Method and computer program product for enhancing medical images

Abstract

Systems, methods, and apparatuses associated with medical image enhancement are described herein. A medical image can include an object of interest, and the techniques disclosed herein can be used to identify the object and enhance the contrast between the object and its surrounding region by adjusting at least the pixels associated with the object. Object identification can be performed using an image filter, a segmentation mask, and / or a deep neural network trained to separate a medical image into multiple layers that respectively include the object of interest and the surrounding region. Once identified, the pixels of the object can be manipulated in various ways to improve the visibility of the object. These ways can include, for example, adding a constant value to the pixels of the object, applying a sharpening filter to these pixels, increasing the weight of these pixels, and / or smoothing the edge region around the object of interest.

Description

Technical Field

[0001] This application relates to the field of medical imaging. Background Technology

[0002] Medical images, such as X-ray fluoroscopy images, are now widely used to visualize internal organs and / or implanted surgical devices. However, due to limitations related to imaging speed and / or radiation dose, medical images acquired using existing medical imaging techniques often contain a significant amount of noise, which affects the visibility of some structures depicted in the medical images (e.g., tubular structures such as blood vessels, catheters, guidewires, etc.). Therefore, there is a need to develop systems and methods capable of detecting and enhancing the visibility of target structures in medical images to improve the usability of medical images. Summary of the Invention

[0003] This document describes systems, methods, and apparatuses associated with enhancing objects of interest in medical images. An apparatus capable of performing an image enhancement task may include one or more processors configured to: acquire a source medical image including an object of interest and a background region surrounding the object of interest; determine from the source medical image a plurality of first pixels associated with the object of interest and a plurality of second pixels associated with the background region; and generate a target medical image based on the plurality of first pixels and the plurality of second pixels to enhance the contrast between the object of interest and the background region in the target medical image. The plurality of first pixels associated with the object of interest may have corresponding first pixel values, the plurality of second pixels associated with the background region may have corresponding second pixel values, and the one or more processors may be configured to: enhance the contrast between the object of interest and the background region by at least adjusting the first pixel values ​​of the plurality of first pixels (e.g., associated with the object of interest) or the second pixel values ​​of the plurality of second pixels (e.g., associated with the background region).

[0004] In the example, one or more processors of the device described herein can be configured to: determine a segmentation mask associated with an object of interest using an artificial neural network or image filter, and determine a plurality of first pixels and a plurality of second pixels based on the segmentation mask. Once the plurality of first pixels associated with the object of interest are determined, the contrast between the object of interest and the background region can be enhanced by adding a constant (e.g., a positive or negative constant depending on the pixel value) to the respective first pixel values ​​and / or by applying a sharpening filter (e.g., an unsharpening mask) to the plurality of first pixels.

[0005] In the examples, the segmentation mask may include values ​​indicating the probability that each of a plurality of first pixels belongs to an object of interest. In these examples, the contrast between the object of interest and the background region can be enhanced by selecting at least a subset of the plurality of first pixels based on the probability that each of the plurality of first pixels belongs to an object of interest, determining the maximum pixel value among the selected pixels, and adjusting the pixel values ​​of each selected pixel based on the maximum pixel value and the probability that each of the selected pixels belongs to an object of interest.

[0006] In the example, one or more processors of the device described herein can be configured to separate a source medical image into a first layer and a second layer using a pre-trained machine learning (ML) model. The first layer may include an object of interest, the second layer may include a background region, and the one or more processors can be configured to determine a plurality of first pixels based on the first layer and a plurality of second pixels based on the second layer. In the example, the one or more processors can also be configured to determine corresponding weights for the first and second layers in the source medical image and to enhance the contrast between the object of interest and the background region by increasing the weight of the first layer in the target medical image or decreasing the weight of the second layer in the target medical image. The weight of the first layer in the target medical image can be increased, for example, by multiplying the corresponding first pixel values ​​of the plurality of first pixels in the target medical image by a value greater than the weight of the first layer in the source medical image. The weight of the second layer in the target medical image can be decreased, for example, by multiplying the corresponding second pixel values ​​of the plurality of second pixels in the target medical image by a value smaller than the weight of the second layer in the source medical image.

[0007] In the example, the ML model for separating source medical images can be trained using a pair of medical images including a first medical image and a second medical image, wherein the first medical image may include a depiction of the object of interest at a first time, the second medical image may include a depiction of the object of interest at a second time, and the ML model can be trained at least based on the motion of the object of interest from the first time to the second time.

[0008] While this document may use X-ray images and tubular anatomical structures or artificial structures as examples to describe embodiments of the present disclosure, those skilled in the art will understand that the techniques disclosed herein can also be used to enhance other types of medical images and / or other types of objects. Attached Figure Description

[0009] The examples disclosed herein can be understood in more detail from the following description, which is given by way of example in conjunction with the accompanying drawings.

[0010] Figure 1 This is a diagram illustrating an example of an object of interest in an enhanced medical image according to one or more embodiments of the disclosure provided herein.

[0011] Figure 2 This is a diagram illustrating an example of identifying and enhancing an object of interest in a medical image according to one or more embodiments of the disclosure provided herein.

[0012] Figure 3 This is a diagram illustrating an example of training a deep neural network to separate a medical image into two layers according to one or more embodiments of the disclosure provided herein.

[0013] Figure 4 This is a flowchart illustrating example operations that can be associated with an object of interest in an enhanced medical image according to one or more embodiments of the disclosure provided herein.

[0014] Figure 5 This is a flowchart illustrating example operations that can be associated with training a neural network to perform one or more tasks described herein.

[0015] Figure 6 This is a block diagram illustrating example components of a device that can be configured to perform the image enhancement tasks described herein. Detailed Implementation

[0016] The present disclosure is illustrated by way of example rather than limitation in the figures.

[0017] Figure 1 Examples of objects of interest 102 in enhanced medical images 104 according to one or more embodiments of the present disclosure are illustrated. As shown, medical image 104 (e.g., source medical image) may include medical scan images such as X-ray fluoroscopy images of the human body, and object of interest 102 may include anatomical structures of the human body and / or artificial devices that may be surgically inserted into the human body. For example, in some examples, object of interest 102 may include one or more tubular structures, such as one or more blood vessels (e.g., coronary arteries), catheters, guidewires, etc., and in other examples, object of interest 102 may include one or more non-tubular structures, such as the left ventricle, myocardium, etc. In addition to object of interest 102, medical image 104 may also include a background region 106 that may surround the object of interest and exhibit contrast with the object of interest, such contrast enabling a medical expert to distinguish the object of interest from the background. However, due to limitations associated with imaging speed and / or radiation dose, the contrast between the object of interest 102 and the background region 106 may not be as clear or distinguishable as desired. Therefore, various techniques may be required to enhance the object of interest in the target medical image 108 so that the structural details of the object of interest can be easily discerned by the human eye.

[0018] According to one or more embodiments of this disclosure, the contrast between object of interest 102 and background region 106 (e.g., the visibility of object of interest) can be enhanced by identifying object of interest at 110 (e.g., using a first set of computer vision and / or machine learning (ML) based techniques) and enhancing object of interest at 112 (e.g., using a second set of computer vision and / or ML techniques). In an example, identifying object of interest at 110 may include determining a plurality of first pixels associated with object of interest 102 and a plurality of second pixels associated with background region 106 from source medical image 104. The plurality of first pixels may be associated with corresponding first pixel values ​​(e.g., each such pixel value may be in the range of 0 to 255), and the plurality of second pixels may be associated with corresponding second pixel values ​​(e.g., each such pixel value may also be in the range of 0 to 255). Based on the determined plurality of first pixels and / or plurality of second pixels, the contrast between object of interest 102 and background region 106 (e.g., in target medical image 108) can be enhanced by adjusting the first pixel values ​​associated with the plurality of first pixels and / or the second pixel values ​​associated with the plurality of second pixels. For example, the contrast between the object of interest 102 and the background region 106 in the target medical image 108 can be enhanced by: adjusting the first pixel value associated with the object of interest by a first amount (e.g., a first constant value) while maintaining the second pixel value associated with the background region; adjusting the second pixel value associated with the background region by a second amount (e.g., a second constant value) while maintaining the first pixel value associated with the object of interest; or adjusting the first pixel value associated with the object of interest and the second pixel value associated with the background region to different values.

[0019] The adjustment amounts or values ​​described herein can be positive or negative, depending on, for example, the corresponding colors / shades used to represent the object of interest and the background region. For instance, if the object of interest 102 is depicted in a brighter color (e.g., with a higher pixel value) compared to a darker background (e.g., with a lower pixel value), the contrast between the object of interest and the background can be enhanced by adding normal values ​​to the pixels of the object of interest and / or adding negative constant values ​​to the pixels of the background region. Conversely, if the object of interest 102 is depicted in a darker color (e.g., with a lower pixel value) compared to a brighter background (e.g., with a higher pixel value) in the image, the contrast between the object of interest and the background can be enhanced by adding negative constant values ​​to the pixels of the object of interest and / or adding normal values ​​to the pixels of the background region.

[0020] Other techniques may also be used to manipulate the pixel values ​​of the object of interest 102 and / or the pixel values ​​of the background region 106 to enhance the object of interest in the target image 108. These techniques may include, for example, applying a sharpening filter (e.g., unsharpening mask) to a plurality of pixels associated with the object of interest, adjusting the pixel values ​​of at least a subset of pixels associated with the object of interest while maintaining the maximum pixel value associated with the object, increasing the weight of pixel values ​​associated with the object of interest in the target image, etc. Further details about these techniques will be provided below in conjunction with a description of techniques for separating (e.g., identifying) objects of interest and background regions in medical images. It should also be noted here that although embodiments of this disclosure may be described herein using X-ray fluoroscopy and / or tubular structures as examples, related techniques may also be applied to enhance other types of medical images (e.g., computed tomography (CT) images, magnetic resonance imaging (MRI) images, etc.) and / or other types of structures (e.g., non-tubular structures). Furthermore, the systems, methods, and apparatus employing the techniques described herein may additionally provide a user interface through which a user can switch between a raw image (e.g., source medical image 104) and an enhanced image (e.g., target medical image 108) based on the user's needs.

[0021] Figure 2 An example is illustrated based on one or more embodiments of the present disclosure, using a source medical image 204 (e.g., Figure 1 Medical images 104) identify and enhance objects of interest 202 (e.g., Figure 1 An example of an object of interest (202) in the image. As shown, an object of interest 202 can be identified from a medical image 204 (e.g., separated from the background region 206) by applying one or more image filtering and / or segmentation techniques to the medical image at 208 and generating a probability mask (e.g., a segmentation mask) indicating the object of interest at 212. Various image filters (e.g., Frangi or hybrid Hessian filters) and / or ML models (e.g., learned using deep neural networks) can be used to achieve these goals. For example, a segmentation neural network can be trained to extract image features associated with the object of interest and generate a segmentation mask (e.g., such as...). Figure 2(As shown in 212) to indicate the region (e.g., pixels) in the medical image 204 corresponding to the object of interest. The segmentation neural network may include a convolutional neural network (CNN) having multiple convolutional layers, one or more pooling layers, and / or one or more fully connected layers. Following the convolutional layers may be batch normalization and / or linear or non-linear activation (e.g., rectified linear units or ReLU activation functions). Each convolutional layer may include multiple convolutional kernels or filters with corresponding weights, the values ​​of which can be learned through a training process, such that the kernels or filters can be used to extract features associated with the object of interest from the medical image 204 upon completion of training. These extracted features may be downsampled by one or more pooling layers to obtain a representation of the features, e.g., in the form of a feature map or feature vector. In some examples, the neural network may also include one or more up-pooling layers and one or more transposed convolutional layers. Through the up-pooling layers, the network can upsample the features extracted from the medical image 204, and process the upsampled features through one or more transposed convolutional layers (e.g., via multiple deconvolution operations) to derive an enlarged or dense feature map or feature vector. The dense feature map or vector can then be used to predict regions (e.g., pixels) in the medical image 204 that may belong to the object of interest 202. The prediction can be represented by a segmentation mask 212, which may include a corresponding probability value (e.g., ranging from z to 1) for each image pixel, indicating whether the image pixel may belong to the object of interest 202 (e.g., with a probability value above a pre-configured threshold) or to the background region 206 (e.g., with a probability value below a pre-configured threshold).

[0022] Once the object of interest 202 is identified using the techniques described above, the object (e.g., the contrast between object 202 and background 206) can be enhanced at 214 by adjusting pixels associated with the object and / or the surrounding background based on pixel color / brightness values ​​(e.g., RGB values) and / or the probability that each pixel belongs to the object of interest or the background. The adjusted pixels can then be used to generate a target medical image 208 including the enhanced object of interest (e.g., ...). Figure 1The target image 208). As described above, the contrast between the object of interest and its surrounding region (e.g., background 206) can be enhanced by adding values ​​(e.g., positive or negative constants) to pixels belonging to the object of interest. The contrast between the object of interest and its surrounding region can also be enhanced by applying a sharpening filter (e.g., an unsharpening mask) to the object of interest (e.g., to sharpen the object of interest in the target image 208). Such a sharpening filter (or unsharpening mask) can be applied, for example, by filtering the source medical image 204 (e.g., by a low-pass filter such as a Gaussian filter) to obtain a blurred or smoothed version of the image, subtracting the blurred image from the source medical image 204 (e.g., to produce a high-pass or edge representation of the source image), and adding the subtraction result back to the source medical image 204 to obtain the target medical image 208 (e.g., with sharpened edges).

[0023] The contrast between the object of interest 202 and the background region 206 can also be enhanced by manipulating at least a subset of the pixels associated with the object of interest according to the following.

[0024] foreground_pixels <- all_pixels [probability > threshold] 1)

[0025] foreground_pixels_stretch<-foreground_pixels.max()–

[0026] a*(foreground_pixels.max()–foreground_pixels)2)

[0027] foreground_pixels<-probability*foreground_pixels_stretch+

[0028] (1 - probability) * foreground_pixels 3)

[0029] `all_pixels[probability>threshold]<-foreground_pixels` 4) where "probability" can represent the probability that a pixel is associated with an object of interest (e.g., based on probability or segmentation mask 212), `all_pixels[probability>threshold]` can represent a set of pixels whose "probability" of being associated with an object of interest is higher than "threshold" (e.g., it can be configurable), `foreground_pixels.max()` can represent the maximum pixel value in the set of pixels represented by `all_pixels[probability>threshold]`, "a" can be a value that can be a configurable constant, and `(foreground_pixels.max()–foreground_pixels)` can represent the difference between the maximum pixel value and the pixel values ​​of each `foreground_pixels`. Therefore, using equations 1)-4), the pixel values ​​of at least a subset of the pixels of the source medical image 204 (e.g., pixels with a specific probability of being associated with an object of interest) can be adjusted in proportion to the difference between the individual pixel values ​​and the maximum pixel value before calculating a weighted sum based on the adjusted pixel values ​​and the original pixel values ​​(e.g., using "probability" as weights) to smooth the edges around the object of interest.

[0030] It should be noted that although techniques for identifying or enhancing objects of interest are described individually herein, one or more of these techniques may be applied together to improve the results of the operation. For example, to enhance the contrast between object 202 and background 206, a sharpening filter (e.g., an unsharpening mask) may be applied in conjunction with the operations illustrated in equations 1)-4) to not only highlight fine details of the object of interest but also smooth the edges surrounding the object of interest.

[0031] In the example, the object augmentation task described herein can be performed by transforming the source medical image (e.g., Figure 1 Image 104 or Figure 2 The process involves separating the image (204) into multiple (e.g., two) layers, identifying the object of interest in one layer, and enhancing the identified object in the target medical image. The layer separation can satisfy the following condition: I(t) = w*I1(t) + (1-w)*I2(t), where I(t) can represent the source medical image at time t, I1(t) can represent the first layer of the source medical image that may include the object of interest, I2(t) can represent the second layer of the source medical image that may include the background region surrounding the object of interest, and w can represent the weight of the first layer in the source medical image. Using these layers, the target medical image (e.g., ...) can be enhanced based on the following formula. Figure 1 Target image 108 or Figure 2The values ​​of w in the target image 208 (e.g., including the weights of the first layer of the object of interest) are used to augment the object of interest:

[0032] I e (t)=w'*I1(t)+(1-w)*I2(t)5)

[0033] Among them, I e (t) can represent the target medical image, and w' can represent the increased weight of the first layer applied to the target medical image (e.g., w' > w) (e.g., by multiplying the pixel values ​​of the object of interest by w'). Although Equation 5) shows that the weights of the second layer in the target medical image can remain the same as in the source medical image, those skilled in the art will understand that the weights of the second layer can also be reduced in the target medical image (e.g., by multiplying the pixel values ​​of the background by a value less than (1-w)) to further enhance the contrast between the object of interest and the background.

[0034] Separating a source medical image into multiple layers can be done using various image processing techniques. Figure 3 An example is illustrated using a deep neural network (DNN) to separate a source medical image into two layers (e.g., layer 1 and layer 2), where the first layer may include an object of interest, and the second layer may include the background surrounding the object of interest. DNN (e.g., Figure 3 The DNN 300 shown may include a convolutional neural network with multiple convolutional layers, one or more pooling layers, and / or one or more fully connected layers. Following the convolutional layers may be batch normalization and / or linear or non-linear activation (e.g., modified linear units or ReLU activation functions). Each convolutional layer may include multiple convolutional kernels or filters with corresponding weights, the values ​​of which can be learned through a training process, allowing the kernels or filters to be used to extract features associated with the object of interest and / or background regions from the source medical image upon completion of training. These extracted features may be downsampled through one or more pooling layers to obtain a representation of the features, e.g., in the form of feature maps or feature vectors. In some examples, the DNN 300 may also include one or more up-pooling layers and one or more transposed convolutional layers. Through the up-pooling layers, the network can upsample the features extracted from the source medical image, and process the upsampled features through one or more transposed convolutional layers (e.g., via multiple deconvolution operations) to derive an amplified or dense feature map or feature vector. The dense feature map or vector can then be used to estimate layers in an image that may include the object of interest or background.

[0035] DNN 300 can be trained unsupervised using pairs of images that are closely related in the temporal domain, enabling image registration based on small and smooth motion fields. For example, the training pairs can be consecutive image frames from medical videos such as X-ray videos or movies, and DNN 300 can be trained to achieve motion and / or fidelity regularization. Figure 3 As shown, DNN 300 can obtain source medical images I that can depict objects of interest at time t1 during training iterations. t1 Features are extracted from the source medical image, and based on the extracted features, two layers (e.g., layer 1 and layer 2) can be estimated to form the source medical image. One of the estimated layers (e.g., layer 1) can include the object of interest, while the other estimated layer (e.g., layer 2) can include a background region that can surround the object of interest. During the same training iterations, DNN 300 can also obtain the source medical image I depicting the object of interest at time t2. t2 Where t1 and t2 can be consecutive time points along the time axis of the medical video, from which image I can be obtained. t1 and I t2 DNN 300 can obtain data from source medical images. t2 Features are extracted, and based on the extracted features, two layers (e.g., layer 1 and layer 2) that can constitute the source medical image are estimated. Similar to image I... t1 Image I t2 One of the estimated layers (e.g., layer 1) may include the object of interest, while image I t2 Another layer in the estimated layer (e.g., layer 2) may include a background region surrounding the object of interest.

[0036] Due to image I t1 and I t2 It may be closely related to time, therefore, with image I t1 Compared to the estimated objects of interest included in layer 1, image I is expected to... t2 The objects of interest included in the estimated layer 1 have small and / or smooth motions. Such motions can be represented by a motion field M, and the parameters (e.g., weights) of the DNN 300 can be learned by regularizing (e.g., optimizing) the motion field M based on the following formula:

[0037] min M (L a (M i (I i (t1)),I i (t2))+L b (M i ))6)

[0038] Where i can have a value of 1 or 2 representing layer 1 or layer 2 of the source medical image, and M i (I i (t1) can represent the image I of the estimated motion M based on the time interval t1 and t2. i (t1)(For example, image I) t1 Transformation of layer 1 or layer 2), L a It can be a measurement of the transformed image M i (I i (t1) and target image I i The loss function of the difference (e.g., mean squared error) between (t2) and L b It can be a loss function used to regularize the motion field M (e.g., L). b It can be a combination of L1 loss on the motion field magnitude (such as L1(M)) and L1 loss on the motion field gradient magnitude (such as L1(grad(M))).

[0039] In some examples, DNN 300 can be trained with the additional goal of achieving fidelity regularization (e.g., to ensure that the two estimated layers can be combined to match the original image):

[0040] min w (L c (w*I1(t)+(1-w)*I2(t),I(t)))7) where w*I1(t)+(1-w)*I2(t) can represent the source medical image constructed using the estimated layer 1 (e.g., denoted as I1(t)) and the estimated layer 2 (e.g., denoted as I2(t)), I(t) can represent the original source medical image, and Lc can be a loss function that measures the difference between the constructed source image and the original source image.

[0041] like Figure 3 As shown, the motion- and / or fidelity-related loss calculated using the above techniques can be backpropagated through the DNN 300 (e.g., loss-based gradient descent) to adjust network parameters (e.g., weights) until one or more training criteria are met (e.g., until one or more losses are minimized). Once trained, the DNN 300 can be used (e.g., at inference time) to receive medical images of a target object and separate the medical image into multiple layers that can each include the target object and the background region surrounding the target object.

[0042] Figure 4 Example operations 400 that can be associated with an object of interest in an enhanced medical image are illustrated. These operations can be performed by a system or device (e.g., Figure 6The illustrated system or device performs this action, for example, as a post-processing step. As shown, operation 400 may include obtaining a source medical image including the object of interest at 402 and identifying the object of interest (e.g., pixels or blocks associated with the object of interest) at 404. Identification may be performed, for example, using image filters and / or a pre-trained artificial neural network (ANN). As described herein, the ANN may include a segmentation network trained to generate a segmentation mask for identifying pixels of interest, or a deep neural network trained to separate the source medical image into multiple layers that include the object of interest and a background region surrounding the object of interest. Once the pixels and / or blocks associated with the object of interest are identified, these pixels and / or blocks may be enhanced at 406 to increase the contrast between the object of interest and the background region. Then, at 408, a target medical image with enhanced contrast and / or better visibility of the object of interest may be generated and used to facilitate interventional procedures and / or downstream image processing tasks such as image registration (e.g., between 2D / 3D CT images and X-ray fluoroscopy images), stent enhancement, path mapping, etc.

[0043] Figure 5 Example operations that can be associated with training a neural network (e.g., a segmentation neural network or a layer-separating DNN as described herein) to perform one or more tasks as described herein are illustrated. As shown, the training operation may include initializing the parameters of the neural network (e.g., weights associated with the individual filters or kernels of the neural network) at 502. The parameters may be initialized, for example, based on samples collected from one or more probability distributions or parameter values ​​from another neural network with a similar architecture. The training operation may also include feeding training data (e.g., paired medical images including objects of interest) to the neural network at 504, and causing the neural network to estimate a segmentation mask or image layer at 506. At 508, one or more suitable loss functions (e.g., the loss functions illustrated in Equations 6 and 7) can be used to determine the loss between the estimate and the desired result, and this loss can be evaluated at 510 to determine whether one or more training termination criteria have been met. For example, if the aforementioned loss is below a predetermined threshold, if the change in loss between two training iterations (e.g., between consecutive training iterations) falls below a predetermined threshold, the training termination criteria can be considered met. If it is determined at 510 that the training termination criteria have been met, training can end. Otherwise, the loss can be backpropagated through the neural network at 512 (e.g., based on gradient descent associated with the loss) before training returns to 506.

[0044] For the sake of simplicity, the training steps are depicted and described in a specific order herein. However, it should be understood that training operations can occur in various orders, simultaneously, and / or with other operations not presented or described herein. Furthermore, it should be noted that not all operations that may be included in the training process are depicted and described herein, and not all exemplified operations need to be performed.

[0045] The systems, methods, and / or apparatuses described herein may be implemented using one or more processors, one or more storage devices, and / or other suitable auxiliary devices (such as display devices, communication devices, input / output devices, etc.). Figure 6 This is a block diagram illustrating an example device 600 that can be configured to perform the image enhancement tasks described herein. As shown, device 600 may include a processor (e.g., one or more processors) 602, which may be a central processing unit (CPU), graphics processing unit (GPU), microcontroller, reduced instruction set computer (RISC) processor, application-specific integrated circuit (ASIC), application-specific instruction set processor (ASIP), physical processing unit (PPU), digital signal processor (DSP), field-programmable gate array (FPGA), or any other circuitry or processor capable of performing the functions described herein. Device 600 may also include communication circuitry 604, memory 606, mass storage device 608, input device 610, and / or communication link 612 (e.g., communication bus) through which one or more components shown in the figure may exchange information.

[0046] Communication circuitry 604 can be configured to send and receive information using one or more communication protocols (e.g., TCP / IP) and one or more communication networks, including local area networks (LANs), wide area networks (WANs), the Internet, and wireless data networks (e.g., Wi-Fi, 3G, 4G / LTE, or 5G networks). Memory 606 may include a storage medium (e.g., a non-transitory storage medium) configured to store machine-readable instructions that, when executed, cause processor 602 to perform one or more functions described herein. Examples of machine-readable media may include volatile or non-volatile memory, including but not limited to semiconductor memory (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)), flash memory, etc.). Mass storage device 608 may include one or more disks, such as one or more internal hard disks, one or more removable disks, one or more magneto-optical disks, one or more CD-ROMs or DVD-ROMs, etc., on which instructions and / or data may be stored for operation of processor 602. Input device 610 may include a keyboard, mouse, voice-controlled input device, touch-sensitive input device (e.g., touch screen), etc., for receiving user input from device 600.

[0047] It should be noted that device 600 can operate as a standalone device or can be connected to other computing devices (e.g., networked or clustered) to perform the functions described herein. And even in Figure 6 Only one example of each component is shown in the figure, and those skilled in the art will understand that device 600 may include multiple instances of one or more components shown in the figure.

[0048] Although this disclosure has been described according to certain embodiments and generally associated methods, changes and variations of the embodiments and methods will be apparent to those skilled in the art. Therefore, the above description of exemplary embodiments does not limit this disclosure. Other changes, substitutions, and modifications are possible without departing from the spirit and scope of this disclosure. Furthermore, unless specifically stated otherwise, discussions using terms such as “analyze,” “determine,” “enable,” “identify,” and “modify” refer to the actions and processes of a computer system or similar electronic computing device that manipulate and transform data representing physical (e.g., electronic) quantities within the registers and memories of the computer system into other data representing physical quantities within the computer system's memory or other such information storage, transmission, or display devices.

[0049] It should be understood that the above description is intended to be illustrative and not restrictive. Many other embodiments will become apparent to those skilled in the art upon reading and understanding the above description. Therefore, the scope of this disclosure should be determined by reference to the appended claims and the full scope of their equivalents.

Claims

1. A method for image enhancement, the method comprising: Obtain a source medical image including an object of interest and a background region surrounding the object of interest, wherein the object of interest and the background region exhibit contrast in the source medical image; A segmentation mask associated with the object of interest is determined; based on the segmentation mask, a plurality of first pixels associated with the object of interest and a plurality of second pixels associated with the background region are determined from the source medical image, wherein the plurality of first pixels have corresponding first pixel values ​​and the plurality of second pixels have corresponding second pixel values; and A target medical image is generated based on the plurality of first pixels and the plurality of second pixels, the target medical image having enhanced contrast between the object of interest and the background region, wherein the contrast between the object of interest and the background region is enhanced by adjusting the first pixel value of at least the plurality of first pixels or the second pixel value of the plurality of second pixels; Wherein, the segmentation mask indicates the corresponding probability that each of the plurality of first pixels belongs to the object of interest, and wherein adjusting the first pixel values ​​of the plurality of first pixels includes: At least a subset of the plurality of first pixels is selected based on the corresponding probability that each of the plurality of first pixels belongs to the object of interest; Determine the maximum pixel value among the selected pixels; and The pixel value of each of the selected pixels is adjusted proportionally to the difference between the maximum pixel value and the pixel value of each of the selected pixels.

2. The method according to claim 1, wherein, The segmentation mask is determined using an artificial neural network or an image filter.

3. The method according to claim 1, wherein, Adjusting the first pixel value of at least the plurality of first pixels or the second pixel value of the plurality of second pixels includes adjusting by adding a constant to each of the first pixel value or the second pixel value or applying a sharpening filter to the plurality of first pixels.

4. The method according to claim 1, further comprising: A machine learning (ML) model is used to separate the source medical image into a first layer and a second layer, wherein the first layer includes the object of interest, the second layer includes the background region, a plurality of first pixels are determined based on the first layer, and a plurality of second pixels are determined based on the second layer.

5. The method according to claim 4, further comprising: The corresponding weights of the first layer and the second layer in the source medical image are determined, and the contrast between the object of interest and the background region is enhanced by increasing the weight of the first layer or decreasing the weight of the second layer.

6. The method according to claim 4, wherein, The ML model is trained using a pair of medical images, including a first medical image and a second medical image, wherein the first medical image includes a depiction of the object of interest at a first time and the second medical image includes a depiction of the object of interest at a second time, and the ML model is trained based at least on the motion of the object of interest from the first time to the second time.

7. The method according to claim 1, wherein, The source medical image includes an X-ray image, and the object of interest has a tubular structure.

8. A computer program product comprising instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1-7.

Images