Image processing method, apparatus, device, and storage medium

Abstract

This invention provides an image processing method, apparatus, device, and storage medium. The method includes: acquiring multiple images containing a target object; performing registration processing on the multiple images to obtain multiple registered images; segmenting the target object in the multiple registered images according to a first template image corresponding to the target reference object to obtain multiple segmented images, each segmented image including multiple segmented regions; extracting features from each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images; fusing the multiple image features corresponding to the same segmented region in the multiple segmented images to obtain multiple fused image features corresponding to each segmented region; and determining the recognition result corresponding to the target object based on the multiple fused image features, thereby improving the quality and effect of image processing.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an image processing method, apparatus, device, and storage medium. Background Technology

[0002] The 21st century is an information age, and images, as the visual foundation for human perception of the world, are an important means for humans to acquire, express, and transmit information. Digital image processing, which uses computers to process, analyze, and understand images to achieve desired results, is mainly applied in fields such as facial recognition, image reconstruction, machine vision, and medical imaging.

[0003] In image processing, feature extraction directly impacts the processing results. Existing image feature extraction techniques require high resolution and signal-to-noise ratio (SNR) from the images being processed. However, in practical applications, many scenarios do not allow for the direct acquisition of images with high SNR and resolution, thus affecting the processing outcomes. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide an image processing method, apparatus, device, and storage medium to improve the quality and effect of image processing, thereby making the recognition results of target objects more accurate.

[0005] In a first aspect, embodiments of the present invention provide an image processing method, comprising:

[0006] Retrieve multiple images containing the target object;

[0007] The multiple images are registered to obtain multiple registered images;

[0008] Based on the first template image corresponding to the target reference object, the target object in the multiple registration images is segmented to obtain multiple segmented images. The target reference object and the target object are of the same type but different in shape. The first template image includes the segmentation result of the target reference object, and the segmented image includes multiple segmented regions.

[0009] Feature extraction is performed on each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images;

[0010] Multiple image features corresponding to the same segmented region in the multiple segmented images are fused to obtain multiple fused image features corresponding to each segmented region;

[0011] Based on the multiple fused image features, the recognition result corresponding to the target object is determined.

[0012] In a second aspect, embodiments of the present invention provide an image processing apparatus, comprising:

[0013] The acquisition module is used to acquire multiple images containing the target object;

[0014] A registration module is used to perform registration processing on the multiple images to obtain multiple registered images;

[0015] The segmentation module is used to segment the target object in the multiple registration images according to the first template image corresponding to the target reference object, so as to obtain multiple segmented images. The target reference object and the target object are of the same type but different in shape. The first template image includes the segmentation result of the target reference object, and the segmented image includes multiple segmented regions.

[0016] The feature extraction module is used to extract features from each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images.

[0017] The fusion module is used to fuse multiple image features corresponding to the same segmentation region in the multiple segmented images to obtain multiple fused image features corresponding to each segmentation region.

[0018] The determination module is used to determine the recognition result corresponding to the target object based on the multiple fused image features.

[0019] Thirdly, embodiments of the present invention provide an electronic device including a processor and a memory, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the image processing method described in the first aspect. The electronic device may also include a communication interface for communicating with other devices or communication networks.

[0020] Fourthly, embodiments of the present invention provide a non-transitory machine-readable storage medium storing executable code, which, when executed by a processor of an electronic device, enables the processor to at least implement the image processing method as described in the first aspect.

[0021] The image processing method provided in this invention first acquires multiple images containing a target object. Then, it performs registration processing on the multiple images to obtain multiple registered images, ensuring that the target object in the multiple registered images has the same spatial location. Next, based on the segmentation result of the target reference object in the first template image, it directly segments the multiple registered images to obtain multiple segmented images. The target reference object and the target object have the same type but different shapes, and the segmented images include multiple segmented regions. Then, it extracts features from each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images. These multiple image features of the same segmented region in the multiple segmented images are then fused to obtain multiple fused image features. Finally, based on the multiple fused image features, it determines the recognition result corresponding to the target object.

[0022] In the above scheme, features are extracted from each segmented region in multiple segmented images containing the target object. Multiple image features from the same segmented region in these multiple images are then fused to obtain multiple fused image features. Based on these fused image features, the recognition result corresponding to the target object is determined. That is, by processing multiple images containing the target object simultaneously, the influence of the image quality of a single image on feature extraction can be effectively avoided. Furthermore, feature extraction from each segmented region allows for targeted feature extraction, resulting in more accurate extracted image features. This not only ensures the quality and effect of image processing, making the recognition result of the target object more accurate, but also reduces the difficulty of image processing, making the image processing method widely applicable to various application scenarios. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart of an image processing method provided in an embodiment of the present invention;

[0025] Figure 2 This is a flowchart illustrating the registration process of multiple images to obtain multiple registered images, provided in an embodiment of the present invention.

[0026] Figure 3This embodiment provides a flowchart for extracting features from each segmented region in multiple segmented images to obtain multiple image features corresponding to each segmented region in multiple segmented images;

[0027] Figure 4 This is a flowchart illustrating an image processing method provided in this embodiment;

[0028] Figure 5 This is a schematic diagram illustrating the application of an image processing method provided by an embodiment of the present invention in a medical setting;

[0029] Figure 6 This is a schematic diagram of the structure of an image processing device provided in an embodiment of the present invention;

[0030] Figure 7 To and Figure 6 The illustrated embodiment provides a schematic diagram of the electronic device corresponding to the image processing apparatus. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in the embodiments of this invention and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. “Multiple” generally includes at least two, but does not exclude the inclusion of at least one.

[0033] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0034] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to identification.” Similarly, depending on the context, the phrases “if determination” or “if identification (of the condition or event of the statement)” can be interpreted as “when determination” or “in response to determination” or “when identification (of the condition or event of the statement)” or “in response to identification (of the condition or event of the statement).”

[0035] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes said element.

[0036] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Where there is no conflict between the embodiments, the following embodiments and features can be combined with each other. Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0037] Image processing technology is the technique of using computers to process image information. It involves manipulating image information to meet human visual and psychological needs, as well as practical application requirements. Image processing mainly includes image digitization, image enhancement and restoration, image data encoding, image segmentation, and image recognition. Image recognition involves extracting features from an image, classifying the image based on its geometric and textural features, and performing structural analysis of the entire image. Typically, image preprocessing is performed before recognition, including filtering noise and interference, improving contrast, enhancing edges, and geometric correction. Image recognition has extremely wide applications, such as in industrial automatic control systems, fingerprint recognition systems, and in medicine, such as cancer cell identification and functional structure recognition.

[0038] However, existing image recognition technologies place high demands on the resolution and signal-to-noise ratio of the images being processed during feature extraction and recognition. In practical applications, many scenarios do not allow for direct acquisition of high signal-to-noise ratio and high-resolution images. Preprocessing of low-quality images is necessary before image recognition operations. However, preprocessing may alter certain image features, affecting the processing quality and effectiveness. To address these issues, the image processing methods provided in the following embodiments of this invention can be used. These methods not only ensure the quality and effectiveness of image processing but also allow direct processing of ordinary low-quality images, reducing the complexity of image processing and enabling the method to be applied to various application scenarios.

[0039] Before describing in detail the image processing methods provided in the various embodiments of the present invention, the application scenarios of image processing are illustrated below:

[0040] Image processing methods have wide applications in practical fields, such as face recognition, image reconstruction, and medical imaging. Taking face recognition as an example, when recognizing a face, an image containing the face is acquired, and the face can be considered the target object. At this point, the image processing device can perform image processing on the acquired image containing the target object to obtain the recognition result corresponding to the target object.

[0041] Taking medical imaging as an example, medical images can be processed. For instance, medical images may contain various functional structures such as the human brain, liver, and lungs, which can be considered as target objects. In this case, the image processing device can process the acquired image containing the target object to obtain the recognition result corresponding to the target object.

[0042] Based on the above description, Figure 1 This is a flowchart illustrating an image processing method provided in an embodiment of the present invention. The image processing method provided in this embodiment can be executed by an image processing device, which can specifically be an electronic device with data processing capabilities, such as a server. Figure 1 As shown, the method includes the following steps:

[0043] S101. Obtain multiple images containing the target object.

[0044] The images can be facial images, or they can include magnetic resonance imaging (MRI) of the human body, computed tomography (CT) scans of any part of the human body, and so on. Furthermore, to ensure image quality and reduce the impact of the quality of a single image on the target object recognition results, multiple images of the target object can be acquired simultaneously. These multiple images can be acquired within a relatively short period of time.

[0045] Since the target object may appear blurry in the image due to factors such as scanning angle during the acquisition process, multiple identical images containing the target object can be directly acquired to better process the image. Optionally, the multiple images containing the target object can be two-dimensional images. If the target object is a three-dimensional object, then the multiple images can also be multiple layers of images corresponding to the target object, so that the complete three-dimensional target object can be described through each layer of images.

[0046] When processing images, multiple images containing the target object can be directly acquired from user input, or multiple images containing the target object can be obtained by directly communicating with data acquisition devices. This can be done by communicating with different data acquisition devices to receive images containing the target object acquired by these devices. In this embodiment of the invention, the specific implementation method for acquiring multiple images containing the target object is not limited. The data acquisition device can be a camera, a superconducting magnetic resonance scanner, etc.

[0047] S102. Perform registration processing on multiple images to obtain multiple registered images.

[0048] In practice, when acquiring images of a target object, the position of the target object in multiple acquired images may be deviated. Therefore, in order to facilitate subsequent image processing, after acquiring multiple images containing the target object, the multiple images are registered to obtain multiple registered images, so that the target object in the multiple registered images has the same spatial position.

[0049] Image registration refers to selecting one image from multiple images as a reference image (i.e., the first image), and aligning all other images to that reference image. In practical applications, the first image can be, for example, the earliest image captured among the multiple images, but it is not limited to this and can be any image among the multiple images. Assuming that images F1, F2, and F3 were captured sequentially, and image F1 is used as the first image, then images F2 and F3 need to be aligned to image F1 respectively to obtain the registered images F2' and F3'.

[0050] In an optional embodiment, an image corresponding to the target reference object can be acquired, and then the image containing the target object can be registered based on the image corresponding to the target reference object. This ensures that the registered image has a spatial location consistent with the target reference object, allowing for better direct segmentation of the target object in multiple registered images based on the segmentation results of the target object. The target reference object is of the same type as the target object but has a different shape; that is, the target reference object is of the same type and has a very similar shape to the target object. A reference object with a similar shape to the target object can be selected as the target reference object. For example, if the target object can be the brain structure of a 6-month-old infant, then the target reference object can be the standard brain structure of a 24-month-old infant. Both are infant brain structures of the same type, but differ in age, resulting in some differences in the shape of the brain structures.

[0051] The template image database can include two-dimensional brain template images corresponding to different ages, and any one of these brain templates can be selected as the image corresponding to the target reference object. Alternatively, based on the shape of the target object, a brain template image with a similar shape can be selected from the template image database and used as the image corresponding to the target reference object.

[0052] Furthermore, for the registration of multiple images, specifically, spatial transformation is performed on the multiple images so that the target object in the transformed and registered image has the same spatial location as the target object in the other images, that is, the target object in the two registered images can achieve spatial consistency. Optionally, traditional gray-scale statistical registration methods and image feature-based registration methods can be used, or deep learning algorithms can be used to register multiple images.

[0053] S103. Based on the first template image corresponding to the target reference object, the target object in multiple registration images is segmented to obtain multiple segmented images. The target reference object and the target object have the same type but different shapes. The first template image includes the segmentation result of the target reference object, and the segmented images include multiple segmented regions.

[0054] To improve the accuracy of segmentation results for registered images, when segmenting multiple registered images, a first template image corresponding to the target reference object can be obtained. Based on the segmentation results of the target reference object in the first template image, the target object in the multiple registered images is segmented to obtain multiple segmented images. Each segmented image includes multiple segmentation regions, which can be areas of primary focus during target object recognition or several important regions of the target object. Furthermore, in practical applications, the number of segmentation regions can be determined based on the characteristics of the target object to be processed, and the registered images can be segmented accordingly. For example, for the brain structure of infants and young children, important brain regions can be divided into 20. Therefore, when segmenting the brain structure of infants and young children in multiple registered images, it can be segmented into 20 regions, meaning the final segmented image includes 20 brain regions.

[0055] Optionally, the first template image can be a three-dimensional image. In practical applications, the target object may be a three-dimensional object. Therefore, when segmenting a registered image containing the target object, obtaining a three-dimensional template image of the target reference object, and then segmenting the registered image based on the segmentation result of the target reference object in the three-dimensional target image, can yield more accurate segmentation results. For example, if the target object is the brain structure of a 6-month-old infant, then the target reference object can be the brain structure of a 24-month-old infant, and its first template image can be a three-dimensional brain segmentation map of a 24-month-old infant. Based on the three-dimensional brain segmentation map, segmentation processing is performed on the registered image containing the target object to obtain the brain segmentation map corresponding to the target object.

[0056] Alternatively, for acquiring the first template image, the image processing device can randomly select an image from an established template image database as the template image. Alternatively, it can select an image from the template image database that has a shape similar to the target object's shape and use that image as the first template image. Continuing with the above example, the template image database may include brain segmentation images corresponding to different ages, and any one of these brain segmentation images can be selected as the first template image. Since infant brain images of different months have different shapes, to ensure the accuracy of image segmentation, an infant brain segmentation image from the template image database that is close to the target object's age can also be selected as the first template image.

[0057] S104. Extract features from each segmented region in multiple segmented images to obtain multiple image features corresponding to each segmented region in multiple segmented images.

[0058] To improve the accuracy of image feature extraction and enable precise target object identification based on the extracted image features, feature extraction is performed by treating each segmented region in each segmented image as a unit. Multiple image features are extracted from each segmented region separately. This allows for the extraction of features from several important regions of the target object, and also enables targeted analysis and processing of the extracted image features from each region, thereby improving the quality of image processing. The image features can include intensity features and texture features, and the number of extracted image features can be set according to actual needs and is not limited.

[0059] For example, if the target object is the brain structure of a 6-month-old infant, multiple images containing the target object can be MRI images of the brain structure of a 6-month-old infant. Each MRI image includes multiple layers of two-dimensional images; that is, the multiple images containing the target object can be multiple layers of two-dimensional images. The two-dimensional images are segmented to obtain multiple brain regions, and image features are extracted in each brain region. 47 image features can be extracted from each brain region.

[0060] There are several ways to extract features from segmented images. For example, convolutional neural networks can be used to extract image features from segmented images, or multiple image features corresponding to each segmented region in a segmented image can be extracted by calculating the intensity and texture features of multiple segmented images.

[0061] S105. Perform fusion processing on multiple image features corresponding to the same segmentation region in multiple segmented images to obtain multiple fused image features corresponding to each segmentation region.

[0062] To ensure the quality and effectiveness of image processing and make the recognition results of the target object more accurate, after obtaining multiple image features corresponding to each segmentation region in multiple segmented images, feature fusion processing is performed on multiple image features corresponding to the same segmentation region in multiple segmented images to obtain multiple fused image features corresponding to each segmentation region.

[0063] Specifically, for example, given 18 segmented images, each image contains 5 segmented regions: segmentation region 1, segmentation region 2, segmentation region 3, segmentation region 4, and segmentation region 5. Each segmented region has 3 image features extracted: image feature A, image feature B, and image feature C. During feature fusion, firstly, image features A from segmentation region 1 are fused, resulting in the fused image feature A corresponding to segmentation region 1. Next, image features B from segmentation region 1 are fused, resulting in the fused image feature B corresponding to segmentation region 1. Then, image features C from segmentation region 1 are fused, resulting in the fused image feature C corresponding to segmentation region 1. This process is repeated for segmentation region 2, segmentation region 3, segmentation region 4, and segmentation region 5.

[0064] The fused image features integrate all the features of each segmented image. By averaging the features of each image, relevant information can be supplemented and noise and redundancy can be removed. This allows the fused image features to enhance the relevant features of the image and better express the relevant information of the target object.

[0065] S106. Based on the multiple fused image features, determine the recognition result corresponding to the target object.

[0066] Finally, the image processing device can further determine the recognition result corresponding to the target object based on multiple fused image features. This includes recognizing specific information about the target object, such as facial information, product details, and brain age corresponding to brain structures. Continuing with the example above, after feature fusion, three fused image features are obtained for each of the five segmented regions. Then, based on the obtained 15 fused image features, the brain age corresponding to the target object is determined.

[0067] Alternatively, multiple fused image features can be input into a pre-trained image recognition model to obtain the recognition result corresponding to the target object. The image recognition model is trained to determine the recognition result of the target object based on multiple image features. The image recognition model can be generated by training a neural network; that is, by using the pre-defined recognition result corresponding to the target object and multiple image features corresponding to the target object to train the neural network, thereby obtaining the image recognition model. After establishing the image recognition model, it can be used to analyze and process the multiple image features of the target object to obtain the recognition result corresponding to the target object.

[0068] In this method, the trained image recognition model analyzes and processes multiple fused image features of the target object to obtain the recognition result corresponding to the target object. This not only effectively ensures the accuracy and reliability of the target object recognition result, but also further improves the stability and reliability of the method.

[0069] In this embodiment of the invention, features are extracted from each segmented region in multiple segmented images containing the target object, and multiple image features of the same segmented region in the multiple segmented images are fused to obtain multiple fused image features. Based on the multiple fused image features, the recognition result corresponding to the target object is determined. That is, by processing multiple images containing the target object simultaneously, the influence of the image quality of a single image on feature extraction can be effectively avoided. This not only ensures the quality and effect of image processing but also reduces the difficulty of image processing, making the image processing method widely applicable to various application scenarios. Furthermore, feature extraction from each segmented region allows for targeted feature extraction, resulting in more accurate extracted image features. Moreover, fusing the image feature information from each image and determining the recognition result of the target object based on the fused image feature information makes the target object recognition result more accurate.

[0070] Optionally, in practice, there are often situations where it is necessary to identify target objects corresponding to different types of users. User types can include gender, identity, age, etc. Specifically, in medical scenarios, user types can include middle-aged men, male infants, young women, female infants, etc., and target objects can be different parts of the body, such as the brain, lungs, liver, etc. Since the same target object varies significantly between different user types, multiple images of the same type of target object can be used to first obtain a target reference object for that user type. When the latest image of the same user type is obtained, this target reference object can be used as a standard for registration and segmentation of the images corresponding to the target object of that user type, improving the quality and effect of image processing, thereby further improving the accuracy of target object identification results.

[0071] For example, the user type could be an infant, and the target object for this user type could be brain structure. Because infants are in a stage of brain development, the shape of their brain structures varies significantly across different months. Furthermore, even within the same month, there are individual differences in the brain structure of each individual infant. Therefore, based on multiple historical images of infant brain structures from different months, an average brain structure—the target reference object—can be generated for each month, and this target reference object can be used as the standard. When a new image of an infant brain structure from a specific month is acquired, the target reference object for that month can be used as the standard to register and segment the brain structures in this latest image.

[0072] The determination of the target reference object mentioned above as a standard, and the registration and segmentation of multiple images containing the target object based on the image corresponding to the target reference object, can be carried out in accordance with the method provided in the following embodiments.

[0073] Figure 2 A flowchart for registering multiple images to obtain multiple registered images, provided in an embodiment of the present invention, is as follows: Figure 2 As shown, the method may include the following steps:

[0074] S201. Obtain the reference images corresponding to multiple reference objects, where the reference objects and the target object have the same type but different shapes.

[0075] To improve the accuracy of subsequent segmentation results for each target object in the segmented image, registration processing can be performed on multiple images based on multiple reference objects similar to the target object. Specifically, multiple reference objects are determined according to the type of the target object, and reference images corresponding to each of these reference objects are obtained. For example, if the target object is the brain structure of a 6-month-old infant, multiple images of 6-month-old infant brain structures can be obtained from the database, and these 6-month-old infant brain structures can be identified as reference objects. These reference objects are of the same type as the target object but differ in shape. Due to individual differences, even when acquiring brain structure images of infants of the same age, there will still be some differences between the individual infant brain structures.

[0076] S202. Calculate the reference image to obtain the target reference object and the second template image corresponding to the target reference object.

[0077] Since the reference objects and the target object differ in shape, using any one of these reference objects as the standard during registration might result in the same registration result regardless of the reference object used. To unify the spatial position of all target objects, the same target reference object can be used for registration of the same type of target object. Therefore, after obtaining the reference images of each reference object, the reference objects in each image can be calculated to determine the target reference object and its corresponding second template image. Subsequent registration of multiple images is then performed based on the target reference object in the second template image. The second template image can be either a two-dimensional or three-dimensional image, as long as it maintains the same dimension as the obtained image containing the target object.

[0078] In practical applications, since the brain structures of different infants in the same month vary, to eliminate individual differences and unify the obtained brain structures into the same spatial coordinates, the reference images corresponding to multiple reference objects can be processed to obtain a standard target reference object and a second template image corresponding to the target reference object. Then, each image containing the target object to be processed can be registered to the second template image of the target reference object, so that the feature points on the target object in each image to be processed have the same spatial position.

[0079] For example, if the target object is the brain structure of a 6-month-old infant, 25 brain structure images of 6-month-old infants with similar image dimensions and brain structure shape and size are selected from the template database according to the type of the target object. The 25 selected brain structure images are processed to obtain an average image, and the average image is determined as the brain template image of a 6-month-old infant.

[0080] S203. Using a preset algorithm, multiple images are registered based on the position information of the target reference object in the second template image to obtain multiple registered images.

[0081] After obtaining the second template image corresponding to the target object, a preset algorithm is used to spatially transform multiple images based on the positional information of the target reference object in the second template image, so that the target object in the multiple registered images has the same positional information. Optionally, the spatial transformation of multiple images can be achieved through operations such as rotation and translation. The preset algorithm can be the open-source toolkit Advanced Normalization Tools (ANTS).

[0082] After obtaining multiple registration images, a first template image corresponding to the target reference object is acquired, which includes the segmentation result of the target reference object. Then, based on the first template image corresponding to the target reference object, the target object in the multiple registration images is segmented to obtain multiple segmented images. Since the multiple registration images are obtained by registering based on the positional information of the target reference object in the second template image (i.e., the target object and the target reference object have the same positional relationship), the registration images can be directly segmented according to the segmentation result of the target reference object in the first template image. This eliminates the need for feature point extraction and correction, directly completing the segmentation of the target object in the registration images to obtain multiple segmented regions. This not only simplifies the image segmentation process but also improves the accuracy of the segmentation results.

[0083] In this embodiment of the invention, by acquiring reference images corresponding to multiple reference objects, calculating on the reference images, obtaining a target reference object and a second template image corresponding to the target reference object, and using a preset algorithm, registering multiple images based on the position information of the target reference object in the second template image to obtain multiple registered images, this not only provides target reference objects that can be used as registration standards for target objects of the same type, but also improves registration efficiency and accuracy. This allows subsequent image segmentation of the registered images to be directly performed based on the first template image corresponding to the target object, simplifying the segmentation process and improving the accuracy of the segmentation results.

[0084] The above embodiments describe how, after registering and segmenting an image containing a target object, to improve the accuracy of the recognition result corresponding to the target object, feature extraction can be performed on several important segmentation regions corresponding to the target object as units. In an optional embodiment, the specific implementation of obtaining the image corresponding to each segmentation region may include: obtaining a first template image, which includes multiple segmentation regions; determining the binary template image corresponding to each segmentation region; and multiplying the binary template image corresponding to each segmentation region with the registered image to be segmented to obtain an image containing only one segmentation region. The binary template image (mask) is an image containing only 0s and 1s, which can be used to multiply with other images to obtain an image containing only the part with a mask value of 1. Specifically, the mask value corresponding to the segmentation region to be extracted can be set to 1, and other regions can be set to 0, thereby obtaining the binary template image corresponding to that segmentation region.

[0085] Specifically, for example, if a segmented image includes five segmentation regions, when extracting features from these five regions, the first step is to determine the binary template image corresponding to each of the five segmentation regions to be extracted. Next, the binary template image corresponding to each of the five segmentation regions is multiplied by the segmentation image to obtain images containing only segmentation region 1, segmentation region 2, segmentation region 3, segmentation region 4, and segmentation region 5, respectively. Then, image feature extraction is performed on the image containing only segmentation region 1 to obtain multiple image features corresponding to segmentation region 1; image feature extraction is performed on the image containing only segmentation region 2 to obtain multiple image features corresponding to segmentation region 2; and so on, sequentially determining multiple image features corresponding to segmentation region 3, segmentation region 4, and segmentation region 5.

[0086] When extracting features from each segmented region, intensity and texture features corresponding to each segmented region can be extracted separately. Histograms can be used to describe intensity features, and based on these histogram features, commonly used statistics such as maximum, minimum, mean, kurtosis, and skewness can be calculated to further describe the intensity features. Texture features are global features that reflect the visual characteristics of homogeneous phenomena in an image, reflecting the slowly changing or periodically varying surface structure of an object. First-order, second-order, and higher-order statistical methods are typically used to quantify and extract image texture features, which are then qualitatively or quantitatively described using image intensity discretization. Furthermore, methods such as variance thresholding, k-best selection, and least absolute shrinkage and selection operator (LASSO) can be used to reduce the dimensionality of feature values ​​and filter them to obtain more representative features.

[0087] In addition, for the specific extraction process of multiple image features corresponding to each segmented region, please refer to, for example... Figure 3 As shown.

[0088] like Figure 3 This embodiment provides a flowchart for extracting features from each segmented region in multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images. For example... Figure 3 As shown, the specific steps may include the following:

[0089] S301. Calculate the first-order statistical features and texture features corresponding to each segmented region in multiple segmented images.

[0090] The first-order statistical features are feature values ​​calculated directly based on the pixel gray-level distribution of the segmented image containing the target object. Specifically, a gray-level histogram is drawn based on the voxel data of each segmented region, the frequency of each gray level is counted, and then the first-order statistical features corresponding to that segmented region are calculated based on the gray-level histogram. Texture features include features calculated based on gray-level co-occurrence matrices, gray-level run-length matrices, etc.

[0091] The Gray-level Co-occurrence Matrix (GLCM) is a common method for describing textures by studying the spatial correlation characteristics of gray levels. Therefore, GLCM is frequently used to describe texture features. After calculating the GLCM corresponding to each segmented region, features of the GLCM are calculated based on it. For example, GLCM features may include GLCM energy, GLCM entropy, homogeneity 1, homogeneity 2, etc.

[0092] The gray-level run-length matrix (GLRLM) is a matrix composed of the lengths of the run-lengths of gray-level values. After calculating the gray-level run-length matrix corresponding to each segmented region, its features are calculated based on the gray-level run-length matrix. Since the segmented image is a two-dimensional image, both GLCM and GLRLM only calculate the matrix in the four directions of the x and y planes.

[0093] S302. Based on first-order statistical features and texture features, determine multiple radiomics features corresponding to each segmented region in multiple segmented images.

[0094] Finally, based on first-order statistical features and texture features (features obtained by calculating gray-level co-occurrence matrix and gray-level run-length matrix), multiple radiomics features corresponding to each segmented region in multiple segmented images are determined. For example, in an optional embodiment, the segmented images are segmented maps of brain structures, with each brain structure image divided into 20 brain regions. For each brain region, 47 radiomics features are obtained by calculating first-order statistical features and texture features. These 47 radiomics features include 14 intensity features and 33 texture features (22 gray-level co-occurrence matrix features and 11 gray-level run-length matrix features).

[0095] In this embodiment, the first-order statistical features and texture features corresponding to each segmented region in multiple segmented images are first calculated. Then, based on the first-order statistical features and texture features, multiple image omics features corresponding to each segmented region in multiple segmented images are determined. This allows for more targeted and accurate extraction of multiple image features within each segmented region corresponding to the target object.

[0096] As mentioned in the above embodiments, image feature extraction processing is performed on each segmented region in the segmented image. In practical applications, due to factors such as scanning angle, the obtained image of the target object may be blurry. Therefore, the image feature analysis of a single image may result in a certain deviation in the target object recognition result. To improve the accuracy of the target object recognition result, after obtaining multiple image features corresponding to each segmented region in each segmented image, the multiple image features corresponding to each segmented region are fused to obtain the fused image features. The image features obtained in this way integrate all the features of each segmented image, which can complement each other with relevant information, remove noise and redundancy, and enable the fused image features to enhance the relevant features of the image and better express the relevant information of the target object.

[0097] Therefore, in Figure 3 Based on the illustrated embodiment, to further improve the accuracy of target object recognition, after obtaining multiple radiomics features corresponding to the same segmented region in multiple segmented images, the mean of these multiple radiomics features corresponding to the same segmented region in the multiple segmented images is calculated to obtain multiple fused radiomics features corresponding to each segmented region. Finally, based on the multiple fused radiomics features, the feature matrix corresponding to the target object is determined; and based on the feature matrix, the recognition result corresponding to the target object is determined.

[0098] The method for determining the feature matrix corresponding to the target object can include stacking multiple fused image features corresponding to each segmented region to obtain the feature matrix. For example, if the target object corresponds to 20 segmented regions, and each segmented region corresponds to 47 fused image features, then the 47 fused image features from the 20 segmented regions are stacked to obtain a 20*47 feature matrix.

[0099] The image processing methods described in the above embodiments can be applied to various application scenarios. For example, in medical settings, when examining human brain structures, a 2D MRI scanner is typically used to scan the brain structure of the subject to obtain 2D MRI image data. The 2D MRI images include T1W and T2W imaging, with each T1W image containing multiple layers of brain structure images, and each T2W image also containing multiple layers of brain structure images. T1W imaging highlights differences in tissue T1 relaxation (longitudinal relaxation), while T2W imaging highlights differences in tissue T2 relaxation (lateral relaxation). Because the brain structure of infants and young children develops and changes rapidly, in the early stages of myelination, the increased content of cholesterol and galactocerebroside in the cell membranes of oligodendrocytes during myelination leads to a greater increase in signal intensity in white matter on T1WI. Therefore, T1WI is more valuable in evaluating myelination in infants under 1 year old (within 6 months). In the later stages of myelination, the decrease in free water content in mature white matter leads to a greater reduction in white matter signal on T2WI, making T2WI more beneficial for evaluating the later stages of myelination. Since the contrast between gray and white matter in T1W and T2W alone is not high, and different age groups focus on different modalities, considering operational uniformity and reducing operational complexity, and to better reflect the differences in gray and white matter development in brain tissue at various stages, T1W and T2W image data can be fused. The combined image is then processed to obtain the recognition results corresponding to the brain structures.

[0100] Before acquiring multiple images containing the target object, this method also includes an image preprocessing process. For details, please refer to [reference needed]. Figure 4 As shown. Figure 4 This is a flowchart illustrating an image processing method provided in this embodiment. To improve image processing quality and effect, the method further includes:

[0101] S401. Acquire magnetic resonance images containing brain structures, including T1W imaging and T2W imaging, wherein T1W imaging includes multiple first images and T2W imaging includes multiple second images.

[0102] In practical applications, after performing an MRI scan on the brain structure to be examined, magnetic resonance images containing the brain structure can be obtained. These images include T1W and T2W imaging. The grayscale of a T1W image is primarily determined by the longitudinal relaxation rate of the tissue, while the grayscale of a T2W image is primarily determined by the lateral relaxation rate of the tissue. Most existing MRI scanners are 2D MRI scanners. 2D MRI scans are performed layer by layer. First, a specific layer is selectively excited using radiofrequency pulses, and then gradient coding is used to spatially locate that layer to achieve imaging. Therefore, when performing an MRI scan on a brain structure, the brain structure can be divided into multiple layers for scanning. The obtained T1W images include multiple first images, and the T2W images include multiple second images.

[0103] Before processing the magnetic resonance image corresponding to the brain structure, the magnetic resonance image of the brain structure input by the user can be received directly, or the magnetic resonance image of the brain structure can be obtained directly from the database. The magnetic resonance image includes T1W imaging and T2W imaging. T1W imaging includes multiple first images, and T2W imaging includes multiple second images.

[0104] S402. Perform scalp removal processing on multiple first images and multiple second images to obtain multiple processed first images and multiple processed second images.

[0105] The brain structures in the first and second images include not only important tissues such as gray and white matter but also the scalp. The scalp may affect the subsequent identification of brain structures. Therefore, after acquiring multiple first images and multiple second images, dpabi is used to remove the scalp from both images, resulting in processed first and second images. dpabi is a brain imaging data processing and analysis toolkit.

[0106] The specific implementation of the peeling operation may include: firstly, calculating the grayscale histogram of the first image / second image, and determining the grayscale threshold, maximum and minimum grayscale values ​​of the image to distinguish between brain tissue and non-brain tissue through the grayscale histogram; then, roughly estimating the centroid of the brain tissue, and obtaining the initial brain tissue based on the grayscale values ​​of the brain and non-brain tissue; finally, constructing the initial brain surface within the brain tissue using three-dimensional triangular facets, establishing tangential force and smoothing force on each triangular facet, and maintaining a certain distance and smoothness on the initial facet under the drive of the two forces on the triangular facets until the brain surface is sufficiently smooth and stable, at which point the segmentation ends, and the processed first image / second image is obtained.

[0107] In one optional embodiment, dpabi is used to perform scalp removal on the first image to obtain a scalp-removed first image, and a binary template image mask of the brain is obtained by setting the non-zero parts of the image to 1. Then, the binary template image mask is used to perform scalp removal processing on other images.

[0108] S403. Process the processed multiple first images and multiple processed second images to obtain multiple first gray matter images corresponding to the processed multiple first images and multiple second gray matter images corresponding to the processed multiple second images.

[0109] First, the brain structure is calibrated by correcting the positions of the first and second images, specifically the gray matter and white matter tissues, according to the second image. Then, tissue segmentation is performed on the corrected first and second images to obtain the corresponding gray matter probability maps for each image. The gray matter probability maps are binarized to obtain gray matter mask images. These gray matter mask images are then multiplied by the corrected first and second images to obtain the first gray matter image corresponding to the first image and the second gray matter image corresponding to the second image, respectively.

[0110] S404. Based on multiple first gray matter images, multiple second gray matter images, multiple first images, and multiple second images, generate multiple combined images containing brain structures.

[0111] Specifically, the gray values ​​are extracted from the first and second grayscale images, sorted from smallest to largest, and the median gray values ​​MG_T1 and MG_T2 are obtained respectively. Then, the new image sT2w after grayscale scaling is calculated using the formula sT2w = MG_T1 / MG_T2 * T2W. Here, MG_T1 represents the median gray value corresponding to the first grayscale image, MG_T2 represents the median gray value corresponding to the second grayscale image, and T2W represents the second image. Finally, the sT2w image and the first image are processed to obtain a combined image. This can be calculated using the formula sT1W / sT2W = (T1W - (M G_T1 / M G_T2 )*T2W) / (T1W+(M G_T1 / M G_T2 The sT1W / sT2W method generates a composite image containing brain structures. Here, sT1W / sT2W represents the composite image, T1W represents the first image, and T2W represents the second image.

[0112] In this embodiment of the method, T1W and T2W images are fused and then the combined image is processed to obtain the recognition result corresponding to the brain structure. Instead of switching images back and forth during image processing based on the type of the target object, this not only reduces the complexity of operation, but also allows the combined image to better reflect the differences in gray and white matter development in brain tissue at different stages, so that the combined image can better reflect the characteristics of various tissue structures of the target object.

[0113] Based on the above embodiments, for ease of understanding, combined with Figure 5 Taking the processing of infant brain structure images in a medical setting to determine the corresponding brain age as an example, the specific implementation process of the image processing method provided above is illustrated. The image containing the target object can be a brain structure image of a 6-month-old infant. The first template image corresponding to the target reference object can be a segmented brain image of a 24-month-old infant. The second template image corresponding to the target reference object can also be a template brain image of a 24-month-old infant. The image containing the target object is processed using the image corresponding to the target reference object to obtain the corresponding brain age of the target object. The specific image processing process may include:

[0114] Step 1: Obtain multiple combined images of the brain structure to be identified.

[0115] Specifically, before acquiring multiple combined images of the brain structure to be identified, the method also includes a data preprocessing step. T1W and T2W images of the brain structure to be identified are acquired, each comprising stacked multi-layered brain structure images. Using the T2W image as a reference, the T1W and T2W images are aligned using ANTsPy (a Python library for biomedical image processing) to obtain a corrected rT1W image. Clinically, T1W and T2W images are typically obtained from a single scan, and the registration process usually involves non-linear transformations such as rotation to eliminate potential head movements during scanning, ensuring that the corresponding positions of important tissues in the brain structure are aligned in both images. Therefore, either the T2W or T1W image can be used as a reference.

[0116] Next, spm12 was used to perform tissue segmentation on the T2W and rT1W images to obtain gray matter probability maps from the segmented T2W and rT1W images. Inhomogeneity correction was then applied to the T2W and rT1W images to obtain the corrected mT2W and mrT1W images. spm12 is a toolbox used for preprocessing NMR data.

[0117] The grayscale probability map obtained from the T2W image segmentation is binarized, with values ​​greater than 0.5 set to 1 and all others set to 0, resulting in a grayscale mask image. This grayscale mask is then multiplied by the mT2W and mrT1W images respectively to obtain the GM-mT2W and GM-mrT1W images. The GM-mT2W and GM-mrT1W images are new images containing only the grayscale components. The values ​​from the GM-mT2W and GM-mrT1W images are then extracted and sorted in ascending order to obtain the median MG_T2 and MG_T1 of the GM-mT2W and GM-mrT1W images, respectively. Finally, the scaled-down image sT2w is calculated using the formula sT2w = MG_T1 / MG_T2 * T2W (where T2W is the non-uniformity-corrected mT2W image).

[0118] Finally, based on the mrT1W and sT2W images, the composite image CI (i.e., sT1W / T2W image) is calculated. Specifically, the composite image CI can be calculated using the formula CI = (mrT1w - sT2w) / (mrT1w + sT2w). The composite image CI includes multiple layers of composite images, meaning it is defined as multiple composite images containing the target object.

[0119] Optionally, the method may further include a scalp removal operation, specifically, applying a binary template image mask of the brain to mrT1W, mT2W, and sT1W / T2W images, masking these images, and obtaining scalp-removed bmrT1W, bmT2W, and bsT1W / T2W images.

[0120] Step 2: Obtain the brain template image corresponding to the target reference object, and use a preset algorithm to register multiple combined images based on the brain template image to obtain multiple registered combined images.

[0121] Because infants' brains develop rapidly and due to individual differences, the shape of each infant's brain structure varies to some extent. Therefore, when processing composite images containing the target object, a 2D brain template image of a 24-month-old infant can be used as a reference for registration processing of multiple images. The 2D brain template image of a 24-month-old infant is an average image obtained by calculating from multiple brain structure images of healthy 24-month-old infants.

[0122] Specifically, using a self-made 2D brain template of a 24-month-old infant as a reference, ANTsPy was used to register the bmrT1W image onto the brain template, obtaining the wbmrT1W image and the transformation matrix. Based on the transformation matrix, the bmT2W image and the bsT1W / T2W image were registered separately to obtain the registered wbmT2W image and wbsT1W / T2W image.

[0123] Step 3: Obtain the brain segmentation map corresponding to the target reference object. Based on the brain segmentation map, segment the brain structure in the registered multiple combined images to obtain multiple segmented images. The segmented images include multiple target brain regions corresponding to the brain structure.

[0124] Specifically, brain segmentation images of 24-month-old infants are acquired. Based on these images, the wbsT1W / T2W images are segmented to obtain 20 brain regions. These include, for example, the left / right corpus callosum, the left / right anterior limb of the internal capsule, the left / right posterior limb of the internal capsule, the left / right basal ganglia, the left / right brainstem, the left / right frontal lobe, the left / right parietal lobe, the left / right temporal lobe, the left / right occipital lobe, and the left / right cerebellar hemispheres. Since the wbsT1W / T2W images contain 18 layers of brain structure images, segmenting these images is equivalent to simultaneously segmenting multiple layers of brain structure images, with each layer being divided into multiple brain regions.

[0125] Step 4: Extract features from each target brain region in multiple segmented images to obtain multiple image features corresponding to each target brain region in multiple segmented images.

[0126] Specifically, taking each layer of brain structure image in 2DwbsT1W / T2W as a unit, texture features, including gray-level co-occurrence matrix and gray-level run-length matrix, were calculated for multiple brain regions in each layer. Then, taking a single brain region as a whole, the first-order statistical features of intensity characteristics corresponding to multiple brain regions in multi-layer brain structure images were calculated. Finally, based on the first-order statistical features and texture features including gray-level co-occurrence matrix and gray-level run-length matrix, 47 radiomics image features corresponding to each of the multiple brain regions were determined.

[0127] Step 5: Perform fusion processing on multiple image features corresponding to the same target brain region in multiple segmented images to obtain multiple fused image features corresponding to each target brain region.

[0128] Specifically, after feature extraction, the average value of the image feature value in each brain region in the multi-layer brain structure image of 2DwbsT1W / T2W is calculated, that is, the average value of each of the 33 texture image features is obtained, and the average value of each image feature is determined as the fused image feature.

[0129] Step 6: Determine the recognition result corresponding to the target object based on the multiple fused image features.

[0130] Specifically, 47 fused image features from 20 brain regions are stitched together to obtain a 1*940 feature matrix. This feature matrix is ​​then input into a brain age recognition model to determine the brain age of the target object. The brain age recognition model is trained based on multiple image features corresponding to brain structure images of healthy infants at various stages, as well as brain age samples corresponding to those brain structures.

[0131] Furthermore, in practical applications, after determining the brain age of the target individual, it's possible to determine whether the infant's brain structure development is normal based on that brain age. If it's determined that the infant's brain structure development is abnormal, developmental trajectory maps corresponding to various image features from different months of the infant can be obtained. Based on these developmental trajectory maps, the specific brain region within the brain structure where the development is problematic can be identified. Specifically, the developmental trajectory maps corresponding to each image feature are determined based on multiple image features from infants of different months. The feature matrices of each infant are expanded and stacked with the feature matrices of other infants to obtain a target feature matrix. Based on this target feature matrix, the developmental trajectory maps corresponding to each image feature are determined.

[0132] The image processing apparatus of one or more embodiments of the present invention will be described in detail below. Those skilled in the art will understand that these image processing apparatuses can all be configured using commercially available hardware components through the steps taught in this invention.

[0133] Figure 6 This is a schematic diagram of the structure of an image processing device provided in an embodiment of the present invention, as shown below. Figure 6 As shown, the device includes: an acquisition module 11, a registration module 12, a segmentation module 13, a feature extraction module 14, a fusion module 15, and a determination module 16.

[0134] The acquisition module 11 is used to acquire multiple images containing the target object.

[0135] The registration module 12 is used to perform registration processing on the multiple images to obtain multiple registered images.

[0136] The segmentation module 13 is used to segment the target object in the multiple registration images according to the first template image corresponding to the target reference object, so as to obtain multiple segmented images. The target reference object and the target object are of the same type but different in shape. The first template image includes the segmentation result of the target reference object, and the segmented image includes multiple segmented regions.

[0137] The feature extraction module 14 is used to extract features from each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images.

[0138] The fusion module 15 is used to perform fusion processing on multiple image features corresponding to the same segmentation region in the multiple segmented images to obtain multiple fused image features corresponding to each segmentation region.

[0139] The determining module 16 is used to determine the recognition result corresponding to the target object based on the multiple fused image features.

[0140] Optionally, the registration module 12 can be specifically used to: acquire reference images corresponding to multiple reference objects, wherein the reference objects are of the same type as the target object but have different shapes; perform calculations on the reference images to obtain a target reference object and a second template image corresponding to the target reference object; and use a preset algorithm to register the multiple images according to the position information of the target reference object in the second template image to obtain multiple registered images.

[0141] Optionally, the feature extraction module 14 can be used to: calculate the first-order statistical features and texture features corresponding to each segmented region in the multiple segmented images respectively; and determine multiple radiomics features corresponding to each segmented region in the multiple segmented images based on the first-order statistical features and texture features.

[0142] Optionally, the fusion module 15 can be specifically used to: calculate the mean of multiple radiomics features corresponding to the same segmentation region in the multiple segmented images to obtain multiple fused radiomics features corresponding to each segmentation region.

[0143] Optionally, the determining module 16 may be specifically used to: determine the feature matrix corresponding to the target object based on the plurality of fused image omics features; and determine the recognition result corresponding to the target object based on the feature matrix.

[0144] Optionally, the target object is a brain structure, and the device may further include a preprocessing module, specifically used for: acquiring magnetic resonance images containing brain structures, the magnetic resonance images including T1W imaging and T2W imaging, the T1W imaging including multiple first images, and the T2W imaging including multiple second images; performing scalp removal processing on the multiple first images and the multiple second images to obtain processed multiple first images and processed multiple second images; processing the processed multiple first images and the processed multiple second images to obtain multiple first gray matter images corresponding to the processed multiple first images and multiple second gray matter images corresponding to the processed multiple second images; and generating multiple combined images containing brain structures based on the multiple first gray matter images, the multiple second gray matter images, the multiple first images, and the multiple second images. The registration module 12 can also be used to: acquire a brain template image corresponding to the target reference object; and register the multiple combined images according to the brain template image using a preset algorithm to obtain multiple registered combined images. The segmentation module 13 can also be used to: acquire a brain segmentation map corresponding to the target reference object; and segment the brain structures in the registered multiple combined images according to the brain segmentation map to obtain multiple segmented images, wherein the segmented images include multiple target brain regions corresponding to the brain structures.

[0145] Figure 6 The device shown can perform Figures 1 to 5 For the methods shown in the embodiments, the parts not described in detail in this embodiment can be referred to the following: Figures 1 to 5 The relevant descriptions of the illustrated embodiments are provided below. For the execution process and technical effects of this technical solution, please refer to [link / reference]. Figures 1 to 5 The descriptions in the illustrated embodiments will not be repeated here.

[0146] The above describes the internal functions and structure of the image processing device. In one possible design, the image segmentation device can be implemented as an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 21 and a memory 22. The memory 22 is used to store data supporting the electronic device in performing the above-described actions. Figures 1 to 7 The image processing method program provided in the illustrated embodiment is configured by the processor 21 to execute the program stored in the memory 22.

[0147] The program includes one or more computer instructions, wherein when the one or more computer instructions are executed by the processor 21, they can perform the following steps:

[0148] Retrieve multiple images containing the target object;

[0149] The multiple images are registered to obtain multiple registered images;

[0150] Based on the first template image corresponding to the target reference object, the target object in the multiple registration images is segmented to obtain multiple segmented images. The target reference object and the target object are of the same type but different in shape. The first template image includes the segmentation result of the target reference object, and the segmented image includes multiple segmented regions.

[0151] Feature extraction is performed on each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images;

[0152] Multiple image features corresponding to the same segmented region in the multiple segmented images are fused to obtain multiple fused image features corresponding to each segmented region;

[0153] Based on the multiple fused image features, the recognition result corresponding to the target object is determined.

[0154] Optionally, the processor 21 is further configured to perform the aforementioned Figures 1 to 5 All or part of the steps in the illustrated embodiments.

[0155] The structure of the electronic device may also include a communication interface 23 for the electronic device to communicate with other devices or communication networks.

[0156] In addition, embodiments of the present invention provide a computer storage medium for storing computer software instructions used by the aforementioned electronic device, which includes instructions for executing the above-mentioned... Figures 1 to 5 The procedure involved in the image processing method in the illustrated embodiment.

[0157] This invention also provides a computer program product, which includes computer program instructions that are read and executed by a processor to perform the above-described... Figures 1 to 5 The image processing method shown in the embodiment is illustrated.

[0158] 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 image data used for processing, stored image data, etc.) involved in this invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An image processing method, characterized by, include: Retrieve multiple images containing the target object; The multiple images are registered to obtain multiple registered images; Based on the first template image corresponding to the target reference object, the target object in the multiple registration images is segmented to obtain multiple segmented images. The target reference object and the target object are of the same type but different in shape. The first template image includes the segmentation result of the target reference object, and the segmented image includes multiple segmented regions. Feature extraction is performed on each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images; Multiple image features corresponding to the same segmented region in the multiple segmented images are fused to obtain multiple fused image features corresponding to each segmented region; Based on the multiple fused image features, the recognition result corresponding to the target object is determined; The multiple images are multi-layered brain structure images; the step of extracting features from each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images includes: Taking each layer of brain structure image as a unit, calculate the texture features corresponding to multiple brain regions in each layer of brain structure image, including gray-level co-occurrence matrix and gray-level run length matrix. Taking a single brain region as a whole, calculate the first-order statistical features of intensity characteristics corresponding to multiple brain regions in the multi-layer brain structure image; Based on the first-order statistical features of the intensity features and the texture features including the gray-level co-occurrence matrix and the gray-level run-length matrix, 47 radiomics image features corresponding to multiple brain regions were determined.

2. The method of claim 1, wherein, The registration process for the multiple images to obtain multiple registered images includes: Acquire reference images corresponding to multiple reference objects, wherein the reference objects are of the same type as the target object but have different shapes; The reference image is calculated to obtain a target reference object and a second template image corresponding to the target reference object; Using a preset algorithm, the multiple images are registered based on the position information of the target reference object in the second template image to obtain multiple registered images.

3. The method of claim 1, wherein, The step of extracting features from each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images includes: Calculate the first-order statistical features and texture features corresponding to each segmented region in multiple segmented images; Based on the first-order statistical features and the texture features, multiple radiomics features corresponding to each segmented region in the multiple segmented images are determined.

4. The method of claim 3, wherein, The step of fusing multiple image features corresponding to the same segmented region in the multiple segmented images to obtain multiple fused image features corresponding to each segmented region includes: The mean values ​​of multiple radiomics features corresponding to the same segmented region in the multiple segmented images are calculated to obtain multiple fused radiomics features corresponding to each segmented region.

5. The method of claim 4, wherein, The step of determining the recognition result corresponding to the target object based on the multiple fused image features includes: Based on the multiple fused image omics features, the feature matrix corresponding to the target object is determined; Based on the feature matrix, the recognition result corresponding to the target object is determined.

6. The method according to claim 1, characterized in that, The target object is a brain structure, and before acquiring multiple images containing the target object, the method further includes: Acquire magnetic resonance images containing brain structures, the magnetic resonance images including T1W imaging and T2W imaging, the T1W imaging including multiple first images and the T2W imaging including multiple second images; The plurality of first images and the plurality of second images are subjected to scalp removal processing to obtain a plurality of processed first images and a plurality of processed second images; The processed plurality of first images and the processed plurality of second images are processed to obtain a plurality of first gray matter images corresponding to the processed plurality of first images and a plurality of second gray matter images corresponding to the processed plurality of second images; Based on the plurality of first gray matter images, the plurality of second gray matter images, the plurality of first images, and the plurality of second images, a plurality of combined images containing brain structures are generated.

7. The method of claim 6, wherein, The registration process for the multiple images to obtain multiple registered images includes: Obtain the brain template image corresponding to the target reference object; Using a preset algorithm, the multiple combined images are registered based on the brain template image to obtain multiple registered combined images; The step of segmenting the target object in the multiple registration images based on the first template image corresponding to the target reference object to obtain multiple segmented images includes: Obtain the brain segmentation map corresponding to the target reference object; Based on the brain segmentation map, the brain structures in the registered combined images are segmented to obtain multiple segmented images, each segmented image including multiple target brain regions corresponding to the brain structures.

8. An image processing apparatus characterized by comprising: include: The acquisition module is used to acquire multiple images containing the target object; A registration module is used to perform registration processing on the multiple images to obtain multiple registered images; The segmentation module is used to segment the target object in the multiple registration images according to the first template image corresponding to the target reference object, so as to obtain multiple segmented images. The target reference object and the target object are of the same type but different in shape. The first template image includes the segmentation result of the target reference object, and the segmented image includes multiple segmented regions. The feature extraction module is used to extract features from each segmented region in the multiple segmented images to obtain multiple image features corresponding to each segmented region in the multiple segmented images. The fusion module is used to fuse multiple image features corresponding to the same segmentation region in the multiple segmented images to obtain multiple fused image features corresponding to each segmentation region. The determining module is used to determine the recognition result corresponding to the target object based on the multiple fused image features; The multiple images are multi-layer brain structure images. The feature extraction module is specifically used to calculate the texture features, including the gray-level co-occurrence matrix and the gray-level run-length matrix, for each layer of brain structure image. Taking a single brain region as a whole, the module calculates the first-order statistical features of the intensity features corresponding to multiple brain regions in the multi-layer brain structure images. Based on the first-order statistical features of the intensity features and the texture features including the gray-level co-occurrence matrix and the gray-level run-length matrix, 47 radiomics image features corresponding to multiple brain regions are determined.

9. An electronic device, characterized in that, include: A memory and a processor; wherein the memory stores executable code, and when the executable code is executed by the processor, the processor performs the image processing method as described in any one of claims 1 to 7.

10. A non-transitory machine-readable storage medium, characterized in that, The non-transitory machine-readable storage medium stores executable code that, when executed by a processor of an electronic device, causes the processor to perform the image processing method as described in any one of claims 1 to 7.

Images