Medical image processing method and ect image processing method

Abstract

The application relates to a medical image processing method and an ECT image processing method. The method comprises the following steps: acquiring a medical image, determining a feature descriptor of the medical image, determining a saliency map corresponding to the medical image according to the feature descriptor, determining a feature similarity distance between the medical image and a full mark image according to the saliency map, and obtaining image quality information of the medical image according to the feature similarity distance. The method can shorten the time for medical image quality evaluation.

Description

Technical Field

[0001] This application relates to the field of medical image technology, and in particular to a medical image processing method and an ECT image processing method. Background Technology

[0002] Medical image quality assessment plays a crucial role in ensuring accurate diagnosis and improving treatment outcomes. Clinically, medical image quality assessment can be composed of both subjective and objective components. Subjective assessment can be conducted by physicians based on their subjective experience and according to pre-determined evaluation criteria, evaluating image quality from a visual perspective. The evaluation results are closer to the actual clinical situation. Objective assessment, on the other hand, usually requires the evaluator to manually locate the liver, aorta, muscle, tumor, etc., delineate the region of interest, and calculate the signal-to-noise ratio, tumor-to-target ratio, lesion detection rate, etc., and then conduct an evaluation based on these results.

[0003] However, subjective assessment is easily affected by the doctor's qualifications and background and the external environment, resulting in poor stability and reproducibility. Objective assessment, on the other hand, is usually time-consuming, requires reviewers to perform a lot of repetitive operations, and incurs costs to organize reviewers to conduct multiple experiments.

[0004] Therefore, there is currently a problem that medical image quality assessment takes a long time. Summary of the Invention

[0005] Therefore, it is necessary to provide a medical image processing method, ECT image processing method, device, computer equipment, computer-readable storage medium, and computer program product that can shorten the processing time in response to the above-mentioned technical problems.

[0006] In a first aspect, this application provides a medical image processing method, comprising:

[0007] Acquire medical images and determine the feature descriptors of the medical images;

[0008] Based on the feature descriptors, determine the saliency map corresponding to the medical image;

[0009] Based on the saliency map, determine the feature similarity distance between the medical image and the perfect score image;

[0010] The image quality information of the medical image is obtained based on the feature similarity distance.

[0011] Secondly, this application provides an ECT image processing method, including:

[0012] Acquire ECT images of the target object;

[0013] The ECT image is segmented to obtain multiple regions of interest;

[0014] Calculate the feature descriptor of the region of interest, and generate a saliency map of the ECT image based on the feature descriptor;

[0015] Based on the saliency map, the image quality information of the ECT image is determined.

[0016] Thirdly, this application provides an ECT image processing method, including:

[0017] Acquire ECT images and anatomical diagrams of the same detection object;

[0018] The anatomical structure diagram is segmented to obtain multiple segmented regions;

[0019] The multiple segmented regions are mapped onto the ECT image to determine multiple regions of interest in the ECT image;

[0020] Calculate the feature descriptor of the region of interest, and generate a saliency map of the ECT image based on the feature descriptor;

[0021] Based on the saliency map, the image quality information of the ECT image is determined.

[0022] Fourthly, this application also provides a medical image processing apparatus, comprising:

[0023] The acquisition module is used to acquire medical images and determine the feature descriptors of the medical images;

[0024] The determination module is used to determine the saliency map corresponding to the medical image based on the feature descriptor;

[0025] The measurement module is used to determine the feature similarity distance between the medical image and the perfect score image based on the saliency map;

[0026] An evaluation module is used to obtain image quality information of the medical image based on the feature similarity distance.

[0027] Fifthly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0028] Acquire medical images and determine the feature descriptors of the medical images;

[0029] Based on the feature descriptors, determine the saliency map corresponding to the medical image;

[0030] Based on the saliency map, determine the feature similarity distance between the medical image and the perfect score image;

[0031] The image quality information of the medical image is obtained based on the feature similarity distance.

[0032] Sixthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0033] Acquire medical images and determine the feature descriptors of the medical images;

[0034] Based on the feature descriptors, determine the saliency map corresponding to the medical image;

[0035] Based on the saliency map, determine the feature similarity distance between the medical image and the perfect score image;

[0036] The image quality information of the medical image is obtained based on the feature similarity distance.

[0037] In a seventh aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0038] Acquire medical images and determine the feature descriptors of the medical images;

[0039] Based on the feature descriptors, determine the saliency map corresponding to the medical image;

[0040] Based on the saliency map, determine the feature similarity distance between the medical image and the perfect score image;

[0041] The image quality information of the medical image is obtained based on the feature similarity distance.

[0042] The aforementioned medical image processing method, ECT image processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product acquire medical images, determine the feature descriptors of the medical images, determine the saliency map corresponding to the medical images based on the feature descriptors, determine the feature similarity distance between the medical images and the perfect score images based on the saliency map, and obtain the image quality information of the medical images based on the feature similarity distance. The feature descriptors can be used to describe the image quality of medical images, generating saliency maps that reflect image quality. Based on the saliency map, the feature similarity distance between the medical images and the perfect score images can be determined, thereby assessing the difference in image quality between the medical images and the perfect score images and obtaining the quality information of the medical images. Since no human intervention is required, the time for medical image quality assessment is shortened. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart illustrating a medical image processing method in one embodiment;

[0045] Figure 2 This is a schematic diagram of a saliency map in one embodiment;

[0046] Figure 3 This is a flowchart illustrating an emission computed tomography (ECT) image quality assessment method based on saliency maps and feature descriptors in one embodiment.

[0047] Figure 4 This is a flowchart illustrating a medical image processing method in another embodiment;

[0048] Figure 5 This is a structural block diagram of a medical image processing device in one embodiment;

[0049] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0052] In one exemplary embodiment, such as Figure 1As shown, a medical image processing method is provided. This embodiment illustrates the method applied to a terminal. It is understood that this method can also be applied to a server, and furthermore, to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0053] Step S102: Acquire medical images and determine the feature descriptors of the medical images.

[0054] The medical image may be, but is not limited to, the ECT image to be evaluated. The feature descriptor may be a parameter describing the image quality, including but not limited to local feature descriptors reflecting local features of the medical image and global feature descriptors reflecting overall features of the medical image.

[0055] In practice, the scanned medical image can be input into the terminal, the terminal can extract the region of interest from the medical image, and generate a local feature descriptor of the medical image based on the region of interest. The terminal can also generate an overall feature descriptor for the entire medical image.

[0056] In practical applications, neural networks can be trained to segment medical images input from terminals and extract regions of interest such as lesions, liver, aortic wall, aortic blood pool, esophagus, spinal cord, head and neck, and brain.

[0057] For areas such as the aortic blood pool, local feature descriptors describing homogeneity can be used. The larger this index is, the more uniform the region of interest.

[0058] ,

[0059] in, This represents the normalized points in the gray-level size zone matrix (GLSZM) of a medical image. The probability distribution, It is a local minimum. The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions.

[0060] For regions such as the aortic wall, esophagus, and spinal cord, local feature descriptors describing continuity can be used. The larger this index is, the more continuous the region of interest is.

[0061] ,

[0062] in, and These are the lengths of the longest and second longest parts of the selected region of interest, respectively.

[0063] For the liver region, a local feature descriptor describing the distribution of false positives can be used. The higher this indicator, the more false positives there are.

[0064] ,

[0065] in, Represents the normalized points in GLSZM The probability distribution, The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions. .

[0066] For structurally complex regions such as the neck and abdomen, local feature descriptors that describe the structure and boundary clarity can be used. The larger this index is, the clearer the structure of the region of interest.

[0067] ,

[0068] in, Represents the normalized points in GLSZM The probability distribution, The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions.

[0069] Based on the maximum density projection map of medical images, an overall feature descriptor can be generated. The higher this index, the greater the noise in the medical image.

[0070] ,

[0071] in, The number of groups representing the intensity of discrete grayscale values. These are the normalized points in GLSZM. The probability distribution, It is a local minimum.

[0072] Step S104: Determine the saliency map corresponding to the medical image based on the feature descriptor.

[0073] A saliency map can be an image that displays the uniqueness of each pixel in a medical image.

[0074] In a specific implementation, the terminal can generate an image based on the mask of the region of interest in the medical image, replace the pixel values ​​of the region of interest in the image with local feature descriptors to obtain an updated image, determine the maximum density projection map of the updated image, and obtain a saliency map.

[0075] The mask, also known as a masking layer, generates an image based on the mask of the region of interest. It can be understood as setting the pixel value of the region of interest in the medical image to 1 (or other relatively large value), and setting the pixel value of other non-interest regions to 0 (or other relatively small value).

[0076] Understandably, since saliency maps display the uniqueness of each pixel in a medical image, while the overall image quality features described by the global feature descriptor do not reflect the uniqueness between different pixels, saliency maps can be generated based solely on local feature descriptors.

[0077] Step S106: Determine the feature similarity distance between the medical image and the perfect score image based on the saliency map.

[0078] The perfect image can be a medical image that has been manually assigned a perfect image quality rating. The feature similarity distance can be a parameter that reflects the feature similarity between images.

[0079] In practice, the terminal can predetermine the perfect score image. After obtaining the saliency map, it can determine the feature similarity distance between the medical image to be evaluated and the perfect score image based on the difference in the overall feature descriptors between them and the difference in the saliency maps between them.

[0080] In practical applications, the overall feature descriptor of the medical image to be evaluated is known as follows: Let the mean vector of the corresponding saliency map be... The covariance matrix is A perfect medical image is pre-generated on the terminal as the perfect image, and its overall feature descriptor is: The mean vector of the corresponding saliency map is The covariance matrix is According to and The differences between them, and and , and The difference between them can be used to obtain the feature similarity distance, and the specific calculation formula is as follows:

[0081] ,

[0082] in, The trace of a matrix is ​​the sum of its diagonal elements.

[0083] Step S108: Obtain the image quality information of the medical image based on the feature similarity distance.

[0084] Among them, image quality information can be information describing the quality of medical images.

[0085] In specific implementation, the terminal can obtain image quality information reflecting the quality of medical images based on the calculated feature similarity distance. For example, the feature similarity distance can be used as image quality information. The smaller the feature similarity distance, the higher the image quality; conversely, the larger the feature similarity distance, the lower the image quality. In addition, image quality information can also be expressed as a function of feature similarity distance, which is not limited in this application.

[0086] Understandably, feature similarity distance can reflect the differences in overall feature descriptors between the medical image to be evaluated and the perfect medical image, as well as the differences in the mean vector and covariance matrix of the saliency map. Since the saliency map is generated based on local feature descriptors, feature similarity distance can reflect the differences in overall and local features between the medical image to be evaluated and the perfect medical image. Based on these differences, the imaging quality of medical images can be evaluated, and the results are more comprehensive and reliable.

[0087] The aforementioned medical image processing method acquires a medical image, determines its feature descriptors, identifies the corresponding saliency map based on the feature descriptors, determines the feature similarity distance between the medical image and the perfect score image based on the saliency map, and obtains the image quality information of the medical image based on the feature similarity distance. The method uses feature descriptors to describe the image quality of a medical image, generates a saliency map reflecting the image quality, and determines the feature similarity distance between the medical image and the perfect score image based on the saliency map. This allows for the assessment of the image quality gap between the medical image and the perfect score image, thus obtaining the quality information of the medical image. Since no manual intervention is required, the time for medical image quality assessment is shortened.

[0088] In an exemplary embodiment, the feature descriptor includes a local feature descriptor; step S102 above may specifically include: extracting the region of interest in the medical image; determining the local feature descriptor of the medical image based on the region of interest; the local feature descriptor reflects the local image quality of the medical image.

[0089] Among these, the local feature descriptor can be a parameter describing the image quality of the region of interest in a medical image. Local image quality can be defined as the image quality of the region of interest.

[0090] In practice, the terminal can segment the medical image to obtain one or more regions of interest (ROIs). For each ROI, the image quality type to be described can be determined based on the corresponding tissue or body part. Then, for the determined image quality type, a corresponding local feature descriptor reflecting the image quality of the ROI can be obtained.

[0091] Image quality types include, but are not limited to, uniformity, continuity, false positive distribution, and sharpness of the region of interest.

[0092] In practical applications, medical images can be segmented based on a pre-trained image segmentation model to obtain masks for each region of interest (ROI) in the medical image. For each ROI, a local feature descriptor can be determined, which may specifically include:

[0093] For areas such as the aortic blood pool, a local feature descriptor describing homogeneity is considered. The larger this index is, the more uniform the region of interest.

[0094] ,

[0095] in, Representing normalized points in a medical image using GLSZM The probability distribution, It is a local minimum. The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions.

[0096] For regions such as the aortic wall, esophagus, and spinal cord, a local feature descriptor describing continuity is considered. The larger this index is, the more continuous the region of interest is.

[0097] ,

[0098] in, and These are the lengths of the longest and second longest parts of the selected region of interest, respectively.

[0099] For the liver region, consider using a local feature descriptor that describes the distribution of false positives. The higher this indicator, the more false positives there are.

[0100] ,

[0101] in, Represents the normalized points in GLSZM The probability distribution, The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions. .

[0102] For structurally complex regions such as the neck and abdomen, local feature descriptors that describe the structure and boundary clarity are considered. The larger this index is, the clearer the structure of the region of interest.

[0103] ,

[0104] in, Represents the normalized points in GLSZM The probability distribution, The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions.

[0105] In this embodiment, by extracting the region of interest (ROI) from the medical image and determining the local feature descriptor of the medical image based on the ROI, the image quality, such as the uniformity, continuity, false positive point distribution, structure, and boundary clarity of the ROI in the medical image, can be quantified. This facilitates automatic evaluation of medical image quality from multiple dimensions and improves the reliability of automatic evaluation of medical images.

[0106] In an exemplary embodiment, the step of extracting the region of interest in a medical image may specifically include: segmenting the medical image based on a trained image segmentation network to obtain the region of interest in the medical image.

[0107] The image segmentation network can be a pre-trained image segmentation model, which can be implemented through, but is not limited to, neural networks.

[0108] In practice, the terminal can pre-train an image segmentation network. When it is necessary to extract the region of interest from a medical image, the medical image can be input into the image segmentation network, and the image segmentation network will automatically segment one or more regions of interest from the medical image.

[0109] In practical applications, a neural network can be pre-trained to segment regions of interest (ROIs) in medical images, such as lesions, liver, aortic wall, aortic blood pool, esophagus, spinal cord, head and neck, and brain. This neural network can output masks of these ROIs. For example, inputting an ECT-reconstructed image into this neural network can automatically generate masks of ROIs such as lesions, liver, aortic wall, aortic blood pool, esophagus, spinal cord, head and neck, and brain.

[0110] It should be noted that the training set for the neural network can be ECT reconstructed images with the region of interest manually delineated.

[0111] In this embodiment, medical images are segmented based on a trained image segmentation network to obtain regions of interest in the medical images, which can improve the speed of image segmentation and thus shorten the time for medical image quality evaluation.

[0112] In an exemplary embodiment, the step of determining the local feature descriptor of a medical image based on the region of interest may specifically include: determining the image quality type to be described for the region of interest based on the tissue location corresponding to the region of interest; the image quality type includes at least one of uniformity, continuity, false positive distribution, and sharpness; and obtaining the local feature descriptor of the medical image based on the image quality type.

[0113] The tissue site can be human tissue or a part of the human body. The image quality type can be the type that describes the image quality of the region of interest.

[0114] In practical implementation, the terminal can identify the tissue site corresponding to the region of interest (ROI). Based on the identified tissue site, it determines the corresponding image quality type of the ROI to be described. For example, when the ROI is identified as the aortic blood pool, it determines the homogeneity of the ROI to be described; when the ROI is identified as the aortic wall, esophagus, or spinal cord, it determines the continuity of the ROI to be described; when the ROI is identified as the liver, it determines the distribution of false positives in the ROI to be described; and when the ROI is identified as a structurally complex region such as the neck or abdomen, it determines the clarity of the structure and boundaries of the ROI to be described. The terminal can predetermine the correspondence between tissue sites and image quality types. After identifying the tissue site, it directly determines the image quality type to be described based on the correspondence. It can also predetermine the mapping relationship corresponding to each image quality type. The mapping relationship can be understood as a calculation formula for the local feature descriptor corresponding to different image quality types. Based on the calculation formula, the local feature descriptor under different image quality types can be obtained.

[0115] In practical applications, for areas such as the aortic blood pool, a local feature descriptor describing uniformity should be considered. The larger this index is, the more uniform the region of interest.

[0116] ,

[0117] in, Representing normalized points in a medical image using GLSZM The probability distribution, It is a local minimum. The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions.

[0118] For regions such as the aortic wall, esophagus, and spinal cord, a local feature descriptor describing continuity is considered. The larger this index is, the more continuous the region of interest is.

[0119] ,

[0120] in, and These are the lengths of the longest and second longest parts of the selected region of interest, respectively.

[0121] For the liver region, consider using a local feature descriptor that describes the distribution of false positives. The higher this indicator, the more false positives there are.

[0122] ,

[0123] in, Represents the normalized points in GLSZM The probability distribution, The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions. .

[0124] For structurally complex regions such as the neck and abdomen, local feature descriptors that describe the structure and boundary clarity are considered. The larger this index is, the clearer the structure of the region of interest.

[0125] ,

[0126] in, Represents the normalized points in GLSZM The probability distribution, The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions.

[0127] In this embodiment, by determining the image quality type of the region of interest based on the tissue location corresponding to the region of interest, and obtaining the local feature descriptor of the medical image based on the image quality type, diverse and personalized descriptions of the image quality of different regions of interest can be provided, increasing the reliability of image quality evaluation.

[0128] In an exemplary embodiment, the feature descriptor includes an overall feature descriptor; step S102 may specifically include: determining the maximum density projection map of the medical image; determining the overall feature descriptor of the medical image based on the grayscale distribution information corresponding to the maximum density projection map; the overall feature descriptor reflects the overall image quality of the medical image.

[0129] The maximum density projection map can be the image obtained by performing maximum density projection on the medical image. Gray-level distribution information can be, but is not limited to, a gray-level region matrix. The overall feature descriptor can be a parameter describing the overall image quality of the medical image. Overall image quality refers to the overall quality of the medical image, including but not limited to noise levels within the image.

[0130] In a specific implementation, the terminal can determine the maximum density projection map of the medical image, determine the gray-level region matrix GLSZM based on the maximum density projection map, and obtain the overall feature descriptor of the medical image based on the probability distribution of the normalized points in GLSZM.

[0131] In practical applications, the overall feature descriptor The calculation formula can be

[0132] ,

[0133] in, The number of groups representing the intensity of discrete grayscale values. These are the normalized points in GLSZM. The probability distribution, It is a local minimum.

[0134] In this embodiment, by determining the maximum density projection map of the medical image and the grayscale distribution information corresponding to the maximum density projection map, the overall feature descriptor of the medical image is determined. This allows for the quantification of the overall image quality of the medical image, facilitating the overall evaluation of the medical image quality and improving the reliability of automatic evaluation of medical images.

[0135] In an exemplary embodiment, step S104 may specifically include: generating a first image based on the region of interest of the medical image; replacing the region of interest in the first image with a local feature descriptor to obtain a second image; and determining the maximum density projection map of the second image to obtain a saliency map.

[0136] The first image can be an image generated by masking a region of interest. The second image can be an image obtained by replacing the pixel values ​​at the masked region of interest in the first image with local feature descriptors.

[0137] In a specific implementation, the terminal can generate a first image based on the mask of the region of interest in the medical image, replace the pixel values ​​of the masked part of the region of interest in the first image with the corresponding local feature descriptors to obtain a second image, and use the maximum density projection map of the second image as the saliency map.

[0138] For example, after identifying the region of interest (ROI) in a medical image, the pixel values ​​of the ROI can be set to 1 (or other relatively large values), while the pixel values ​​of other non-ROI regions can be set to 0 (or other relatively small values), resulting in a first image. For the pixel values ​​set to 1 in the first image, they can be replaced with the corresponding local feature descriptors of the ROI, resulting in a second image. For instance, if the ROI is the aortic vascular pool, then the pixel values ​​of the ROI in the resulting second image would be... The pixel values ​​of other non-interest regions are 0. The second image is then subjected to maximum density projection to obtain a saliency map.

[0139] Figure 2 A schematic diagram of a saliency map is provided for reference. Figure 2 The pixel values ​​of regions of interest, such as the aortic blood pool, can be replaced with descriptors. The pixel values ​​of regions of interest, such as the aortic wall, esophagus, and spinal cord, are replaced with descriptors. Replace the pixel values ​​of regions of interest, such as the liver, with descriptors. Replace pixel values ​​of regions of interest such as the neck and abdomen with descriptors. Then, the maximum density projection map is calculated to obtain the saliency map.

[0140] In this embodiment, a first image is generated based on the region of interest (ROI) of the medical image. The ROI in the first image is then replaced with local feature descriptors to obtain a second image. The maximum density projection map of the second image is determined to obtain a saliency map. This allows the local feature descriptors of the ROI to be converted into a description using a saliency map, which facilitates comparison with a perfect score medical image and shortens the evaluation time of the medical image.

[0141] In an exemplary embodiment, step S106 may specifically include: determining the mean vector and covariance matrix of the saliency map; and determining the feature similarity distance based on the difference in the overall feature descriptors between the medical image and the perfect score image, as well as the difference in the mean vector and covariance matrix between the saliency map and the perfect score saliency map.

[0142] Among them, the perfect saliency map can be a saliency map generated based on a medical image whose image quality is rated as perfect by a human setting.

[0143] In specific implementation, the overall feature descriptor of the medical image to be evaluated is known as follows: Let the mean vector of the corresponding saliency map be... The covariance matrix is A perfect medical image is pre-generated on the terminal as the perfect image, and its overall feature descriptor is: The mean vector of the corresponding full-score saliency plot is The covariance matrix is According to and The differences between them, and and , and The difference between them can be used to obtain the feature similarity distance, and the specific calculation formula is as follows:

[0144] ,

[0145] in, The trace of a matrix is ​​represented by the sum of its diagonal elements. In practical applications, it can be expressed as... and Add weighting coefficients to adjust the proportion of local features in image quality assessment.

[0146] In this embodiment, by determining the mean vector and covariance matrix of the saliency map, and based on the differences in the overall feature descriptors between the medical image and the perfect score image, as well as the differences in the mean vector and covariance matrix between the saliency map and the perfect score saliency map, the feature similarity distance is determined. This can comprehensively describe the feature similarity between the medical image to be evaluated and the perfect score medical image, thereby improving the reliability of image quality assessment.

[0147] To facilitate a deeper understanding of the embodiments of this application by those skilled in the art, a specific example will be used for illustration below.

[0148] This application proposes an ECT image quality assessment method based on saliency maps and feature descriptors. It constructs multi-dimensional descriptors to quantify the subjective and objective evaluations of medical image quality. At the same time, it integrates multiple descriptors using a two-dimensional saliency map and uses the feature similarity distance of the saliency map as a comprehensive evaluation of image quality. This method is independent of equipment and algorithms, has strong generalization ability and interpretability, and can be freely configured according to the requirements of clinicians and technicians, reducing the human and material costs of medical image assessment.

[0149] Figure 3 A flowchart illustrating an ECT image quality assessment method based on saliency maps and feature descriptors is provided. Figure 3 The method includes the following steps:

[0150] Step S201: Perform a static ECT scan and reconstruct the image obtained from the scan to obtain the medical image to be evaluated;

[0151] Step S202: Train a segmentation neural network to segment and extract different regions of interest, including background, target organs and tissues, tumors, etc.

[0152] Step S203: Construct feature descriptors for the region of interest;

[0153] Step S204: Generate a saliency map of the medical image using the constructed feature descriptors;

[0154] Step S205: Construct a saliency map of the perfect score image and calculate the feature similarity distance between the saliency maps of the medical image to be evaluated and the perfect score image;

[0155] Step S206: Image quality assessment is performed based on feature similarity distance.

[0156] In one embodiment, step S202 may specifically include: training a segmentation neural network to extract regions of interest (ROIs) of the background and the tumor. Specifically, this may involve training a segmentation neural network to segment the following ROIs: lesion, liver, aortic wall, aortic blood pool, esophagus, spinal cord, head and neck, and brain. The network input may be an ECT reconstructed image; the training target may be a mask of the aforementioned ROIs; and the training set may be obtained by manually delineating the ROIs on an ECT reconstructed image.

[0157] In one embodiment, step S203 may specifically include: constructing feature descriptors for the region of interest. These feature descriptors are divided into the following two categories:

[0158] (1) Local feature descriptor:

[0159] The local feature descriptor is generated based on the local region of interest extracted in step S202, representing the local image quality, and may specifically include:

[0160] Description of homogeneity: For areas such as the aortic blood pool, the following features are considered to quantitatively describe their homogeneity. The larger the index, the more homogeneous the region of interest.

[0161] ,

[0162] in, Representing normalized points in a medical image using GLSZM The probability distribution, It is a local minimum. The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions.

[0163] Describing Continuity: For regions such as the aortic wall, esophagus, and spinal cord, the following features are considered to quantitatively describe their continuity. The larger the index, the more continuous the region of interest.

[0164] ,

[0165] in, and These are the lengths of the longest and second longest parts of the selected region of interest, respectively.

[0166] Describing the distribution of false positives: For the liver region, the following feature is considered to describe the distribution of false positives; a higher value indicates a greater number of false positives.

[0167] ,

[0168] in, Represents the normalized points in GLSZM The probability distribution, The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions. .

[0169] Describing the clarity of structure and boundaries: For structurally complex regions such as the neck and abdomen, consider using the following features to describe the clarity of their structure and boundaries. The higher the value of this index, the clearer the structure of the region of interest.

[0170] .

[0171] in, Represents the normalized points in GLSZM The probability distribution, The number of groups representing the intensity of discrete grayscale values. The number of types representing the size of discrete grayscale regions.

[0172] (2) Overall feature descriptor:

[0173] The overall feature descriptor is generated based on the maximum density projection map of the 3D ECT image, representing the overall image quality. A larger value indicates greater image noise. Consider using the following features to describe it.

[0174] ,

[0175] in, The number of groups representing the intensity of discrete grayscale values. These are the normalized points in GLSZM. The probability distribution, It is a local minimum.

[0176] In one embodiment, step S204 may specifically include: generating a two-dimensional saliency map using feature descriptors.

[0177] Based on the aforementioned feature descriptors, a corresponding two-dimensional saliency map is generated. The saliency map is a maximum density projection map generated by segmenting and masking all regions of interest in step S202, and the values ​​of the masked portions are the corresponding feature descriptors.

[0178] In one embodiment, step S205 may specifically include: generating a full-score saliency map and calculating the feature similarity distance of the saliency map.

[0179] First, a two-dimensional saliency map of the perfect-score image is generated. The feature similarity distance between the medical image to be evaluated and the perfect-score image is defined as the Fréchet Inception Distance. The difference between their global feature descriptors is then added. The formula for calculating the feature similarity distance is as follows:

[0180] ,

[0181] in, and These are global feature descriptors for the perfect score image and the medical image to be evaluated, respectively. Assume the perfect score image... and images to be evaluated It is a multidimensional Gaussian distribution with the following parameters: and ,in and These are the mean vector and the covariance matrix, respectively. The trace of the matrix (the sum of its diagonal elements).

[0182] It should be noted that the segmentation of the region of interest mentioned above can be performed not only on ECT images, but also on computed tomography (CT) or magnetic resonance (MR) images. The results are then mapped onto the ECT image. In one embodiment, ECT and CT (or MR) images of the same object can be acquired, the CT (or MR) image can be segmented to obtain the segmented region, and the segmented region can be mapped onto the ECT image to obtain the region of interest of the ECT image.

[0183] It should also be noted that perfect scores can be obtained by users training their own networks. For example, high-quality images can be collected daily, used to train a network model, and then the trained network model can be used to generate perfect scores. Perfect scores can also be generated through simulation. For example, a noise-free image can be generated through Monte Carlo simulation and used as the perfect score image.

[0184] Furthermore, the weights of each descriptor in the saliency map can be manually configured or obtained using unsupervised clustering. For example, they can be assigned to the weights in the formula for calculating feature similarity distance. and Add weight coefficients. The specific values ​​of the weight coefficients can be configured manually by the user or obtained through unsupervised clustering.

[0185] The aforementioned ECT image quality assessment method based on saliency maps and feature descriptors can not only automate the clinical evaluation process, save labor costs, and shorten the evaluation time, but also quantify subjective evaluation indicators, increase the stability and generalization ability of evaluation results. In addition, it can assist in the selection of hyperparameters for image reconstruction algorithms and efficiently evaluate new images generated by new equipment and algorithms.

[0186] In an exemplary embodiment, an ECT image processing method is provided, which is illustrated by taking the application of the method to a terminal as an example, and includes the following steps:

[0187] Step S301: Obtain the ECT image of the object to be detected;

[0188] Step S302: Segment the ECT image to obtain multiple regions of interest;

[0189] Step S303: Calculate the feature descriptor of the region of interest, and generate a saliency map of the ECT image based on the feature descriptor;

[0190] Step S304: Based on the saliency map, determine the image quality information of the ECT image.

[0191] The object to be detected can be the object scanned by the ECT device.

[0192] In practice, an ECT image is obtained by scanning the object to be detected. This ECT image is then input into a terminal, which segments multiple regions of interest (ROIs) from the image. For each ROI, the terminal calculates corresponding feature descriptors, replaces the pixel values ​​corresponding to the masked ROI with the calculated feature descriptors, and generates a maximum density projection map as the saliency map of the ECT image. The image quality information of the ECT image is then determined based on the saliency map. The feature descriptors corresponding to different ROIs are local feature descriptors, including but not limited to descriptors describing features such as uniformity, continuity, false positive distribution, and sharpness of the ROIs.

[0193] In this embodiment, by acquiring the ECT image of the object to be detected, segmenting the ECT image to obtain multiple regions of interest, calculating the feature descriptors of the regions of interest, and generating a saliency map of the ECT image based on the feature descriptors, the image quality information of the ECT image is determined based on the saliency map. Different saliency maps and image quality information can be obtained for different regions of interest, making the image quality assessment compatible with human body parts. Different features are used to assess image quality in different human body parts, increasing the accuracy of image quality assessment.

[0194] In an exemplary embodiment, step S304 may specifically include: obtaining a reference saliency map; calculating the similarity between the saliency map of the ECT image and the reference saliency map; and determining the image quality information of the ECT image based on the similarity.

[0195] The reference saliency map can be a saliency map that has received a perfect score in image quality assessment. Both the saliency map and the reference saliency map reflect the local features of the region of interest. Similarity refers to the degree of similarity, including but not limited to histogram statistics, mean squared error, cosine similarity, and similarity distance.

[0196] In practice, the terminal can obtain a reference saliency map and calculate the similarity between the saliency map of the generated ECT image and the reference saliency map. This calculation includes, but is not limited to, methods such as histogram comparison, mean squared error, convolutional neural networks, and similarity distance. Based on the similarity, the image quality information of the ECT image is determined. For example, for a region of interest, the higher the similarity between the saliency map and the reference saliency map, the better the quality of that region of interest in the ECT image. Conversely, the lower the similarity between the saliency map and the reference saliency map, the worse the quality of that region of interest in the ECT image.

[0197] In this embodiment, by obtaining a reference saliency map, the similarity between the saliency map of the ECT image and the reference saliency map is calculated, and the image quality information of the ECT image is determined based on the similarity. The image quality information of the region of interest in the ECT image can be quickly determined by using the similarity of the saliency maps, thus shortening the time for image quality assessment.

[0198] In one exemplary embodiment, another ECT image processing method is provided, which is illustrated by taking the application of this method to a terminal as an example, and includes the following steps:

[0199] Step S401: Obtain ECT images and anatomical structure diagrams of the same detection object;

[0200] Step S402: Segment the anatomical structure diagram to obtain multiple segmented regions;

[0201] Step S403: Map multiple segmented regions onto the ECT image to determine multiple regions of interest in the ECT image;

[0202] Step S404: Calculate the feature descriptor of the region of interest and generate a saliency map of the ECT image based on the feature descriptor;

[0203] Step S405: Based on the saliency map, determine the image quality information of the ECT image.

[0204] The anatomical structure image can be an image reflecting the anatomical structure of the object being tested, including but not limited to CT or MR images. The segmented region can be the image region obtained after segmentation.

[0205] In practice, the same object can be scanned to obtain its ECT image and anatomical structure map. There is a certain mapping relationship between the ECT image and the anatomical structure map. The ECT image and the anatomical structure map are input into the terminal. The terminal segments the anatomical structure map into multiple segments and maps each segment to the corresponding region of the ECT image according to the mapping relationship, thus obtaining multiple regions of interest (ROIs) of the ECT image. For different ROIs, the terminal can calculate the corresponding feature descriptors, replace the pixel values ​​corresponding to the ROI mask with the calculated feature descriptors, and generate the corresponding maximum density projection map as the saliency map of the ECT image. Then, the image quality information of the ECT image is determined based on the saliency map.

[0206] In this embodiment, by acquiring ECT images and anatomical structures of the same detection object, the anatomical structure image is segmented to obtain multiple segmented regions. These multiple segmented regions are then mapped onto the ECT image to determine multiple regions of interest (ROIs) in the ECT image. Feature descriptors for the ROIs are calculated, and a saliency map of the ECT image is generated based on the feature descriptors. Based on the saliency map, the image quality information of the ECT image is determined. The ROIs in the ECT image can be obtained by segmenting the anatomical structure image, which reduces the complexity of ROI identification.

[0207] In one embodiment, such as Figure 4 As shown, a medical image processing method is provided. Taking the application of this method to a terminal as an example, the method includes the following steps:

[0208] Step S501: Acquire a medical image, extract the region of interest in the medical image, determine the local feature descriptor of the medical image based on the region of interest, determine the maximum density projection map of the medical image, and determine the overall feature descriptor of the medical image based on the gray-scale distribution information corresponding to the maximum density projection map.

[0209] Step S502: Generate a first image based on the region of interest, replace the region of interest in the first image with a local feature descriptor to obtain a second image, determine the maximum density projection map of the second image, and obtain a saliency map;

[0210] Step S503: Determine the mean vector and covariance matrix of the saliency map. Based on the difference in the overall feature descriptors between the medical image and the perfect score image, and the difference in the mean vector and covariance matrix between the saliency map and the perfect score saliency map, determine the feature similarity distance.

[0211] Step S504: Obtain image quality information of the medical image based on the feature similarity distance.

[0212] In practice, the terminal can acquire the medical image to be evaluated, segment the medical image using a trained image segmentation model, extract the region of interest, determine the local feature descriptor of the medical image based on the region of interest, and determine the maximum density projection map of the medical image. Based on the maximum density projection map, the terminal determines the gray-level region matrix, and then obtains the overall feature descriptor of the medical image based on the gray-level region matrix. After that, the terminal can generate a saliency map based on the local feature descriptor. Based on the difference in the overall feature descriptor between the medical image and the perfect score image, as well as the difference in the mean vector and covariance matrix between the saliency map and the corresponding perfect score saliency map, the terminal calculates the feature similarity distance and uses the feature similarity distance to determine the medical image quality assessment result.

[0213] The aforementioned medical image processing method can use feature descriptors to describe the image quality of medical images, generate a saliency map that reflects the image quality, determine the feature similarity distance between the medical image and the perfect score image based on the saliency map, and evaluate the image quality gap between the medical image and the perfect score image to obtain the quality information of the medical image. Since no human intervention is required, the time for medical image quality assessment is shortened.

[0214] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0215] Based on the same inventive concept, this application also provides a medical image processing apparatus for implementing the medical image processing method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more medical image processing apparatus embodiments provided below can be found in the limitations of the medical image processing method described above, and will not be repeated here.

[0216] In one exemplary embodiment, such as Figure 5 As shown, a medical image processing device is provided, including: an acquisition module 602, a determination module 604, a measurement module 606, and an evaluation module 608, wherein:

[0217] Acquisition module 602 is used to acquire medical images and determine the feature descriptors of the medical images;

[0218] The determining module 604 is used to determine the saliency map corresponding to the medical image based on the feature descriptor;

[0219] The measurement module 606 is used to determine the feature similarity distance between the medical image and the perfect score image based on the saliency map;

[0220] Evaluation module 608 is used to obtain image quality information of the medical image based on the feature similarity distance.

[0221] In an exemplary embodiment, the acquisition module 602 is further configured to extract the region of interest in the medical image; determine the local feature descriptor of the medical image based on the region of interest; and the local feature descriptor reflects the local image quality of the medical image.

[0222] In an exemplary embodiment, the acquisition module 602 is further configured to segment the medical image based on a trained image segmentation network to obtain the region of interest in the medical image.

[0223] In an exemplary embodiment, the acquisition module 602 is further configured to determine the image quality type to be described for the region of interest based on the tissue location corresponding to the region of interest; the image quality type includes at least one of uniformity, continuity, false positive distribution, and sharpness; and obtain the local feature descriptor of the medical image based on the image quality type.

[0224] In an exemplary embodiment, the acquisition module 602 is further configured to determine the maximum density projection map of the medical image; and to determine the overall feature descriptor of the medical image based on the grayscale distribution information corresponding to the maximum density projection map; the overall feature descriptor reflects the overall image quality of the medical image.

[0225] In an exemplary embodiment, the determining module 604 is further configured to generate a first image based on the region of interest of the medical image; replace the region of interest in the first image with a local feature descriptor to obtain a second image; and determine the maximum density projection map of the second image to obtain the saliency map.

[0226] In an exemplary embodiment, the measurement module 606 is further configured to determine the mean vector and covariance matrix of the saliency map; and to determine the feature similarity distance based on the difference in the overall feature descriptors between the medical image and the perfect score image, and the difference in the mean vector and covariance matrix between the saliency map and the perfect score saliency map.

[0227] In one exemplary embodiment, an ECT image processing apparatus is provided, comprising:

[0228] The first image acquisition module is used to acquire the ECT image of the object to be detected;

[0229] The first image segmentation module is used to segment the ECT image to obtain multiple regions of interest;

[0230] The first image calculation module is used to calculate the feature descriptor of the region of interest and generate a saliency map of the ECT image based on the feature descriptor;

[0231] The first quality assessment module is used to determine the image quality information of the ECT image based on the saliency map.

[0232] In an exemplary embodiment, the quality assessment module described above is further configured to: obtain a reference saliency map; calculate the similarity between the saliency map of the ECT image and the reference saliency map; and determine the image quality information of the ECT image based on the similarity.

[0233] In one exemplary embodiment, another ECT image processing apparatus is provided, comprising:

[0234] The second image acquisition module is used to acquire ECT images and anatomical structure diagrams of the same detection object;

[0235] The second image segmentation module is used to segment the anatomical structure image to obtain multiple segmented regions;

[0236] The second region mapping module is used to map the multiple segmented regions to the ECT image to determine multiple regions of interest in the ECT image;

[0237] The second image calculation module is used to calculate the feature descriptor of the region of interest and generate a saliency map of the ECT image based on the feature descriptor;

[0238] The second quality assessment module is used to determine the image quality information of the ECT image based on the saliency map.

[0239] Each module in the aforementioned medical image processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0240] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a medical image processing method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0241] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0242] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0243] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0244] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0245] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0246] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0247] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0248] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A medical image processing method, characterized in that, The method includes: Acquire medical images and determine the feature descriptors of the medical images; Based on the feature descriptors, determine the saliency map corresponding to the medical image; Based on the saliency map, determine the feature similarity distance between the medical image and the perfect score image; The image quality information of the medical image is obtained based on the feature similarity distance.

2. The method according to claim 1, characterized in that, The feature descriptor includes a local feature descriptor; determining the feature descriptor of the medical image includes: Extract the region of interest from the medical image; Based on the region of interest, the local feature descriptor of the medical image is determined; the local feature descriptor reflects the local image quality of the medical image.

3. The method according to claim 2, characterized in that, Extracting the region of interest from the medical image includes: The medical image is segmented based on a trained image segmentation network to obtain the region of interest in the medical image.

4. The method according to claim 2, characterized in that, Determining the local feature descriptor of the medical image based on the region of interest includes: Based on the tissue location corresponding to the region of interest, the image quality type to be described for the region of interest is determined; the image quality type includes at least one of uniformity, continuity, false positive distribution, and sharpness; Based on the image quality type, the local feature descriptor of the medical image is obtained.

5. The method according to claim 1, characterized in that, The feature descriptor includes a global feature descriptor; determining the feature descriptor of the medical image includes: Determine the maximum density projection map of the medical image; Based on the grayscale distribution information corresponding to the maximum density projection map, the overall feature descriptor of the medical image is determined; the overall feature descriptor reflects the overall image quality of the medical image.

6. The method according to claim 1, characterized in that, The step of determining the saliency map corresponding to the medical image based on the feature descriptor includes: A first image is generated based on the region of interest in the medical image; The region of interest in the first image is replaced with a local feature descriptor to obtain the second image; The maximum density projection map of the second image is determined to obtain the saliency map.

7. The method according to claim 1, characterized in that, The step of determining the feature similarity distance between the medical image and the perfect score image based on the saliency map includes: Determine the mean vector and covariance matrix of the saliency map; The feature similarity distance is determined based on the difference in the overall feature descriptors between the medical image and the perfect score image, and the difference in the mean vector and covariance matrix between the saliency map and the perfect score saliency map.

8. An ECT image processing method, characterized in that, The method includes: Acquire ECT images of the target object; The ECT image is segmented to obtain multiple regions of interest; Calculate the feature descriptor of the region of interest, and generate a saliency map of the ECT image based on the feature descriptor; Based on the saliency map, the image quality information of the ECT image is determined.

9. The method according to claim 8, characterized in that, The process of determining the image quality information of the ECT image based on the saliency map includes: Obtain a reference saliency map; Calculate the similarity between the saliency map of the ECT image and the reference saliency map; The image quality information of the ECT image is determined based on the similarity.

10. An ECT image processing method, characterized in that, The method includes: Acquire ECT images and anatomical diagrams of the same detection object; The anatomical structure diagram is segmented to obtain multiple segmented regions; The multiple segmented regions are mapped onto the ECT image to determine multiple regions of interest in the ECT image; Calculate the feature descriptor of the region of interest, and generate a saliency map of the ECT image based on the feature descriptor; Based on the saliency map, the image quality information of the ECT image is determined.

Images