconstructing a truncated medical image, wherein the instructions, when executed by a processor, cause the processor to perform a method comprising: accessing a medical image of a subject from memory, the medical image comprising voxels; determining that the medical image has a truncated portion of the subject, the voxels of the truncated portion having non-tissue image values; and generating a reconstructed medical image using a trained machine learning model and the medical image, the trained machine learning model being trained to generate an untruncated reconstructed medical image, the reconstructed medical image having tissue image values ​​in the voxels of the truncated portion.

[0097] Example 12B: An apparatus for reconstructing a truncated medical image, the apparatus comprising: one or more processors; and a memory accessible by the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the apparatus to perform a method comprising: accessing a medical image of a subject from the memory, the medical image comprising voxels; determining that the medical image has a truncated portion of the subject, the voxels of the truncated portion having non-tissue image values; and generating a reconstructed medical image using a trained machine learning model and the medical image, the trained machine learning model being trained to generate an untruncated reconstructed medical image, the reconstructed medical image having tissue image values ​​in the voxels of the truncated portion.

[0098] Example 13: A computer-implemented method for reconstructing truncated medical images, the method comprising: accessing from memory multiple medical images of multiple subjects, the medical images belonging to the torso of each of the subjects, the medical images including voxels; processing the medical images by truncating a portion of each medical image to obtain truncated medical images, wherein the truncated portion of each medical image corresponds to the right and left sides of each subject; designating a set of the truncated medical images as training truncated medical images; and training a machine learning model to obtain a trained machine learning model, wherein the machine learning model is trained using the training truncated medical images, wherein the machine learning model is trained to generate untruncated reconstructed medical images.

[0099] Example 14: According to the method of Example 13, the method further includes: designating a set of truncated medical images as test truncated medical images, wherein the test truncated medical images are different from the training truncated medical images; generating reconstructed medical images using the trained machine learning model and the test truncated medical images; comparing the reconstructed medical images with the medical images corresponding to the training truncated medical images to obtain a comparison result; and if the comparison result is unsatisfactory, retraining the trained machine learning model to obtain a retrained machine learning model.

[0100] Example 15: The method according to Example 13, wherein the machine learning model is a generative adversarial network.

[0101] Example 16: According to the method described in Example 13, the machine learning model is a generative adversarial network, which includes a deconvolutional neural network as a generator and a convolutional neural network as a discriminator.

[0102] Example 17: The method according to Example 13, wherein the trained machine learning model is a deep learning neural network.

[0103] Example 18: The method described in Example 13, wherein the machine learning model is an unsupervised machine learning model.

[0104] Example 19: The method according to Example 13, wherein the machine learning model includes a generative adversarial network (GAN), MedGAN, super-resolution GAN, pix2pix GAN, cycleGAN, discoGAN, fila-sGAN, projection adversarial network (PAN), variational autoencoder (VAE), or unsupervised neural network.

[0105] Example 19A: A computer-implemented method for reconstructing a truncated medical image, the method comprising: accessing a medical image of a subject from memory, the medical image being of the subject's torso and including voxels; determining that the medical image has a truncated portion of the subject, the truncated portion of the medical image corresponding to the right and left sides of the subject, the voxels of the truncated portion having non-tissue image values; generating a reconstructed medical image using a trained machine learning model and the medical image, the trained machine learning model being trained to generate an untruncated reconstructed medical image, the voxels of the reconstructed medical image having tissue image values ​​in the truncated portion; and generating, based on the reconstructed medical image, a plurality of transducer layouts for applying a tumor therapeutic electric field to the subject.

[0106] Example 19B: A non-transitory processor-readable medium containing thereon a set of instructions for reconstructing a truncated medical image, wherein the instructions, when executed by a processor, cause the processor to perform the following methods: accessing a medical image of a subject from memory, the medical image being of the subject's torso and including voxels; determining that the medical image has a truncated portion of the subject, the truncated portion of the medical image corresponding to the right and left sides of the subject, the voxels of the truncated portion having non-tissue image values; generating a reconstructed medical image using a trained machine learning model and the medical image, the trained machine learning model being trained to generate an untruncated reconstructed medical image, the voxels of the reconstructed medical image having tissue image values ​​in the truncated portion; and based on the reconstructed medical image, generating a plurality of transducer layouts for applying a tumor therapeutic electric field to the subject.

[0107] Example 20: An apparatus for reconstructing a truncated medical image, the apparatus comprising: one or more processors; and a memory accessible by the one or more processors, the memory storing instructions which, when executed by the one or more processors, cause the apparatus to perform a method comprising: accessing from the memory a medical image of a subject, the medical image being of the subject's torso, the medical image comprising voxels; determining that the medical image has a truncated portion of the subject, the truncated portion of the medical image corresponding to the right and left sides of the subject, the voxels of the truncated portion having non-tissue image values; generating a reconstructed medical image using a trained machine learning model and the medical image, the trained machine learning model being trained to generate an untruncated reconstructed medical image, the voxels of the reconstructed medical image having tissue image values ​​in the truncated portion; and, based on the reconstructed medical image, generating a plurality of transducer arrangements for applying a tumor therapeutic electric field to the subject.

[0108] Example 21: A method, machine, article of manufacture and / or system substantially as shown and described.

[0109] Alternatively, for each embodiment described herein, a voltage generating component supplies an electrical signal to the transducer having an alternating current waveform with a frequency in the range of about 50 kHz to about 1 MHz, and is suitable for delivering TTField therapy to the subject's body.

[0110] Unless otherwise stated herein or clearly contradicted by the context, embodiments illustrated under any heading or in any part of this disclosure may be combined with embodiments illustrated under the same or any other heading or other part of this disclosure. For example, but not limited to, embodiments described in the form of dependent claims for a given embodiment (e.g., a given embodiment described in the form of independent claims) may be combined with other embodiments (described in the form of independent claims or dependent claims).

[0111] Various modifications, alterations, and changes can be made to the described embodiments without departing from the scope of the invention as defined by the claims. It is intended that the invention is not limited to the described embodiments, but has the full scope defined by the language of the following claims and their equivalents.

Claims

1. A computer-implemented method for reconstructing truncated medical images, the method comprising: Accessing medical images of the subject from memory, the medical images including voxels; The medical image is determined to have a truncated portion of the subject, and the voxel of the truncated portion has non-tissue image values; A reconstructed medical image is generated using a trained machine learning model and the medical image, the trained machine learning model being trained to generate an untruncated reconstructed medical image, the voxels in the truncated portion of the reconstructed medical image having tissue image values.

2. The method of claim 1, wherein determining that the medical image has a truncated portion of the subject comprises: The subject depicted in the medical image is determined to have at least two flat surfaces.

3. The method of claim 1, wherein determining that the medical image has a truncated portion of the subject comprises: Determine the circumference of the subject in the medical image; Determine whether the perimeter of the subject has a straight segment with a length greater than or equal to a straight segment threshold; If the perimeter of the subject has a straight segment with a length greater than or equal to the straight segment threshold, then the medical image is designated as having the truncated portion of the subject; as well as If the perimeter of the subject does not have a straight segment with a length greater than or equal to the straight segment threshold, then the medical image is designated as not having the subject's truncated portion.

4. The method of claim 1, wherein determining that the medical image has a truncated portion of the subject comprises: Determine the circumference of the subject in a slice of the medical image; The subject's circumference is compared with the subject's expected circumference; If the subject's circumference exceeds the tolerance range of the subject's expected circumference, the medical image is designated as having the subject's truncated portion; as well as If the subject's circumference does not exceed the tolerance range of the subject's expected circumference, then the medical image is designated as not having the subject's truncated portion.

5. The method of claim 1, wherein determining that the medical image has a truncated portion of the subject comprises: The medical images are displayed on the monitor; as well as Receive user input, which identifies the medical image as having a truncated portion of the subject.

6. The method of claim 1, wherein the non-tissue image values ​​of the truncated portion are the same image values.

7. The method according to claim 1, further comprising: Define a region of interest (ROI) in at least one of the medical image or the reconstructed medical image for applying a tumor therapeutic electric field to the subject; A three-dimensional model of the subject is created based on the reconstructed medical image, the three-dimensional model of the subject including the region of interest; Based on the three-dimensional model of the subject, a plurality of transducer layouts for applying a tumor therapeutic electric field to the subject are generated; Select at least one of the transducer layouts as the recommended transducer layout. The recommended transducer layout is presented. Receive user selection of at least one recommended transducer layout; and A report is provided for the at least one selected recommended transducer layout.

8. A computer-implemented method for reconstructing truncated medical images, the method comprising: Access multiple medical images of multiple subjects from memory, the medical images belonging to the torso of each of the subjects, the medical images including voxels; The medical images are processed by truncating a portion of each medical image to obtain truncated medical images, wherein the truncated portion of each medical image corresponds to the right and left sides of each subject; Designate a set of the truncated medical images as training truncated medical images; as well as A machine learning model is trained to obtain a trained machine learning model, wherein the machine learning model is trained using truncated medical images, and wherein the machine learning model is trained to generate untruncated reconstructed medical images.

9. The method according to claim 8, further comprising: A set of the truncated medical images is designated as test truncated medical images, wherein the test truncated medical images are different from the training truncated medical images; The trained machine learning model and the truncated medical image used for testing are used to generate reconstructed medical images. The reconstructed medical image is compared with the medical image corresponding to the truncated medical image used for training to obtain a comparison result; as well as If the comparison result is unsatisfactory, the trained machine learning model is retrained to obtain a retrained machine learning model.

10. The method of claim 8, wherein the machine learning model is a generative adversarial network.

11. The method of claim 8, wherein the machine learning model is a generative adversarial network, the generative adversarial network comprising a deconvolutional neural network as a generator and a convolutional neural network as a discriminator.

12. The method of claim 8, wherein the trained machine learning model is a deep learning neural network.

13. The method of claim 8, wherein the machine learning model is an unsupervised machine learning model.

14. The method of claim 8, wherein the machine learning model comprises a generative adversarial network (GAN), MedGAN, super-resolution GAN, pix2pix GAN, cycleGAN, discoGAN, fila-sGAN, projection adversarial network (PAN), variational autoencoder (VAE), or unsupervised neural network.

15. A computer-implemented method for reconstructing truncated medical images, the method comprising: Accessing medical images of a subject from memory, the medical images being of the subject's torso, the medical images including voxels; The medical image is determined to have a truncated portion of the subject, the truncated portion of the medical image corresponding to the right and left sides of the subject, and the voxel of the truncated portion has non-tissue image values; A reconstructed medical image is generated using a trained machine learning model and the medical image, wherein the trained machine learning model is trained to generate an untruncated reconstructed medical image, and the voxels in the truncated portion of the reconstructed medical image have tissue image values. as well as Based on the reconstructed medical image, a layout of multiple transducers is generated for applying a tumor therapeutic electric field to the subject.

Images