A medical image compression method, related device and storage medium

Abstract

Embodiments of the present application relate to the field of medical image processing, and provide a medical image compression method, related device and storage medium, the method comprising: segmenting a lesion region and a non-lesion region of an image; performing hash matching on the lesion region of the image and the lesion region of a reference image; if the lesion region of the image matches the hash value of the reference image, sending different pixels of the image; if not, dividing the lesion region of the image into M sub-lesion regions, and performing hash matching on the M sub-lesion regions and the reference image respectively; if the sub-region of the image matches the hash value of the reference image, sending different pixels of the sub-region; if the sub-region of the image does not match the hash value of the reference image, compressing the sub-region of the image in a first compression manner; and sending the compressed sub-region. The embodiments of the present application can send medical images in multiple ways, reducing the resources consumed in transmitting medical images.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of medical image processing, and more specifically to a medical image compression method, related devices, and storage media. Background Art

[0002] Medical images typically include lesion areas and non-lesion areas. Lesion areas refer to abnormal image regions that differ in morphology or density from surrounding normal tissue. Non-lesion areas refer to image regions outside the lesion area in a medical image. After acquisition, medical images need to be transmitted. How to transmit more clinically valuable medical images using fewer resources is an urgent problem to be solved. Summary of the Invention

[0003] The embodiments of the present application provide a medical image compression method, related devices, and storage medium, which can enable the compressed data to occupy fewer resources and retain the high resolution of the medical image.

[0004] In a first aspect, an embodiment of the present application provides a medical image compression method, the method comprising: acquiring a medical image sequence; segmenting the lesion area and non-lesion area of ​​the image in the medical image sequence; performing hash matching on the lesion area of ​​the current image and the lesion area of ​​the reference image; if the hash value of the lesion area of ​​the current image matches the lesion area of ​​the reference image, sending a first pixel; wherein the first pixel is a different pixel between the lesion area of ​​the current image and the lesion area of ​​the reference image; the current image is an image in the medical image sequence; if the hash value of the lesion area of ​​the current image does not match the lesion area of ​​the reference image, dividing the lesion area of ​​the current image There are M sub-lesion areas, and the M sub-lesion areas are hash-matched with the sub-lesion areas of the reference image respectively; if the hash value of the first sub-area of ​​the current image matches the hash value of the first sub-area of ​​the reference image, the second pixel is sent; wherein the second pixel is a different pixel between the first sub-area of ​​the current image and the first sub-area of ​​the reference image, and the first sub-area is a sub-area of ​​the M sub-lesion areas; M is a positive integer; if the hash value of the second sub-area of ​​the current image does not match the hash value of the second sub-area of ​​the reference image, the second sub-area of ​​the current image is compressed using the first compression method; and the compressed second sub-area is sent; the second sub-area is a sub-area of ​​the M sub-lesion areas.

[0005] In combination with the first aspect, in some embodiments, the method further includes: performing hash matching on the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image; if the hash values ​​of the non-lesion area of ​​the current image do not match those of the non-lesion area of ​​the reference image, dividing the non-lesion area of ​​the current image into N sub-non-lesion areas, and performing hash matching on the N sub-non-lesion areas with the sub-non-lesion areas of the reference image respectively; wherein N is a positive integer; if the hash values ​​of the third sub-area of ​​the current image do not match those of the third sub-area of ​​the reference image, determining whether the number of third pixels is greater than a first threshold; if the number of third pixels is greater than the first threshold, compressing the third sub-area of ​​the current image using a second compression method; and sending the compressed third sub-area of ​​the current image; wherein the third pixel is a pixel in the third sub-area of ​​the current image that is different from the pixel in the third sub-area of ​​the reference image; the third sub-area is a sub-area of ​​the N sub-non-lesion areas; if the number of third pixels is not greater than the first threshold, sending the third pixel.

[0006] In combination with the first aspect, in some embodiments, the method further includes: if the hash value of the non-lesion area of ​​the current image matches the hash value of the non-lesion area of ​​the reference image, sending a fourth pixel; wherein the fourth pixel is a different pixel between the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image; if the hash value of the fourth sub-area of ​​the current image matches the hash value of the fourth sub-area of ​​the reference image, sending a fifth pixel; wherein the fifth pixel is a different pixel between the fourth sub-area of ​​the current image and the fourth sub-area of ​​the reference image, and the fourth sub-area is a sub-area among the N sub-non-lesion areas.

[0007] In combination with the first aspect, in some embodiments, the compression ratio of the first compression method is smaller than the compression ratio of the second compression method.

[0008] In combination with the first aspect, in some embodiments, the lesion area and non-lesion area of ​​an image in a medical image sequence are segmented, including: using a deep learning segmentation model with an improved U-Net architecture to segment the lesion area and non-lesion area of ​​the medical image sequence, where the modality of the images in the medical image sequence is computed tomography (CT), magnetic resonance imaging (MRI), or positron emission tomography (PET); the deep learning segmentation model with the improved U-Net architecture extracts features of the current image according to the modality to which the current image belongs; the deep learning segmentation model with the improved U-Net architecture performs feature fusion on the feature extraction results of images of multiple modalities; and the loss function of the deep learning segmentation model with the improved U-Net architecture is a combination of Diess loss and focal loss.

[0009] In combination with the first aspect, in some embodiments, the deep learning segmentation model of the improved U-Net architecture performs feature extraction on the current image according to the modality to which the current image belongs, including: the deep learning segmentation model of the improved U-Net architecture recognizes that the modality of the current image is CT; the deep learning segmentation model of the improved U-Net architecture calculates the gradient amplitude map of the current image; the deep learning segmentation model of the improved U-Net architecture uses the gradient amplitude map as an attention signal to generate an attention weight map; the deep learning segmentation model of the improved U-Net architecture applies the attention weight map to the original feature map of the current image or the feature map after preliminary convolution to obtain a weighted feature map; in the convolution layer, the deep learning segmentation model of the improved U-Net architecture uses multiple dilation kernels with different expansion rates to process the weighted feature map to obtain a first image, wherein the first image is the image after feature extraction of the current image.

[0010] In combination with the first aspect, in some embodiments, the method performs hash matching on the lesion area of ​​the current image and the lesion area of ​​the reference image, including: using DCT-based perceptual hash coding for the lesion area of ​​the current image; using DCT-based perceptual hash coding for the lesion area of ​​the reference image; hash matching the encoded lesion area of ​​the current image and the lesion area of ​​the reference image; hash matching the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image, including: using color histogram-based fast hash coding for the non-lesion area of ​​the current image and color histogram-based fast hash coding for the non-lesion area of ​​the reference image; hash matching the encoded non-lesion area of ​​the current image and the non-lesion area of ​​the reference image.

[0011] In a second aspect, a medical image compression device is provided, which includes: an input and output module configured to acquire a medical image sequence; a processing module configured to segment a lesion region and a non-lesion region of an image in the medical image sequence; the processing module is further configured to perform hash matching on the lesion region of the current image and the lesion region of the reference image; if the hash values of the lesion region of the current image and the lesion region of the reference image match, the first pixel is sent; the first pixel is a pixel different between the lesion region of the current image and the lesion region of the reference image; the current image is an image in the medical image sequence; the processing module is further configured to, if the hash values of the lesion region of the current image and the lesion region of the reference image do not match, divide the lesion region of the current image into M sub-lesion regions, and perform hash matching on the M sub-lesion regions and the sub-lesion regions of the reference image respectively; the input and output module is further configured to, if the hash values of the first sub-region of the current image and the first sub-region of the reference image match, send the second pixel; the second pixel is a pixel different between the first sub-region of the current image and the first sub-region of the reference image, and the first sub-region is a sub-region in the M sub-lesion regions; M is a positive integer; the processing module is further configured to, if the hash values of the second sub-region of the current image and the second sub-region of the reference image do not match, compress the second sub-region of the current image in a first compression manner; the input and output module is further configured to send the compressed second sub-region; and the second sub-region is a sub-region in the M sub-lesion regions.

[0012] In a third aspect, a computer readable storage medium is provided, which includes instructions that, when executed on a computer, cause the computer to perform the method according to the first aspect or any one of the implementation forms of the first aspect.

[0013] In a fourth aspect, a computer program product is provided, which includes computer instructions that, when executed by a processor, implement the method according to the first aspect or any one of the implementation forms of the first aspect.

[0014] In a fifth aspect, an embodiment of the present application provides a computing device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the method according to the first aspect or any one of the implementation forms of the first aspect when executing the computer program.

[0015] In the first to fifth aspects above, by segmenting the medical image into lesion and non-lesion areas, and based on the matching results of the lesion area of ​​the image with the lesion area of ​​the reference image, the lesion area image is transmitted by sending the pixels that differ from the reference image when similar. When they are dissimilar, the lesion area is partitioned, and the sub-lesion areas of the partitioned image are compared with the sub-lesion areas of the reference image. If the sub-lesion area is similar to the reference image, the pixels that differ from the sub-lesion area of ​​the reference image are sent. If they are dissimilar, the sub-lesion area is compressed and the compressed sub-lesion area is sent. Thus, the medical image can be compressed while ensuring the transmission quality of the medical image, saving transmission resources. Furthermore, the determination of compressed pixels is accelerated, thereby improving image compression efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The objects, features and advantages of the embodiments of the present application will become readily understood by reading the detailed description of the embodiments of the present application with reference to the accompanying drawings, wherein:

[0017] Figure 1 A schematic diagram of an image processing system for a medical image compression method according to an embodiment of the present application;

[0018] Figure 2 A schematic flow chart of a medical image compression method according to an embodiment of the present application;

[0019] Figure 3 A schematic diagram of performing hash matching on a lesion area of ​​a current image and a lesion area of ​​a reference image in a medical image compression method according to an embodiment of the present application;

[0020] Figure 4 This is a schematic structural diagram of a medical image compression device according to an embodiment of the present application;

[0021] Figure 5 A schematic diagram of the structure of a computing device according to an embodiment of the present application.

[0022] In the drawings, the same or corresponding reference numerals denote the same or corresponding parts. DETAILED DESCRIPTION

[0023] In the description and claims of the embodiments of the present application and the accompanying drawings, the terms "first," "second," and the like are used to distinguish similar objects (e.g., the first sub-region and the second sub-region are each represented as different sub-regions, and similarly for other purposes), and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product, or device. The division of modules in the embodiments of the present application is merely a logical division, and other divisions may be used in actual implementations. For example, multiple modules may be combined or integrated into another system, or some features may be omitted or not implemented. In addition, the coupling or direct coupling or communication connection shown or discussed may be through some interface, or indirect coupling between modules. The communication connection may be electrical or other similar forms, which are not limited in the embodiments of the present application. Moreover, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed into multiple circuit modules, and some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiment of the present application.

[0024] The present invention provides a medical image compression method, related device and storage medium, which can be applied to an image processing system capable of transmitting medical images. Figure 1 The image processing system may include a medical image compression device (e.g., a terminal device) and an image processing device (e.g., a server). The medical image compression device is configured to send a compressed medical image (e.g., a first pixel, a second pixel, or a compressed second sub-region as described below) to the image processing device. The image processing device may receive the compressed medical image and decompress it to perform subsequent processing on the medical image.

[0025] The solutions provided in the embodiments of the present application involve technologies such as artificial intelligence (AI), computer vision (CV), and machine learning (ML), and are specifically described through the following embodiments:

[0026] AI refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, artificial intelligence is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. Artificial intelligence is the study of the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0027] AI technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0028] Computer vision (CV) is the science of making machines "see." Specifically, it refers to machine vision, which uses cameras and computers to replace the human eye in identifying, tracking, and measuring objects. This involves further image processing, which transforms the computer's image into an image more suitable for human observation or transmission to instrumentation. As a scientific discipline, computer vision studies related theories and technologies, aiming to build artificial intelligence systems capable of extracting information from images or multidimensional data. Computer vision technologies typically include adversarial perturbation generation, image recognition, image semantic understanding, image retrieval, optical character recognition (OCR), video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, and common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0029] In related art, medical image compression devices use the same compression method for all pixels in a captured medical image. However, different pixels in a medical image represent different medical information, and the importance of each pixel varies. Using the same compression method for all pixels results in a high rate of damage to important information, a low overall compression rate for the medical image, and a waste of transmission resources.

[0030] Based on this, an embodiment of the present application provides a medical image compression method, which segments a medical image into lesion areas and non-lesion areas, and based on the matching results of the lesion area of ​​the image and the lesion area of ​​the reference image, transmits the lesion area image by sending the pixels that are different from the reference image when they are similar. When they are not similar, the lesion area is partitioned, and the sub-lesion area of ​​the partitioned image is compared with the sub-lesion area of ​​the reference image. If the sub-lesion area is similar to the reference image, the pixels that are different from the sub-lesion area of ​​the reference image are sent. If they are not similar, the sub-lesion area is compressed and the compressed sub-lesion area is sent. Thus, the medical image can be compressed while ensuring the transmission quality of the medical image, saving transmission resources. It also speeds up the judgment of compressed pixels and improves the image compression efficiency.

[0031] Furthermore, this embodiment of the present application first compares the lesion area of ​​the current image with the lesion area of ​​the reference image, and then compares the non-lesion area of ​​the current image with the non-lesion area of ​​the reference image. Pixels that differ between the lesion area of ​​the current image and the reference image are sent first. This speeds up the transmission of key information (i.e., the lesion area).

[0032] Reference Figure 2 , Figure 2 This is a flow chart of a medical image compression method provided in an embodiment of the present application. This method can be executed by a medical image compression device and can be applied to scenarios where medical images need to be transmitted. The method includes the following steps:

[0033] Step 101: A medical image compression apparatus acquires a medical image sequence.

[0034] Medical image compression devices can capture multiple medical images to form a medical image sequence. For example, when a patient undergoes a computed tomography (CT) scan, multiple CT images are captured. The image modalities in a medical image sequence can include CT, magnetic resonance imaging (MRI), or positron emission tomography (PET).

[0035] The images in the medical image sequence may be pancreas images, kidney images, intestinal images, lung images, etc.

[0036] The reference image may be an image in a medical image sequence, or a preset image. For example, the preset image may be an image of the same human body region as the image acquired by the medical image compression device. For example, if the image acquired by the medical image compression device is an image of the liver, the preset image may be a standard liver image. The standard liver image may be an image of a liver diseased area or an image of a healthy liver.

[0037] In some embodiments, the standard image is one of the top 10% of images in a historically acquired liver CT image set that are most similar to other liver CT images in the set. The similarity between a historically acquired image set and other images in the set can be determined with reference to the description of related art and will not be repeated here. By selecting the standard image as the reference image, images in the medical image sequence can be made more similar to the standard image, thereby requiring fewer pixels from the images in the medical image sequence to be transmitted.

[0038] Step 102: The medical image compression apparatus segments the lesion area and the non-lesion area of ​​the images in the medical image sequence.

[0039] In some embodiments, the medical image compression apparatus can identify regions of interest (i.e., lesion regions) in an image and segment the images in a medical image sequence into images of the region of interest and images of non-regions of interest. The specific manner in which the medical image compression apparatus segments the images in the medical image sequence into lesion regions and non-lesion regions can be found in the description of subsequent embodiments and is not further described here.

[0040] It is understood that when the reference image is a preset image, after the medical image compression device segments the image into lesion areas and non-lesion areas, the lesion areas and non-lesion areas of the reference image can be segmented based on the positions of the regions of interest in the images in the medical image sequence, thereby obtaining the lesion areas in the reference image corresponding to the images in the medical image sequence.

[0041] Step 103 : The medical image compression apparatus performs hash matching on the lesion area of ​​the current image and the lesion area of ​​the reference image.

[0042] The medical image compression device can read the medical image sequence in sequence and perform subsequent processing on the images in the medical image sequence. The current image can be an image in the medical image sequence being processed by the medical image compression device.

[0043] Reference Figure 3 , Figure 3 Schematic diagram of hash matching between the lesion area of ​​the current image and the lesion area of ​​the reference image.

[0044] In some embodiments, the medical image compression device hash-encodes the lesion region of the current image using a first encoding method, such as perceptual hash encoding based on a discrete cosine transform (DCT). The medical image compression device also hash-encodes the lesion region of the reference image using the first encoding method. The medical image compression device performs hash matching on the encoded lesion region of the current image and the lesion region of the reference image.

[0045] Exemplarily, DCT-based perceptual hash coding is applied to the lesion area of ​​the current image. This can be achieved by performing a DCT transform on the lesion area to obtain a DCT coefficient matrix for the lesion area. Calculating the mean of the first DCT coefficients, comparing each first DCT coefficient with the mean, and obtaining a hash string for the first lesion area. The first DCT coefficient is the DCT coefficient in the upper left corner of the DCT coefficient matrix. If the first DCT coefficient is greater than the mean, the hash value corresponding to the first DCT coefficient is set to 1. If the first DCT coefficient is less than the mean, the hash value corresponding to the first DCT coefficient is set to 0. This results in a hash code for the lesion area of ​​the current image. Similarly, a hash code for the lesion area of ​​the reference image can also be obtained.

[0046] The medical image compression device can determine whether the lesion area of ​​the current image matches the lesion area of ​​the reference image by calculating whether the hash prefix of the lesion area of ​​the current image is the same as the hash prefix of the lesion area of ​​the reference image. Taking the example that the hash code of the lesion area of ​​the current image and the hash code of the lesion area of ​​the reference image are both 64 bits, it is assumed that the hash prefix is ​​the first 32 bits of the hash code. That is, if the first 32 bits of the code are the same, it is determined that the hash values ​​of the lesion area of ​​the current image and the lesion area of ​​the reference image match. If the first 32 bits of the code are different, it is determined that the hash values ​​of the lesion area of ​​the current image and the lesion area of ​​the reference image do not match.

[0047] If the hash values ​​of the lesion region of the current image match those of the lesion region of the reference image, then step 1041 is executed. If the hash values ​​of the lesion region of the current image do not match those of the lesion region of the reference image, then step 1042 is executed.

[0048] Step 1041: The medical image compression device sends a first pixel.

[0049] The first pixel is a pixel that is different between the lesion area of ​​the current image and the lesion area of ​​the reference image.

[0050] The medical image compression device can perform pixel-level difference detection on the lesion area of ​​the current image and the lesion area of ​​the reference image to obtain the first pixel of the lesion area of ​​the current image.

[0051] For example, the current image is a lung CT image, and when compared with the lesion area of ​​the reference image, only 1% of the pixels are different, and only 1% of the pixels are transmitted.

[0052] By sending the first pixel instead of sending pixels that are repeated in the lesion area of ​​the reference image, the sending of repeated and useless pixels is reduced, thereby reducing the resources occupied by medical image transmission. In addition, the image processing device can also receive less data information when receiving medical images. For example, after decompressing the first pixel, the complete lesion area information of the current image can be obtained based on the existing pixels of the lesion area other than the first pixel. The existing pixels of the lesion area other than the first pixel can be obtained through the reference image. For example, the image processing device can read the reference image. As a result, the medical image compression device can transmit the key lesion area information of subsequent images more quickly, allowing the image processing device to receive more important medical image information as quickly as possible.

[0053] In some embodiments, sending the first pixel may be implemented as sending the position of the first pixel in the lesion area of ​​the current image and the color value of the pixel. The color value of the pixel may be a grayscale value, a color value in a red, green, and blue color space, etc.

[0054] In some embodiments, sending the first pixel can be implemented as sending a hash code of the lesion area of ​​the current image. For example, the hash code of the lesion area of ​​the current image can be the above-mentioned 64-bit hash code. For another example, the hash code can be a hash code of bits 33 to 64. By sending fewer hash codes, the image receiving end can obtain a 64-bit hash code of the lesion area of ​​the current image based on the hash code of the lesion area of ​​the reference image, thereby obtaining the lesion area of ​​the current image.

[0055] In step 1042 , the medical image compression apparatus divides the lesion region of the current image into M sub-lesion regions, and performs hash matching on the M sub-lesion regions with the sub-lesion regions of the reference image respectively.

[0056] Where M is a positive integer. Take M=4 as an example, refer to Figure 3 The medical image compression apparatus divides the lesion region of the current image (e.g., image 1) into four sub-lesion regions, namely sub-lesion region 1 to sub-lesion region 4. Accordingly, the medical image compression apparatus may divide the sub-lesion region of the reference image into sub-lesion regions 1 to sub-lesion region 4 according to the four sub-lesion regions of the current image.

[0057] After the medical image compression device divides the sub-lesion area of ​​the reference image into sub-lesion area 1 to sub-lesion area 4, when subsequently processing the next image of the medical image sequence (such as image 2 after image 1), the image can be divided into sub-lesion area 1 to sub-lesion area 4 according to the division method of the sub-lesion area of ​​the reference image.

[0058] The medical image compression device can determine whether the sub-lesion area of ​​the current image matches the sub-lesion area of ​​the reference image by calculating whether the hash prefix of the sub-lesion area of ​​the current image is the same as the hash prefix of the sub-lesion area of ​​the reference image. Taking the hash code of the sub-lesion area of ​​the current image and the hash code of the sub-lesion area of ​​the reference image as an example, it is assumed that the hash prefix is ​​the first 16 bits of the hash code. That is, if the first 16 bits are the same, it is determined that the hash values ​​of the sub-lesion area of ​​the current image match the hash values ​​of the sub-lesion area of ​​the reference image. If the first 16 bits are different, it is determined that the hash values ​​of the sub-lesion area of ​​the current image do not match the hash values ​​of the sub-lesion area of ​​the reference image.

[0059] If the hash values ​​of the first subregion of the current image and the first subregion of the reference image match, step 1051 is executed. If the hash values ​​of the second subregion of the current image and the second subregion of the reference image do not match, step 1052 is executed. The first subregion and the second subregion are subregions of the M sub-lesion regions. It should be understood that the first subregion does not refer to a specific subregion, but rather to each subregion whose hash value matches the corresponding subregion of the reference image. To distinguish it from other regions (such as subregions whose hash values ​​do not match the corresponding subregion of the reference image), the term "first subregion" is used here. Similarly, the second subregion does not refer to a specific subregion. The number of first subregions is determined based on the number of subregions whose hash values ​​match the corresponding subregion of the reference image. The number of second subregions is determined based on the number of subregions whose hash values ​​do not match the corresponding subregion of the reference image. For example, the current image is divided into 4 sub-regions, 3 sub-regions match the sub-region hash values ​​of the corresponding reference image, and 1 sub-region does not match the sub-region hash value of the corresponding reference image, then the current image includes 3 first sub-regions and 1 second sub-region.

[0060] Reference Figure 3 For example, if sub-lesion region 1 of the current image matches sub-lesion region 1 of the reference image (an example of a first sub-region), step 1051 is continued for sub-lesion region 1 of the current image. If sub-lesion region 2 of the current image does not match sub-lesion region 2 of the reference image (an example of a second sub-region), steps 1052 and 1503 are continued for sub-lesion region 2 of the current image.

[0061] Step 1051: The medical image compression apparatus sends a second pixel.

[0062] The second pixel is a different pixel between the first sub-region of the current image and the first sub-region of the reference image.

[0063] Taking the sub-lesion region 1 of the current image as the first sub-region of the current image as an example, if the hash values ​​of the sub-lesion region 1 of the current image and the sub-lesion region 1 of the reference image match, the medical image compression device can perform pixel-level difference detection on the sub-lesion region 1 of the current image and the sub-lesion region 1 of the reference image to obtain the second pixel of the sub-lesion region 1 of the current image.

[0064] In some embodiments, sending the second pixel may be implemented by sending the position of the second pixel in the first sub-region of the current image and the color value of the pixel.

[0065] In some embodiments, sending the second pixel may be implemented as sending a hash code of the first sub-region of the current image.

[0066] Step 1052: The medical image compression apparatus compresses the second sub-region of the current image using a first compression method.

[0067] Step 1053: The medical image compression apparatus sends the compressed second sub-region.

[0068] In some embodiments, the medical image compression apparatus compresses the second sub-region of the current image by using an EZW encoding method based on wavelet transform.

[0069] Through the solution of the embodiment of the present application, the medical image compression device compares the images in the medical image sequence with the reference images multiple times, and adopts different transmission schemes for the lesion area based on the comparison results, thereby reducing the resources consumed in sending medical images and improving the efficiency of sending medical images.

[0070] In some other embodiments of the present application, the medical image compression apparatus also transmits the non-lesion area. The following describes how the medical image compression apparatus transmits the pixels of the non-lesion area. The present embodiment further includes step 201.

[0071] In step 201 , the medical image compression apparatus performs hash matching on the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image.

[0072] Referring to the above embodiments and Figure 3 The medical image compression device can segment the lesion area and the non-lesion area of ​​the reference image to obtain the non-lesion area of ​​the reference image corresponding to the non-lesion area of ​​the current image.

[0073] In some embodiments, the medical image compression apparatus performs hash encoding on the non-lesion region of the current image using a second encoding method. For example, the second encoding method is a fast hash algorithm (CH-hash) based on a color or grayscale histogram. The medical image compression apparatus also performs hash encoding on the non-lesion region of the reference image using the second encoding method.

[0074] Exemplarily, the non-lesion area of ​​the current image is subjected to fast hash coding based on color histogram, which can be implemented as follows: grayscale normalization of the non-lesion area. There are S grayscale levels in total. Count the number of pixels of each grayscale level in the S grayscale levels. Divide the S grayscale levels into D grayscale groups, and obtain the number of pixels in the D grayscale groups and the pixel average of the D grayscale groups. Compare the number of pixels in the D grayscale groups with the pixel average of the D grayscale groups to obtain the first non-lesion area hash string. If the number of pixels in the Fth grayscale group is greater than the pixel average of the D grayscale groups, the hash value of the Fth grayscale group is 1. If the number of pixels in the Fth grayscale group is not greater than the pixel average of the D grayscale groups, the hash value of the Fth grayscale group is 0. 1≤F≤D<S, where F, D, and S are integers. Similarly, the hash code of the non-lesion area of ​​the reference image can also be obtained.

[0075] Similar to performing a hash match between the lesion area of ​​the current image and the lesion area of ​​the reference image, the medical image compression device can determine whether the non-lesion area of ​​the current image matches the non-lesion area of ​​the reference image by calculating whether the hash prefix of the non-lesion area of ​​the current image is the same as the hash prefix of the non-lesion area of ​​the reference image. If the hash values ​​of the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image match, the medical image compression device may no longer process the current image. The medical image compression device may continue to process the next image in the medical image sequence, using the next image in the medical image sequence as the current image, and re-execute step 102 and subsequent steps. If the hash values ​​of the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image do not match, step 2021 is executed.

[0076] In this embodiment of the present application, if the hash values ​​of the non-lesion area of ​​the current image match those of the non-lesion area of ​​the reference image, it indicates that the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image are highly similar. Considering that the non-lesion area is less important for medical diagnosis, the differences in the non-lesion area can be discontinued, thereby saving communication resources and allowing more useful information to be transmitted within limited transmission resources.

[0077] In step 2021 , the medical image compression apparatus divides the non-lesion area of ​​the current image into N sub-non-lesion areas, and performs hash matching on the N sub-non-lesion areas with the sub-non-lesion areas of the reference image.

[0078] wherein N is a positive integer. Taking N=4 as an example, referring to FIG. 2, the medical image compression device divides the lesion region of the current image into four sub-non-lesion regions, which are sub-non-lesion region 1 to sub-non-lesion region 4. Correspondingly, the medical image compression device can divide the sub-non-lesion region of the reference image into sub-non-lesion region 1 to sub-non-lesion region 4 according to the four sub-non-lesion regions of the current image. Figure 3

[0079] The medical image compression device can determine whether the sub-non-lesion region of the current image matches the sub-non-lesion region of the reference image by calculating whether the hash prefix of the sub-non-lesion region of the current image is the same as the hash prefix of the sub-non-lesion region of the reference image. If the fourth sub-region of the current image matches the fourth sub-region hash value of the reference image, the medical image compression device can no longer process the current image. The medical image compression device can continue to process the next image in the medical image sequence, take the next image in the medical image sequence as the current image, and re-execute step 102 and the subsequent steps. If the third sub-region of the current image does not match the third sub-region hash value of the reference image, step 2031 is executed. The third sub-region is a sub-region in the N sub-non-lesion regions. Similar to the first sub-region and the second sub-region described above, the third sub-region is not a specific sub-region, but refers to each sub-region.

[0080] Step 2031, the medical image compression device determines whether the number of third pixels is greater than a first threshold.

[0081] wherein the third pixel is a pixel in the third sub-region of the current image that is different from the third sub-region of the reference image.

[0082] Referring to FIG. 2, the medical image compression device can determine whether the proportion of the third pixel in the third sub-region is greater than the first threshold. For example, the first threshold is 10%. Taking the sub-non-lesion region 2 of the current image as the third sub-region of the current image and the total number of pixels of the sub-non-lesion region 2 as an example. Assuming that the first threshold is 10% of the total number of pixels of the sub-non-lesion region 2, when the number of third pixels is greater than or equal to 410, it indicates that the number of third pixels is greater than the first threshold. When the number of third pixels is less than 410, it indicates that the number of third pixels is not greater than the first threshold. Assuming that the first threshold is 20% of the total number of pixels of the sub-non-lesion region 2, when the number of third pixels is greater than 820, it indicates that the number of third pixels is greater than the first threshold. Figure 3 For another example, the medical image compression device can determine whether the total number of third pixels in the third sub-region exceeds the first threshold. For example, the first threshold is 500 pixels. Thus, the medical image compression device can determine whether the number of third pixels is greater than the first threshold.

[0083] For another example, the medical image compression device can determine whether the total number of third pixels in the third sub-region exceeds the first threshold. For example, the first threshold is 500 pixels. Thus, the medical image compression device can determine whether the number of third pixels is greater than the first threshold.​

[0084] If the number of the third pixels is greater than the first threshold, then steps 2041 and 2042 are performed. If the number of the third pixels is not greater than the first threshold, then step 2043 is performed.

[0085] In step 2041 , the medical image compression apparatus compresses the third sub-region of the current image using a second compression method.

[0086] Step 2042: The medical image compression apparatus sends the compressed third sub-region of the current image.

[0087] Step 2043: The medical image compression device sends the third pixel.

[0088] Exemplarily, the second compression method is a predictive coding method based on a color histogram.

[0089] In some embodiments, the compression ratio of the first compression method is lower than the compression ratio of the second compression method. For example, the compression ratio of the first compression method is 20:1, and the compression ratio of the second compression method is 100:1. Because the second sub-region is the lesion area, the medical information reflected by the lesion area is more important than the medical information of the non-lesion area. Using a lower compression ratio can reduce the loss of important medical information and preserve the integrity of the medical data of the lesion area to a greater extent.

[0090] In other embodiments, if the hash value of the non-lesion area of ​​the current image matches the hash value of the non-lesion area of ​​the reference image, the medical image compression apparatus may execute step 2022 .

[0091] In step 2022, the medical image compression apparatus sends a fourth pixel, where the fourth pixel is a different pixel between the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image.

[0092] If the hash value of the fourth subregion of the current image matches the hash value of the fourth subregion of the reference image, the medical image compression apparatus may execute step 2032. The fourth subregion is a subregion of the N sub-non-lesion regions. Similar to the first to third subregions described above, the fourth subregion does not refer to a specific subregion, but rather refers to all subregions.

[0093] Step 2032: The medical image compression apparatus sends the fifth pixel.

[0094] The fifth pixel is a different pixel between the fourth sub-region of the current image and the fourth sub-region of the reference image.

[0095] Through this embodiment, the medical image compression device can transmit pixels of a non-lesion area that differs from the reference image, allowing the receiving device to receive the complete current image. For example, in scenarios where network conditions are good and the speed of acquiring medical images is not critical, the medical image compression device can transmit the complete current image.

[0096] In the above embodiment, the medical image compression device transmits pixels in the lesion area first, followed by pixels in the non-lesion area. This allows important information to be transmitted first. This allows doctors to receive important information first at the receiving end, reducing the time doctors have to wait for data and accelerating patient diagnosis.

[0097] The above embodiment describes performing a multi-level hash comparison between the current image and the reference image, thereby transmitting pixels in the current image that differ from the reference image. In some embodiments, similar to transmitting the first and second pixels, transmitting the third through fifth pixels, etc., can also be implemented by transmitting the positions and color values ​​of the third through fifth pixels in the lesion area of ​​the current image. Alternatively, it can be implemented by transmitting the hash code of the area containing the third through fifth pixels.

[0098] In some embodiments, the length of the hash code may be adaptively adjusted. The adaptive adjustment of the hash code length is described below.

[0099] In some implementation examples, the computing device adjusts the target image hash code length according to the texture complexity or information complexity of the target image.

[0100] Texture complexity is used to indicate the complexity of the texture in the target image, and information complexity is used to indicate the randomness of the color distribution in the target image.

[0101] Exemplarily, the target area may be the lesion area of ​​the current image, the first sub-area of ​​the current image, the third sub-area of ​​the current image, the non-lesion area of ​​the current image, the fourth sub-area of ​​the current image, etc.

[0102] Exemplarily, the texture complexity of the target image is implemented as a local binary pattern (LBP) or a histogram of gradients (HOG) of the target image.

[0103] For example, the target image is a grayscale image, and the information complexity is used to indicate the randomness of the grayscale distribution in the target image. The information complexity of the target region is implemented as the entropy value of the grayscale distribution of the target image.

[0104] In some examples, if the texture complexity of the target area is greater than the second threshold, the hash length is set to the first length. If the texture complexity of the target area is not greater than the second threshold and greater than the third threshold, the hash length is set to the second length; if the texture complexity of the target area is not greater than the third threshold, the hash length is set to the third length. The first length is greater than the second length and greater than the third length. The second threshold is greater than the third threshold. For example, the first length is 64, the second length is 32, and the third length is 16. Taking the texture complexity of the target image as the LBP of the target image as an example, the second threshold is 5 and the third threshold is 3. Taking the texture complexity of the target image as the HOG of the target image as an example, the second threshold is 0.4 and the third threshold is 0.2.

[0105] In other examples, if the information complexity of the target region is greater than a fourth threshold, the hash length is set to the first length. If the information complexity of the target region is not greater than the fourth threshold, the hash length is set to the second length. Exemplarily, the fourth threshold is 5.

[0106] In other examples, if the texture complexity of the target area is greater than the second threshold and the information complexity is greater than the fourth threshold, the hash length is set to the first length. If the information complexity of the target area is not greater than the fourth threshold but the texture complexity of the target area is greater than the second threshold, the hash length is set to the second length. If the texture complexity of the target area is not greater than the second threshold but the information complexity of the target area is greater than the fourth threshold, the hash length is set to the second length. If the texture complexity of the target area is not greater than the second threshold and the information complexity is not greater than the fourth threshold, the hash length is set to the third length. Wherein, the first length is greater than the second length and greater than the third length.

[0107] If the target area has high texture complexity and information complexity, it means that the image in the target area reflects more medical information, so a longer hash is used to retain more medical information. If the target area has low texture complexity and information complexity, it means that the image in the target area reflects less medical information, so a shorter hash is used to retain less medical information, saving resources.

[0108] In other embodiments, the computing device may output a hash code length (e.g., 16, 32, 64) based on image features (e.g., texture complexity, information complexity) through a machine learning model.

[0109] Exemplarily, the machine learning model is a regression model, such as a random forest.

[0110] The machine learning model can also output a hash code length (e.g., 16, 32, or 64) based on the target image classification. For example, if the target image is an image of a lesion area, a 32- or 64-bit hash code is output. If the target image is an image of a non-lesion area, a 16- or 32-bit hash code is output.

[0111] The above embodiment describes how to transmit the lesion and non-lesion areas of a medical image. The following describes how to segment the lesion and non-lesion areas of a current image. In some embodiments, step 102 of the above method can identify regions of interest (i.e., lesion areas) in images of multiple modalities and perform image segmentation. Step 102 involves segmenting the lesion and non-lesion areas of images in a medical image sequence using a deep learning segmentation model using an improved U-Net architecture (hereinafter referred to as the first model). The first model is described below.

[0112] After receiving the medical image sequence, the first model can perform data preprocessing, such as pixel normalization, registration, cropping, and scaling. Registration refers to spatially aligning images from different modalities so that the same anatomical structure is consistently positioned across modalities. Cropping and scaling refers to cropping and scaling images to a uniform size.

[0113] The medical image sequence may include images of various modalities including CT images, PET images, and MRI images.

[0114] The first model can extract features from images in a medical image sequence based on the modality to which the images belong. The modality of the images in the medical image sequence is computed tomography (CT), magnetic resonance imaging (MRI), or positron emission tomography (PET). For example, the medical image sequence may include lung CT images, lung MRI images, and lung PET images.

[0115] In some embodiments, the first model extracts features from CT images using filters or attention mechanisms that target edge and shape information. For example, a gradient attention mechanism and a spatial convolution group are used for feature extraction. For MRI images, a feature extraction method based on texture and soft tissue contrast is used. For PET images, a feature extraction method based on metabolically active regions is used.

[0116] In some embodiments, the encoder of the first model extracts features of the images in the medical image sequence. After the input images are extracted, the features of the images in the medical image sequence of multiple modalities are fused in the intermediate layer of the first model or the decoder. For example, the encoder of the first model includes P layers, wherein Q layers in the P layers include feature maps of the images, the scales of the feature maps generated by each of the Q layers are different, the feature maps generated by each of the Q layers are fused to the layers of the decoder through the jump connection respectively, P is greater than Q, and P and Q are positive integers.

[0117] For example, the feature fusion of the multi-modal images can be implemented by splicing the feature maps output by the branches of different modalities in the same spatial size level in the channel dimension to form a feature map with more channels.

[0118] For another example, the feature fusion of the multi-modal images can be implemented by taking the average, maximum or minimum value of the pixels in the same position of the feature maps output by the branches of different modalities in the same spatial size level.

[0119] For another example, the feature fusion of the multi-modal images can be implemented by weighted summation of the feature maps output by the branches of different modalities in the same spatial size level through an attention module. The attention module can learn the weights of the features of different modal images and realize the feature fusion of the multi-modal images.

[0120] In some embodiments, the encoder of the first model generates feature maps of different scales in different layers. The feature map of each layer contains the corresponding features of the layer. Through multi-scale jump connection, the feature maps of different scales are fused into the corresponding layers of the decoder, so that the decoder can utilize the information of different scales at the same time, thereby improving the accuracy of segmentation.

[0121] In some embodiments, the loss function of the first model is a modality consistency loss function. The modality consistency loss function is a loss function for multi-modal learning, which aims to promote the consistency of the prediction results from different modalities to a certain extent.

[0122] For example, the modality consistency loss function is a combination of Dice loss and focal loss. For example, the modality consistency loss function is obtained by weighted summation of the Dice loss and the focal loss.

[0123] The Dice loss can make the overlapping part of the prediction result and the true label as large as possible. The focal loss can improve the segmentation performance of the model on the lesions that are difficult to classify. Therefore, by applying the combination of the Dice loss and the focal loss, both the overlapping degree of the segmentation result and the problem of sample imbalance can be considered, and the lesion area and the non-lesion area can be better distinguished.

[0124] In some embodiments, the modality of the current image is CT, and the processing of the current image by the first model includes steps 301 to 305 .

[0125] In step 301 , the first model identifies that the modality of the current image is CT.

[0126] Step 302: The first model calculates a gradient magnitude map of the CT image.

[0127] In step 303, the first model uses the gradient magnitude map as an attention signal to generate an attention weight map.

[0128] In step 304, the first model applies the attention weight map to the original feature map of the current image or the feature map after preliminary convolution to obtain a weighted feature map.

[0129] Step 305: In the convolution layer, the first model uses multiple dilated convolution kernels with different expansion rates to process the weighted feature map to obtain a first image.

[0130] The first image is an image obtained after CT image features are extracted.

[0131] The above method focuses more on extracting edge information of the CT image, so that the first model can better extract the features of the CT image.

[0132] In some embodiments, the modality of the current image is MRI, and the processing of the current image by the first model includes steps 401 to 404 .

[0133] In step 401 , the first model identifies that the modality of the current image is MRI.

[0134] MRI typically includes multiple sequences, such as T1, T2, and FLAIR. Each sequence reflects different tissue characteristics. For example, the T2 sequence is sensitive to edema, while the FLAIR sequence suppresses cerebrospinal fluid signals.

[0135] In step 402, the first model extracts MRI features through phase congruency attention to obtain a weighted feature map, wherein the weighted sequence includes T1-weighted image, T2-weighted image or FLAIR sequence.

[0136] The first model can input the image into the feature extraction module corresponding to the sequence to obtain the features of the sequence according to the sequence to which the MRI belongs. For example, the sequence to which the MRI belongs is the T1 sequence. The first model inputs the image into the feature extraction module corresponding to the T1 sequence to extract the features of the sequence. After the features of each sequence are extracted, phase consistency attention can be used to assign different weights to the feature channels from different sequences, and then perform weighted fusion. Phase consistency attention can use frequency domain phase information to enhance the edge contrast of the image.

[0137] In step 403 , the first model performs texture analysis on the weighted feature map using one or more of a gray level co-occurrence matrix, a local binary pattern, or a Gabor filter to obtain texture features of the lesion area and normal tissue.

[0138] In step 404 , the first model uses multi-scale analysis on lesion regions of different sizes in the MRI image to extract features of the lesion regions to obtain a second image.

[0139] Exemplarily, multi-scale analysis is implemented by outputting MRI features from different layers of a U-Net encoder. Furthermore, multi-scale analysis is implemented by processing the texture-analyzed MRI image using convolution kernels of different sizes and dilated convolution.

[0140] The second image is an image obtained by extracting MRI image features.

[0141] MRI images can demonstrate high soft tissue contrast. Phase congruent attention extraction can enhance the edge contrast of lesion areas, providing rich anatomical details and pathological change information. Different weighting sequences can highlight different types of tissue characteristics. Texture analysis and multi-scale analysis of the weighted MRI feature map can further extract MRI image features. Thus, through the above method, the first model can better extract MRI image features.

[0142] In some embodiments, the modality of the current image is PET, and the processing of the current image by the first model includes steps 501 and 502 .

[0143] Step 501: The first model identifies the modality of the current image as PET.

[0144] PET is a functional imaging technique that reflects the metabolic activity of tissues by tracking the distribution and uptake of radioactive tracers in the body.

[0145] In step 502 , the first model extracts features related to fluorodeoxyglucose (FDG) uptake values ​​in PET to obtain a third image.

[0146] The third image is an image obtained after feature extraction from the PET image.

[0147] The second neural network model learns high FDG uptake areas and extracts features from areas with high FDG uptake values. The high FDG uptake areas are marked in the training set images.

[0148] Among them, the characteristics related to FDG uptake value can be determined by one or more of the following: mean standardized uptake value (SUV), maximum SUV, metabolite volume (MTV), total lesion glycolysis (TLG), etc.

[0149] In some embodiments, the first model may also utilize super-resolution techniques based on deep learning (e.g., a generative adversarial network (GAN)) to improve the effective spatial resolution of the PET image before executing step 502. Because PET images have relatively low spatial resolution, they are typically blurry. This step can yield clearer edges and internal detail features, thereby improving feature extraction from PET images.

[0150] The above scheme extracts features related to FDG uptake values, thereby achieving feature extraction of PET images, so that in subsequent operations, the first model can separate lesion areas from non-lesion areas based on the PET image features.

[0151] By employing a targeted feature extraction module for each modality, the encoder can more effectively extract modality-specific information related to the lesion area during downsampling and feature extraction. These features are passed to the decoder via the U-Net's U-shaped structure. The decoder then uses this information for pixel-level classification, ultimately outputting the segmented lesion and non-lesion areas.

[0152] A medical image compression method in an embodiment of the present application is described above. A medical image compression device (such as a computer) that executes the above medical image compression method is introduced below.

[0153] See Figure 4 ,like Figure 4 The medical image compression device in the embodiment of the present application can realize the above-mentioned Figure 2The steps of the medical image compression method executed in the corresponding embodiment. The functions implemented by the medical image compression device can be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules can be software and / or hardware. The medical image compression device may include an input and output module 601 and a processing module 602. The functions of the processing module 602 and the input and output module 601 can be implemented by reference to Figure 2 The operations performed in the corresponding embodiments are not described in detail here. For example, the processing module 602 can be used to control the operations of the input / output module 601 such as sending, receiving, and obtaining.

[0154] The input and output module 601 is configured to acquire a medical image sequence.

[0155] The processing module 602 is configured to segment the lesion area and the non-lesion area of ​​an image in the medical image sequence.

[0156] Processing module 602 is further configured to perform a hash match between the lesion region of the current image and the lesion region of the reference image. If the hash values ​​of the lesion region of the current image and the lesion region of the reference image match, a first pixel is sent. The first pixel is a pixel that differs between the lesion region of the current image and the lesion region of the reference image. The current image is an image in a medical image sequence.

[0157] The processing module 602 is also configured to divide the lesion area of ​​the current image into M sub-lesion areas if the hash value of the lesion area of ​​the current image does not match the hash value of the lesion area of ​​the reference image, and hash match the M sub-lesion areas with the sub-lesion areas of the reference image respectively.

[0158] The input / output module 601 is further configured to send a second pixel if the hash value of the first subregion of the current image matches the hash value of the first subregion of the reference image. The second pixel is a pixel that is different between the first subregion of the current image and the first subregion of the reference image, where the first subregion is a subregion of the M sub-lesion regions, and M is a positive integer.

[0159] The processing module 602 is further configured to compress the second subregion of the current image using a first compression method if the hash value of the second subregion of the current image does not match the hash value of the second subregion of the reference image.

[0160] The input / output module 601 is further configured to send the compressed second sub-region, which is a sub-region among the M sub-lesion regions.

[0161] In some embodiments, the processing module 602 is further configured to perform a hash match on the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image. If the hash values ​​of the non-lesion area of ​​the current image do not match those of the non-lesion area of ​​the reference image, the non-lesion area of ​​the current image is divided into N sub-non-lesion areas, and the N sub-non-lesion areas are hash matched with the sub-non-lesion areas of the reference image, respectively. Wherein N is a positive integer. If the hash values ​​of the third sub-area of ​​the current image do not match those of the third sub-area of ​​the reference image, it is determined whether the number of the third pixels is greater than the first threshold. If the number of the third pixels is greater than the first threshold, the third sub-area of ​​the current image is compressed using the second compression method.

[0162] The input / output module 601 is further configured to transmit the compressed third sub-region of the current image. The third pixel is a pixel in the third sub-region of the current image that is different from the third sub-region of the reference image. If the number of the third pixels is not greater than the first threshold, the third pixel is transmitted.

[0163] In some embodiments, the input / output module 601 is further configured to send a fourth pixel if the hash value of the non-lesion area of ​​the current image matches the hash value of the non-lesion area of ​​the reference image, wherein the fourth pixel is a pixel that is different between the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image.

[0164] The input / output module 601 is further configured to send a fifth pixel if the hash value of the fourth subregion of the current image matches the hash value of the fourth subregion of the reference image. The fifth pixel is a pixel that is different between the fourth subregion of the current image and the fourth subregion of the reference image, and the fourth subregion is a subregion of the N sub-non-lesion regions.

[0165] In some embodiments, the compression ratio of the first compression mode is smaller than the compression ratio of the second compression mode.

[0166] In some embodiments, the processing module 602 is further configured to segment the lesion area and non-lesion area of ​​a medical image sequence using a deep learning segmentation model with an improved U-Net architecture, where the modality of the images in the medical image sequence is computed tomography (CT), magnetic resonance imaging (MRI), or positron emission tomography (PET). The deep learning segmentation model with an improved U-Net architecture extracts features from the current image based on the modality to which the current image belongs. The deep learning segmentation model with an improved U-Net architecture performs feature fusion on the feature extraction results of images of multiple modalities. The loss function of the deep learning segmentation model with an improved U-Net architecture is a combination of Diess loss and focal loss.

[0167] In some embodiments, the processing module 602 is further configured to use an improved U-Net architecture deep learning segmentation model to identify that the modality of the current image is CT. The improved U-Net architecture deep learning segmentation model calculates the gradient amplitude map of the current image. The improved U-Net architecture deep learning segmentation model uses the gradient amplitude map as an attention signal to generate an attention weight map. The improved U-Net architecture deep learning segmentation model applies the attention weight map to the original feature map of the current image or the feature map after preliminary convolution to obtain a weighted feature map. In the convolution layer, the improved U-Net architecture deep learning segmentation model uses a plurality of dilation kernels with different expansion rates to process the weighted feature map to obtain a first image, wherein the first image is the image after feature extraction of the current image.

[0168] In some embodiments, the processing module 602 is further configured to apply DCT-based perceptual hash coding to the lesion area of ​​the current image. Apply DCT-based perceptual hash coding to the lesion area of ​​the reference image. Hash matching is performed on the encoded lesion area of ​​the current image and the lesion area of ​​the reference image. Hash matching is performed on the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image, including: applying color histogram-based fast hash coding to the non-lesion area of ​​the current image and applying color histogram-based fast hash coding to the non-lesion area of ​​the reference image. Hash matching is performed on the encoded non-lesion area of ​​the current image and the non-lesion area of ​​the reference image.

[0169] In the embodiment of the present application, the processing module 602 segments the medical image into lesion areas and non-lesion areas. Based on the matching results of the lesion area of ​​the image and the lesion area of ​​the reference image, the lesion area image is transmitted by sending the pixels that differ from the reference image when similar. When dissimilar, the lesion area is partitioned and the sub-lesion areas of the partitioned image are compared with the sub-lesion areas of the reference image. If the sub-lesion area is similar to the reference image, the input / output module 601 sends the pixels that differ from the sub-lesion area of ​​the reference image. If they are dissimilar, the sub-lesion area is compressed and the compressed sub-lesion area is sent. This achieves compression of the medical image while ensuring the transmission quality of the medical image, saving transmission resources. It also speeds up the determination of compressed pixels and improves image compression efficiency.

[0170] The medical image compression device 60 in the embodiment of the present application is described above from the perspective of modular functional entities. The medical image compression device in the embodiment of the present application is described below from the perspective of hardware processing.

[0171] It should be noted that Figure 4The physical device corresponding to the input / output module 601 shown may be a transceiver, a radio frequency circuit, a communication module, an input / output (I / O) interface, etc., and the physical device corresponding to the processing module 602 may be a processor.

[0172] Figure 4 The medical image compression device 60 shown has the following features: Figure 5 When the structure shown is Figure 5 The processor and transceiver in the device can implement the same or similar functions as the processing module 602 and the input / output module 601 provided in the aforementioned device embodiment corresponding to the device.

[0173] An embodiment of the present application also relates to a chip system, which includes at least one processor and an interface circuit, wherein the processor includes multiple vector storage units, and the processor is used to execute instructions and / or data interaction through the interface circuit, so that the chip system executes the method of any of the above embodiments.

[0174] In a possible implementation, the chip system may also directly include a memory, in which a computer program or computer instructions are stored.

[0175] For example, the memory may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache memory. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM).

[0176] The embodiment of the present application also relates to a processor comprising a plurality of storage units, which are used to call computer programs or computer instructions stored in the memory, so that the processor executes the method described in any of the above embodiments.

[0177] For example, in the embodiment of the present application, the processor is an integrated circuit chip with signal processing capability. For example, the processor can be an FPGA, a general processor, a DSP, an ASIC or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, a SoC, a CPU, a network processor (NP), a micro controller unit (MCU), a PLD or other integrated chip, which can implement or execute the disclosed methods, steps and logic block diagrams in the embodiment of the present application. In a possible implementation manner, the embodiment of the present application further provides a computer readable storage medium, which stores program codes, when the program codes run on the computer, the computer executes the above method embodiment.

[0178] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0179] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and module described above can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

[0180] In the several embodiments provided by the embodiment of the present application, it should be understood that the disclosed system, device and method can be implemented by other manners. For example, the above-described device embodiment is only schematic, for example, the division of the modules is only a logical function division, and there can be another division manner in actual implementation, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed modules can be indirect coupling or communication connection through some interface, device or module, and can be electrical, mechanical or other forms.

[0181] The modules described as separate components can or can not be physically separate, and the components shown as modules can or can not be physical modules, that is, they can be located in one place, or can be distributed on a plurality of network modules. According to actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment of the present application.

[0182] In addition, the functional modules in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into a module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0183] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0184] The computer program product includes one or more computer instructions. When the computer program is loaded and executed on a computer, the process or function described in accordance with the embodiments of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium may be any available medium that can be stored on a computer or a data storage device such as a server or data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, hard disk, or magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0185] The above is a detailed introduction to the technical solutions provided in the embodiments of the present application. Specific examples are used in the embodiments of the present application to illustrate the principles and implementation methods of the embodiments of the present application. The description of the above embodiments is only used to help understand the methods and core ideas of the embodiments of the present application. At the same time, for those skilled in the art, according to the ideas of the embodiments of the present application, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the embodiments of the present application.

Claims

1. A medical image compression method, characterized in that: The method comprises: Acquire medical image sequences; Segmenting the lesion area and the non-lesion area of ​​the image in the medical image sequence; Performing hash matching on the lesion area of ​​the current image and the lesion area of ​​the reference image; if the hash values ​​of the lesion area of ​​the current image and the lesion area of ​​the reference image match, sending a first pixel; wherein the first pixel is a pixel that is different between the lesion area of ​​the current image and the lesion area of ​​the reference image; the current image is an image in the medical image sequence; If the hash value of the lesion area of ​​the current image does not match the hash value of the lesion area of ​​the reference image, dividing the lesion area of ​​the current image into M sub-lesion areas, and performing hash matching on each of the M sub-lesion areas with the sub-lesion areas of the reference image; If the hash value of the first subregion of the current image matches the hash value of the first subregion of the reference image, a second pixel is sent; wherein the second pixel is a different pixel between the first subregion of the current image and the first subregion of the reference image, and the first subregion is a subregion of the M sub-lesion regions; M is a positive integer; If the hash value of the second sub-region of the current image does not match the hash value of the second sub-region of the reference image, compressing the second sub-region of the current image using a first compression method; The compressed second sub-region is sent; the second sub-region is a sub-region in the M sub-lesion regions.

2. The method according to claim 1, characterized in that The method further comprises: Performing hash matching on the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image; If the hash values ​​of the non-lesion area of ​​the current image do not match those of the non-lesion area of ​​the reference image, dividing the non-lesion area of ​​the current image into N sub-non-lesion areas, and performing hash matching on each of the N sub-non-lesion areas with the sub-non-lesion areas of the reference image; wherein N is a positive integer; If the hash value of the third subregion of the current image does not match the hash value of the third subregion of the reference image, determining whether the number of third pixels is greater than a first threshold; wherein the third pixel is a pixel in the third subregion of the current image that is different from the pixel in the third subregion of the reference image; and the third subregion is a subregion of the N sub-non-lesion regions; If the number of the third pixels is greater than the first threshold, compressing the third sub-region of the current image using a second compression method; Sending the compressed third sub-region of the current image; If the number of the third pixels is not greater than the first threshold, the third pixels are sent.

3. The method according to claim 2, characterized in that Also includes: If the hash value of the non-lesion area of ​​the current image matches the hash value of the non-lesion area of ​​the reference image, sending a fourth pixel; wherein the fourth pixel is a different pixel between the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image; If the hash value of the fourth sub-region of the current image matches the hash value of the fourth sub-region of the reference image, the fifth pixel is sent; wherein the fifth pixel is a different pixel between the fourth sub-region of the current image and the fourth sub-region of the reference image, and the fourth sub-region is a sub-region of the N sub-non-lesion regions.

4. The method according to claim 2 or 3, characterized in that A compression ratio of the first compression method is smaller than a compression ratio of the second compression method.

5. The method according to claim 1, wherein The segmenting of the lesion area and the non-lesion area of ​​the image in the medical image sequence includes: Segmenting the lesion area and the non-lesion area of ​​the medical image sequence using a deep learning segmentation model with an improved U-Net architecture, where the images in the medical image sequence belong to a modality of computed tomography (CT), magnetic resonance imaging (MRI), or positron emission tomography (PET); The deep learning segmentation model of the improved U-Net architecture performs feature extraction on the current image according to the modality to which the current image belongs; The deep learning segmentation model of the improved U-Net architecture performs feature fusion on the feature extraction results of images of multiple modalities; The loss function of the deep learning segmentation model of the improved U-Net architecture is a combination of Diess loss and focal loss.

6. The method according to claim 5, characterized in that The deep learning segmentation model of the improved U-Net architecture performs feature extraction on the current image according to the modality to which the current image belongs, including: The deep learning segmentation model of the improved U-Net architecture identifies the modality of the current image as CT; The deep learning segmentation model of the improved U-Net architecture calculates the gradient magnitude map of the current image; The deep learning segmentation model of the improved U-Net architecture uses the gradient magnitude map as an attention signal to generate an attention weight map; The deep learning segmentation model of the improved U-Net architecture applies the attention weight map to the original feature map of the current image or the feature map after preliminary convolution to obtain a weighted feature map; In the convolution layer, the deep learning segmentation model of the improved U-Net architecture uses multiple hole convolution kernels with different expansion rates to process the weighted feature map to obtain a first image, wherein the first image is the image after feature extraction of the current image.

7. The method according to claim 5, characterized in that The performing hash matching on the lesion area of ​​the current image and the lesion area of ​​the reference image includes: applying DCT-based perceptual hash coding to the lesion area of ​​the current image; applying DCT-based perceptual hash coding to the lesion area of ​​the reference image; and performing hash matching on the encoded lesion area of ​​the current image and the lesion area of ​​the reference image; The hash matching of the non-lesion area of ​​the current image and the non-lesion area of ​​the reference image includes: using fast hash coding based on color histogram for the non-lesion area of ​​the current image, and using fast hash coding based on color histogram for the non-lesion area of ​​the reference image; and hash matching of the encoded non-lesion area of ​​the current image and the non-lesion area of ​​the reference image.

8. A medical image compression device, characterized in that: The device includes: an input-output module configured to acquire a sequence of medical images; A processing module configured to segment the lesion area and the non-lesion area of ​​the image in the medical image sequence; The processing module is further configured to perform hash matching on the lesion area of ​​the current image and the lesion area of ​​the reference image; if the hash values ​​of the lesion area of ​​the current image and the lesion area of ​​the reference image match, send a first pixel; wherein the first pixel is a pixel that is different between the lesion area of ​​the current image and the lesion area of ​​the reference image; the current image is an image in the medical image sequence; The processing module is further configured to, if the hash value of the lesion area in the current image does not match the hash value of the lesion area in the reference image, divide the lesion area in the current image into M sub-lesion areas, and perform hash matching on the M sub-lesion areas with the sub-lesion areas of the reference image respectively; The input / output module is further configured to send a second pixel if the hash value of the first sub-region of the current image matches the hash value of the first sub-region of the reference image; wherein the second pixel is a different pixel between the first sub-region of the current image and the first sub-region of the reference image, and the first sub-region is a sub-region of the M sub-lesion regions; M is a positive integer; The processing module is further configured to compress the second subregion of the current image using a first compression method if the hash value of the second subregion of the current image does not match the hash value of the second subregion of the reference image; The input-output module is further configured to send the compressed second sub-region; the second sub-region is a sub-region in the M sub-lesion regions.

9. A computer-readable storage medium, characterized in that The method comprises instructions which, when executed on a computer, cause the computer to perform the method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images