Organ deformation compensation for medical image registration

Abstract

Systems and methods for medical image registration are provided. First and second input medical images of one or more anatomical objects are received. For each respective anatomical object of the one or more anatomical objects: a region of interest comprising the respective anatomical object is detected in one of the first or second input medical images, the region of interest is extracted from the first input medical image and the second input medical image, and a motion distribution of the respective anatomical object is determined from one of the region of interest extracted from the first input medical image or the region of interest extracted from the second input medical image using a motion model specific to the respective anatomical object. The first and second input medical images are registered based on the motion distribution of each respective anatomical object of the one or more anatomical objects to generate a fused image.

Description

Technical Field

[0001] This invention relates generally to medical image registration, and more particularly to organ deformation compensation for medical image registration. Background Technology

[0002] Registration of medical images is fundamental to many medical imaging analysis tasks. However, registration of medical images can be significantly affected by patient motion patterns, such as breathing or cardiac motion. These patient motion patterns cause various distortions of organs and other anatomical objects in medical images, resulting in misalignment between medical images, and often occurs in patient regions where there is a high demand for accurate registration. Conventional registration techniques often fail to accurately compensate for this distortion. In particular, many conventional registration techniques focus on aligning structures with high intensity differences while neglecting misalignment of structures with low intensity differences. Summary of the Invention

[0003] According to one or more embodiments, a system and method for medical image registration are provided. A first input medical image and a second input medical image of one or more anatomical objects (e.g., organs) are received. For each of the one or more anatomical objects: a region of interest (ROI) including the corresponding anatomical object is detected in one of the first or second input medical images; the ROI is extracted from the first and second input medical images; and a motion distribution of the corresponding anatomical object is determined using a motion model specific to the corresponding anatomical object, based on either the ROI extracted from the first or second input medical image. The first and second input medical images are registered based on the motion distribution of each of the one or more anatomical objects to generate a fused image.

[0004] In one embodiment, the first input medical image and the second input medical image are registered by determining a regularization term for the one or more anatomical objects based on the distance between the motion distribution and the prior distribution of each corresponding anatomical object, and minimizing a loss function that includes the regularization term of the one or more anatomical objects. The regularization term can be determined by summing the distances between the motion distribution and the prior distribution of each corresponding anatomical object.

[0005] In one embodiment, the motion model specific to a given anatomical object includes a variational autoencoder containing an encoder, and the encoder is used to determine the motion distribution of the given anatomical object. The machine learning network can be trained by: training the encoder of the variational autoencoder to generate code representing the encoding of deformation between regions of interest extracted from a first training image and regions of interest extracted from a second training image; and training the decoder of the variational autoencoder to generate a deformation field from the code and the regions of interest extracted from the first training image, the deformation field representing the deformation between the regions of interest extracted from the first training image and the regions of interest extracted from the second training image.

[0006] In one embodiment, for each corresponding anatomical object, a motion model specific to the corresponding anatomical object is learned by receiving a first training image and a second training image of the corresponding anatomical object, detecting a region of interest (ROI) including the corresponding anatomical object in the first training image and detecting a region of interest including the corresponding anatomical object in the second training image, extracting the ROI from the first training image and the second training image, and training a machine learning network to model the motion of the corresponding anatomical object based on the ROI extracted from the first training image and the ROI extracted from the second training image, as a motion model specific to the corresponding anatomical object.

[0007] In one embodiment, a region of interest including a corresponding anatomical object can be detected by segmenting the corresponding anatomical object in one of a first input medical image or a second input medical image, and centering the region of interest around the segmented corresponding anatomical object.

[0008] These and other advantages of the present invention will be apparent to those skilled in the art from the following detailed description and accompanying drawings. Attached Figure Description

[0009] Figure 1 A method for medical image registration according to one or more embodiments is shown;

[0010] Figure 2 A method for learning an anatomical object-specific motion model is illustrated according to one or more embodiments;

[0011] Figure 3 An exemplary variational autoencoder according to one or more embodiments is shown; and

[0012] Figure 4 A high-level block diagram of the computer is shown. Detailed Implementation

[0013] This invention generally relates to methods and systems for organ deformation compensation in medical image registration. Embodiments of the invention are described herein to provide a visual understanding of such methods and systems for organ deformation compensation in medical image registration. Digital images typically consist of digital representations of one or more objects (or shapes). The digital representation of an object is often described herein in terms of identifying and manipulating that object. This manipulation is a virtual manipulation performed in the memory or other circuitry / hardware of a computer system. Therefore, it is to be understood that embodiments of the invention can be performed within a computer system using data stored within the computer system.

[0014] Furthermore, it should be understood that while the embodiments discussed herein may relate to organ deformation compensation for medical image registration, the invention is not limited thereto. Embodiments of the invention can be applied to the registration of any type of image by compensating for deformation or motion of any object of interest.

[0015] In medical image registration, the goal is to find a spatial transformation to transform a first image so that it can be aligned with a second image. The first image is often referred to as the moving image, and the second image is often referred to as the stationary image. Typically, motion (such as, for example, respiratory motion or cardiac motion) causes deformation of the anatomical object depicted in the medical image, leading to inaccuracies in such medical image registration. Some embodiments of the invention learn an anatomical object-specific motion model and use this motion model to determine a regularization term for the anatomical object used for medical image registration, thereby compensating for deformation or other motion of the anatomical object. Advantageously, the regularization term provides medical image registration with a high degree of accuracy.

[0016] Figure 1 A method 100 for medical image registration according to one or more embodiments is illustrated. Method 100 can be used with any suitable computing device (such as, for example, ...). Figure 4 The computer (402) is used to execute this.

[0017] At step 102, motion models specific to the corresponding anatomical objects in one or more anatomical objects are learned. Step 102 can be performed offline or during a preprocessing phase. In one embodiment, by performing... Figure 2 The method involves 200 steps to learn motion models specific to the corresponding anatomical object.

[0018] refer to Figure 2 This illustrates a method 200 for learning a motion model specific to an anatomical object, according to one or more embodiments. Method 200 can be generated by any suitable computing device (such as, for example, ...). Figure 4 The computer (402) is used to execute this.

[0019] At step 202, a first training image and a second training image of one or more specific anatomical objects are received. The first training image may correspond to the representation of... I The motion image of 0, and the second training image can correspond to the representation as I 1. Fixed image. First training image. I 0 and second training images I 1 depicts one or more specific anatomical objects in various deformed states due to, for example, respiratory or cardiac movements. In one embodiment, the first training image I 0 and second training images I 1 is a sequence of images acquired within a time period. The one or more anatomical objects may include any anatomical structure of the patient, such as organs (e.g., lungs, heart, liver, kidneys, bladder, etc.), blood vessels, bones, etc.

[0020] In one embodiment, the first training image I 0 and second training images I 1. They have the same modality. First training image. I 0 and second training images I 1 can have any suitable modality, such as, for example, X-ray, magnetic resonance imaging (MRI), computed tomography (CT), ultrasound (US), single-photon emission computed tomography (SPECT), positron emission tomography (PET), or any other suitable modality or combination of modalities. First training image I 0 and second training images I 1. These images can be directly acquired from the image acquisition device used to acquire them (e.g., Figure 4 The image acquisition device 414 receives the image. Alternatively, the first training image... I 0 and second training images I 1. Images can be received by loading previously acquired images from storage devices or memory of a computer system (e.g., a picture archiving and communication system, PACS), or by receiving images that have already been transmitted from a remote computer system. It should be understood that, although regarding the first training image... I 0 and second training images I Method 200 is described in section 1, but method 200 can be performed using any number of training images.

[0021] At step 204, for the corresponding specific anatomical object among the one or more specific anatomical objects, in the first training image I 0 or second training image IThe detection in 1 includes a region of interest (ROI) for a specific anatomical object. The ROI can be detected manually via input from a user (e.g., a clinician) or automatically using any suitable known technique. In one embodiment, the specific anatomical object is segmented from training images according to a selective and iterative method (SIMPLE) for performance level estimation, and the ROI is centered around the segmented specific anatomical object.

[0022] At step 206, regions of interest including the corresponding specific anatomical objects are extracted from the first training image and the second training image.

[0023] At step 208, a machine learning network is trained to model the motion of a specific anatomical object based on regions of interest extracted from a first training image and regions of interest extracted from a second training image, thereby learning a motion model specific to that specific anatomical object. This machine learning network can be any suitable machine learning method for modeling the motion of an anatomical object. In one embodiment, according to one or more embodiments, the machine learning network is a variational autoencoder (VAE), such as... Figure 3 The VAE 300 shown is used to provide a probabilistic motion model specific to a particular anatomical object.

[0024] VAE 300 includes encoder 306 and decoder 310. Encoder 306 is the receiving sub-image 302 and 304 as input and output code z A 0.308 neural network. Sub-image 302 and 304 are from the first training image I 0 and second training images I 1. Extracted region of interest o Code z 0 308 indicates that it is generated by the encoder. 306 from the first training image I 0 and second training images I A low-dimensional vector of the mean and variance of a 1-sampled multivariate Gaussian. Decoder 310 is the receiving sub-image 302 and code z 0 308 as input and output speed v 312 and Deformation Field ϕ A 314-bit neural network. Speed. v312 is the raw output of VAE 300 and is non-diffeomorphic. Speed v 312 includes the velocity value for each pixel. Deformation field ϕ 314 is calculated using exponential operations and represents a sub-image. 302 and Deformation between 304. By making the motion sub-image 302 appearance information for decoder 310 is available, deformation field ϕ 314 is more likely to encode deformation information than appearance information. For example... Figure 3 As shown, the deformation field ϕ 314 was applied to the sub-image 302 to reconstruct sub-images 304. Train VAE 300 according to Equation 1 to optimize the code. z 0 308, thus optimally transforming the sub-image. 302 so as to be compatible with sub-images 304 alignment.

[0025] (Equation 1).

[0026] in p ( z 0) is z The prior distribution is 0, and it is assumed to follow a multivariate unit Gaussian distribution. (Prior distribution) p ( z 0) refers to the distribution learned by VAE 300, which represents the distribution of all possible motions of a particular anatomical object (as observed in the training set).

[0027] It should be understood that method 200 can return to step 204 and can iteratively repeat steps 204-208 to learn a motion model specific to each of the one or more specific anatomical objects.

[0028] Return to reference Figure 1 In step 104, a first input medical image and a second input medical image of the one or more anatomical objects are received. The first input medical image may correspond to a motion image. M Furthermore, the second input medical image can correspond to a fixed image in the dataset to be registered. F The first input medical image M Second input medical images FThe same one or more anatomical objects are depicted in various deformed states due to, for example, respiratory or cardiac movements. In one embodiment, a first input medical image... M Second input medical images F It is a sequence of images acquired within a certain time period. The first input medical image. M Second input medical images F It can have any suitable modality.

[0029] First input medical image M Second input medical images F These images can be directly obtained from the image acquisition device used to acquire them (e.g., Figure 4 The image acquisition device 414 receives the image. Alternatively, the first input medical image... M Second input medical images F Images can be received by loading previously acquired images from storage devices or memory of a computer system (e.g., a picture archiving and communication system PACS) or by receiving images that have been transmitted from a remote computer system.

[0030] At step 106, for each corresponding anatomical object, in the first input medical image M Or a second input medical image F One of the detection methods includes the region of interest of the corresponding anatomical object. o Area of ​​Interest o It can be detected manually via input from a user (e.g., a clinician) or automatically using any suitable known technology. In one embodiment, it can be detected by input from a first input medical image. M Or a second input medical image F The corresponding anatomical object is segmented, and the region of interest is centered around the segmented anatomical object to automatically determine the region of interest, as mentioned above. Figure 2 As described in step 204 of method 200.

[0031] At step 108, the region of interest is extracted from the first input medical image and the second input medical image.

[0032] At step 110, a motion model specific to the corresponding anatomical object (learned at step 102) is used, based on the first input medical image. M Extracted region of interest and from the second input medical image F The extracted region of interest is used to determine the motion distribution of the corresponding anatomical object. In one embodiment where the motion model is a VAE, it is based on a function... The trained encoder of the motion model (For example, Figure 3 encoder 306) Applied to from the first input medical image M Extracted region of interest and from the second input medical image F Extracted regions of interest to determine the motion distribution of the corresponding anatomical object. z 0. Specifically, the encoder Receive from the first input medical image M Extracted region of interest and such as through deformation field Modified from the first input medical image M The extracted region of interest is used as input, and the output is the motion distribution. z 0. Deformation field Indicates the first input medical image M With the second input medical image F Global motion between them. Advantageously, the same image (e.g., the first input medical image) is calculated. M Motion distribution on ) z 0 enables multimodal registration using only one modality trained on a motion model specific to the corresponding anatomical object. It should be understood that the motion distribution... z 0 can be replaced by according to The motion distribution obtained by reversing z 0 determines, where It is trained using fixed images.

[0033] Then, by changing the training weights of the network in each iteration... θ To minimize motion distribution z 0= With prior distribution p ( z The distance between 0) is minimized. z 0 and prior distribution p ( z The distance between 0) ensures the distribution of motion. z 0 represents the possible motion of the corresponding anatomical object. In one embodiment, this distance is the Kullback-Leibler divergence. However, other distance metrics can be applied. Such as, for example, optimal transmission loss, generative adversarial networks, adversarial autoencoders, or any other suitable distance metric.

[0034] Method 100 can return to step 106 and can iteratively repeat steps 106-110 (e.g., for whole-body registration) for each corresponding anatomical object in the one or more anatomical objects to determine the motion distribution of each anatomical object in the one or more anatomical objects.

[0035] At step 112, the first input medical image and the second input medical image are registered based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects to generate a fused image.

[0036] In one embodiment, an anatomical object-specific regularization term The motion distribution was determined for each corresponding anatomical object. z 0= With prior distribution p ( z The sum of the distances between 0) is shown in Equation 2:

[0037] (Equation 2)

[0038] in o It is the region of interest for the corresponding anatomical object. It is a distance metric (e.g., Kullback-leibler divergence). From the first input medical image M Extracted region of interest o , It is the first input medical image M With the second input medical image F The global movement between them, and p ( z 0) is the prior distribution. Then, the first input medical image and the second input medical image are registered according to the loss function in Equation 3:

[0039] (Equation 3)

[0040] in It is a distance metric (e.g., Kullback-leibler divergence). F It is the second input medical image. M It is the first input medical image. It is the divergence field of the motion model. It is a space regularization term. It is the regularization term for the one or more anatomical objects defined in Equation 2, and λ 1 and λ 2 are respectively for and The parameters are weighted. The loss function in Equation 3 includes a regularization term for the one or more anatomical objects, and can therefore be used to compensate for organ deformation to train a machine learning network to register a first input medical image and a second input medical image to generate a fused image.

[0041] At step 114, the merged image is output. For example, the merged image can be output by displaying the merged image on a display device of the computer system, storing the merged image in the memory or storage device of the computer system, or by transmitting the merged image to a remote computer system.

[0042] Advantageously, embodiments of the invention provide anatomically specific regularization terms for medical image registration. This allows the registration process to focus on the region of interest and compensate for motion of anatomical objects within that region. Because some embodiments of the invention involve detecting bounding boxes in one of the input medical images, these embodiments are computationally less expensive than conventional techniques that require segmentation of anatomical objects in both images. Embodiments of the invention allow for medical image registration with a high degree of accuracy.

[0043] In some embodiments, instead of prior distribution p ( z 0), the average posterior distribution can be used. After learning the motion model for each anatomical object, the mean posterior distribution can be extracted. .

[0044] The systems, apparatus, and methods described herein can be implemented using digital circuitry or one or more computers that utilize known computer processors, memory units, storage devices, computer software, and other components. Typically, a computer includes a processor for executing instructions and one or more memories for storing instructions and data. A computer may also include or be coupled to one or more mass storage devices, such as one or more disks, internal hard disks and removable disks, magneto-optical disks, optical disks, etc.

[0045] The systems, apparatus, and methods described herein can be implemented using computers operating in a client-server relationship. Typically, in such systems, the client computer is located remotely from the server computer and interacts via a network. The client-server relationship can be defined and controlled by computer programs running on the respective client and server computers.

[0046] The systems, apparatus, and methods described herein can be implemented within a network-based cloud computing system. In such a system, a server or another processor connected to the network communicates with one or more client computers via the network. Client computers can communicate with the server via, for example, a web browser application residing and operating on the client computer. Client computers can store data on the server and access that data via the network. Client computers can transmit requests for data or for online services to the server via the network. The server can perform the requested service and provide data to (one or more) client computers(s). The server can also transmit data adapted to cause client computers to perform specified functions (e.g., perform calculations, display specified data on a screen, etc.). For example, the server can transmit data adapted to cause client computers to perform one or more steps or functions of the methods and workflows described herein (including...). Figure 1-2 A request for one or more steps or functions). Certain steps or functions of the methods and workflows described herein (including...) Figure 1-2 One or more steps or functions of the methods and workflows described herein may be performed by a server or by another processor in a network-based cloud computing system. Figure 1-2 One or more steps) can be performed by a client computer in a web-based cloud computing system. The steps or functions of the methods and workflows described herein (including...) Figure 1-2 One or more steps can be performed by a server and / or by a client computer in a web-based cloud computing system in any combination.

[0047] The systems, apparatuses, and methods described herein can be implemented using a computer program product tangibly embodied in an information carrier (e.g., embodied in a non-transitory machine-readable storage device) for execution by a programmable processor; and the methods and workflow steps described herein (including Figure 1-2 One or more steps or functions of a computer can be implemented using one or more computer programs that can be executed by such a processor. A computer program is a set of computer program instructions that can be used directly or indirectly in a computer to perform an activity or produce a result. A computer program can be written in any form of programming language (including compiled or interpreted languages) and can be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0048] Figure 4This document depicts a high-level block diagram of an example computer 402 that can be used to implement the systems, apparatus, and methods described herein. Computer 402 includes a processor 404 operatively coupled to a data storage device 412 and a memory 410. Processor 404 controls this operation by executing computer program instructions that define the overall operation of computer 402. The computer program instructions may be stored in the data storage device 412 or other computer-readable medium and loaded into memory 410 when execution of the computer program instructions is desired. Therefore, Figure 1-2 The methods and workflow steps or functions can be defined by computer program instructions stored in memory 410 and / or data storage device 412, and can be controlled by processor 404 that executes the computer program instructions. For example, the computer program instructions can be implemented to be programmed by those skilled in the art to perform... Figure 1-2 The computer-executable code describes the methods, workflow steps, or functions. Therefore, by executing computer program instructions, processor 404 performs... Figure 1-2 The computer 402 may include methods and workflow steps or functions. The computer 402 may also include one or more network interfaces 406 for communicating with other devices via a network. The computer 402 may also include one or more input / output devices 408 (e.g., monitor, keyboard, mouse, speaker, buttons, etc.) that enable a user to interact with the computer 402.

[0049] Processor 404 may include both general-purpose microprocessors and special-purpose microprocessors, and may be the sole processor of computer 402 or one of multiple processors. For example, processor 404 may include one or more central processing units (CPUs). Processor 404, data storage device 412 and / or memory 410 may include one or more application-specific integrated circuits (ASICs) and / or one or more field-programmable gate arrays (FPGAs), supplemented by or incorporated therein.

[0050] Both data storage device 412 and memory 410 include tangible, non-transitory computer-readable storage media. Both data storage device 412 and memory 410 may include high-speed random access memory, such as dynamic random access memory (DRAM), static random access memory (SRAM), double data rate synchronous dynamic random access memory (DDR RAM), or other random access solid-state memory devices, and may include non-volatile memory, such as one or more disk storage devices (e.g., internal hard disks and removable disks), magneto-optical disk storage devices, optical disk storage devices, flash memory devices, semiconductor memory devices (e.g., erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), digital versatile disc read-only memory (DVD-ROM) discs), or other non-volatile solid-state storage devices.

[0051] Input / output device 408 may include peripheral devices such as printers, scanners, displays, etc. For example, input / output device 408 may include display devices (such as cathode ray tube (CRT) or liquid crystal display (LCD) monitors) for displaying information to a user, a keyboard, and pointing devices such as mice or trackballs through which the user can provide input to computer 402.

[0052] Image acquisition device 414 can be connected to computer 402 to input image data (e.g., medical images) into computer 402. It is possible to implement image acquisition device 414 and computer 402 as a single device. It is also possible for image acquisition device 414 and computer 402 to communicate wirelessly via a network. In a possible embodiment, computer 402 may be remotely located relative to image acquisition device 414.

[0053] Any or all of the systems and apparatus discussed herein may be implemented using one or more computers (such as computer 402).

[0054] Those skilled in the art will recognize that actual computer or computer system implementations may have other structures and may include other components, and for illustrative purposes, Figure 4 It is a high-level representation of some components in this type of computer.

[0055] The foregoing specific embodiments should be understood in each respect as illustrative and exemplary, not restrictive, and the scope of the invention disclosed herein is not determined by these specific embodiments, but rather by the claims as interpreted under the full breadth permitted by patent law. It is to be understood that the embodiments shown and described herein merely illustrate the principles of the invention, and various modifications can be made by those skilled in the art without departing from the scope and spirit of the invention. Various other combinations of features can be implemented by those skilled in the art without departing from the scope and spirit of the invention.

Claims

1. A method comprising: Receive a first input medical image and a second input medical image depicting the same one or more anatomical objects in various deformed states; For each of the one or more anatomical objects: Detect the region of interest including the corresponding anatomical object in one of the first input medical image or the second input medical image. Extracting regions of interest from the first and second input medical images, and The motion distribution of the corresponding anatomical object is determined using a motion model specific to the corresponding anatomical object, based on either a region of interest extracted from a first input medical image or a region of interest extracted from a second input medical image. as well as The first input medical image and the second input medical image are registered based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects. The registration of the first input medical image and the second input medical image based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects includes: The regularization term for the one or more anatomical objects is determined based on the distance between the motion distribution and the prior distribution of each corresponding anatomical object.

2. The method of claim 1, wherein determining the regularization term for the one or more anatomical objects based on the distance between the motion distribution and the prior distribution of each corresponding anatomical object comprises: The distance between the motion distribution and the prior distribution of each corresponding anatomical object is summed.

3. The method of claim 1, wherein registering the first input medical image and the second input medical image based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects comprises: Minimize the loss function that includes the regularization term of the one or more anatomical objects.

4. The method of claim 1, wherein the motion model specific to the corresponding anatomical object comprises a variational autoencoder including an encoder, and wherein determining the motion distribution of the corresponding anatomical object using the motion model specific to the corresponding anatomical object, based on either a region of interest extracted from a first input medical image or a region of interest extracted from a second input medical image, comprises: The encoder is used to determine the motion distribution of the corresponding anatomical object.

5. The method of claim 1, further comprising, for each corresponding anatomical object, learning a motion model specific to that anatomical object in the following manner: Receive the first and second training images of the corresponding anatomical object; Detect the region of interest including the corresponding anatomical object in one of the first training image or the second training image; Extract the region of interest from the first training image and the second training image; as well as A machine learning network is trained to model the motion of a corresponding anatomical object based on regions of interest extracted from a first training image and regions of interest extracted from a second training image, as a motion model specific to the corresponding anatomical object.

6. The method of claim 5, wherein training a machine learning network to model the motion of a corresponding anatomical object based on a region of interest extracted from a first training image and a region of interest extracted from a second training image as a motion model specific to the corresponding anatomical object comprises: The encoder of the variational autoencoder is trained to generate code representing the encoding of the deformation between a region of interest extracted from a first training image and a region of interest extracted from a second training image; as well as The decoder of the variational autoencoder is trained to generate a deformation field from the code and a region of interest extracted from a first training image, the deformation field representing the deformation between the region of interest extracted from the first training image and a region of interest extracted from a second training image.

7. The method of claim 1, wherein detecting a region of interest comprising the corresponding anatomical object in one of the first input medical image or the second input medical image comprises: Segment the corresponding anatomical object from either the first input medical image or the second input medical image; as well as The region of interest is centered around the corresponding anatomical object that has been segmented.

8. The method of claim 1, wherein the one or more anatomical objects comprise one or more organs.

9. An apparatus comprising: A device for receiving a first input medical image and a second input medical image depicting the same one or more anatomical objects in various deformed states; A means for detecting a region of interest comprising the corresponding anatomical object in one of a first input medical image or a second input medical image for each of the one or more anatomical objects; A device for extracting regions of interest from a first input medical image and a second input medical image for each corresponding anatomical object; A device for determining the motion distribution of a corresponding anatomical object using a motion model specific to that anatomical object for each corresponding anatomical object, based on either a region of interest extracted from a first input medical image or a region of interest extracted from a second input medical image; as well as A means for registering a first input medical image and a second input medical image based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects; The means for registering a first input medical image and a second input medical image based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects includes: A device for determining the regularization term of one or more anatomical objects based on the distance between the motion distribution of each corresponding anatomical object and the prior distribution.

10. The apparatus of claim 9, wherein the means for determining the regularization term of the one or more anatomical objects based on the distance between the motion distribution of each corresponding anatomical object and the prior distribution comprises: A device for summing the distance between the motion distribution and the prior distribution of each corresponding anatomical object.

11. The apparatus of claim 9, wherein the means for registering the first input medical image and the second input medical image based on the motion distribution of each corresponding anatomical object among the one or more anatomical objects comprises: A means for minimizing a loss function that includes a regularization term comprising the one or more anatomical objects.

12. The apparatus of claim 9, wherein the motion model specific to the corresponding anatomical object comprises a variational autoencoder including an encoder, and wherein means for determining the motion distribution of the corresponding anatomical object using the motion model specific to the corresponding anatomical object, based on a region of interest extracted from a first input medical image or a region of interest extracted from a second input medical image, comprises: A device for determining the motion distribution of a corresponding anatomical object using the encoder.

13. A non-transitory computer-readable medium storing computer program instructions, which, when executed by a processor, cause the processor to perform operations including the following steps: Receive a first input medical image and a second input medical image depicting the same one or more anatomical objects in various deformed states; For each of the one or more anatomical objects: Detect the region of interest including the corresponding anatomical object in one of the first input medical image or the second input medical image. Extracting regions of interest from the first and second input medical images, and The motion distribution of the corresponding anatomical object is determined using a motion model specific to the corresponding anatomical object, based on either a region of interest extracted from a first input medical image or a region of interest extracted from a second input medical image. as well as The first input medical image and the second input medical image are registered based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects. The registration of the first input medical image and the second input medical image based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects includes: The regularization term for the one or more anatomical objects is determined based on the distance between the motion distribution and the prior distribution of each corresponding anatomical object.

14. The non-transitory computer-readable medium of claim 13, wherein registering the first input medical image and the second input medical image based on the motion distribution of each corresponding anatomical object in the one or more anatomical objects comprises: Minimize the loss function that includes the regularization term of the one or more anatomical objects.

15. The non-transitory computer-readable medium of claim 13, further comprising, for each corresponding anatomical object, learning a motion model specific to that anatomical object in the following manner: Receive the first and second training images of the corresponding anatomical object; Detect the region of interest including the corresponding anatomical object in one of the first training image or the second training image; Extract the region of interest from the first training image and the second training image; as well as A machine learning network is trained to model the motion of a corresponding anatomical object based on regions of interest extracted from a first training image and regions of interest extracted from a second training image, as a motion model specific to the corresponding anatomical object.

16. The non-transitory computer-readable medium of claim 15, wherein training a machine learning network to model the motion of a corresponding anatomical object based on a region of interest extracted from a first training image and a region of interest extracted from a second training image as a motion model specific to the corresponding anatomical object comprises: The encoder of the variational autoencoder is trained to generate code representing the encoding of the deformation between a region of interest extracted from a first training image and a region of interest extracted from a second training image; as well as The decoder of the variational autoencoder is trained to generate a deformation field from the code and a region of interest extracted from a first training image, the deformation field representing the deformation between the region of interest extracted from the first training image and a region of interest extracted from a second training image.

17. The non-transitory computer-readable medium of claim 13, wherein detecting a region of interest comprising a corresponding anatomical object in one of the first input medical image or the second input medical image comprises: Segment the corresponding anatomical object from either the first input medical image or the second input medical image; as well as The region of interest is centered around the corresponding anatomical object that has been segmented.

Images