Joint compression method for mass medical image cold data

Abstract

The invention provides a mass medical image cold data joint compression method, which comprises the following steps of: performing fine-grained feature extraction on a large-scale medical image through a feature extraction network, thereby establishing a feature database of the medical image, and storing the extracted image features according to classification; extracting features of a medical image to be compressed by using the feature extraction network, and performing similarity matching; sorting the slices of the reference image and the to-be-compressed image based on the image brightness so as to optimize the compression sequence; and finally, assisting lossy coding of the current image according to the reference image, and assisting lossless coding of the current image according to the original image.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image compression, and in particular relates to a method for jointly compressing massive medical image cold data. Background Art

[0002] Cold data refers to data that is less frequently accessed and less important to business and applications. This data usually needs to be stored for a long time, but does not need to be frequently accessed and processed. In the field of medical imaging, cold data usually includes patients' historical imaging records, long-term tracking data, and research data. Although these data are accessed less frequently, they are of great value in medical research, disease tracking, and patient historical information queries, so they need to be stored long-term and reliably. Most existing compression methods process a single image or a single set of slices, and fail to fully utilize the similarities and correlations between massive data, resulting in poor compression effects. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide a joint compression method for massive medical image cold data, which can effectively utilize the similarities and correlations between massive data, reduce the transmission code stream, and improve the compression ratio. The scheme of the present invention first designs a feature extraction network to extract fine-grained features from large-scale medical images and construct an efficient and retrievable feature library for medical images. Subsequently, the medical images are predicted and sorted based on features such as image brightness, and the compression order of the images is optimized. As a supporting design for the compression scheme, finally, a low-code-rate lossless encoding and decoding scheme for medical images is designed, which uses reference images to perform lossy encoding and decoding on the compressed images in turn, and then performs lossless encoding and decoding on the residuals, so as to achieve lossless encoding and decoding effects with fewer bits. The present invention effectively solves the problem of efficient compression in the storage and transmission of massive medical image cold data.

[0004] The technical solution specifically adopted by the present invention to solve the technical problem is:

[0005] A massive medical image cold data joint compression method, which uses a feature extraction network to perform fine-grained feature extraction on large-scale medical images, thereby establishing a feature database of medical images, and storing the extracted image features by category;

[0006] Then, the feature extraction network is used on the medical image to be compressed to extract features and perform similarity matching;

[0007] Then, based on the image brightness, the slices of the reference image and the image to be compressed are sorted to optimize the compression order;

[0008] Finally, the reference image is used to assist the lossy encoding of the current image, and then the original image is used to assist the lossless encoding of the current image.

[0009] Furthermore, the feature extraction network is used to extract features, which is defined as:

[0010] F i =ε(X i ,θ)

[0011] Where ε represents a feature extractor, which is used to convert the image into a feature representation, preferably using the I-frame encoding technology of a video encoder, and X i represents the input image, θ represents the parameter set of the feature extractor, F i Represents the extracted feature vector or feature map.

[0012] Furthermore, a feature database of medical images is established, and the extracted image features are stored by category as follows:

[0013] DB={(F i ,C i )|i=1,2,…N}

[0014] Where N represents the total amount of medical imaging data, F i Represents the feature vector or feature map of the i-th image, C i Indicates the corresponding classification label; the data in the feature library is classified by C i storage.

[0015] Furthermore, the feature extraction network is used for the medical image to be compressed to extract features and perform similarity matching as follows:

[0016] F cur =ε(I cur ,θ)

[0017]

[0018] Among them, I cur represents the image to be compressed, F cur Represents the feature vector or feature map extracted from the image to be compressed, F * is the feature vector with the smallest distance to the current image feature F, I ref is the corresponding image data, id ref is the corresponding image identifier, dist(·) is the difference between the query feature F and the feature F in the database i function of the distance between them.

[0019] Furthermore, the ordering of the slices of the reference image and the image to be compressed based on a certain indicator is expressed as:

[0020] I ref =Sort(I ref ,metric)

[0021] I cur =Sort(I cur ,metric)

[0022] Among them, Sort(·) is a sorting algorithm, and metric is a sorting indicator, such as image brightness, organ area, and image similarity.

[0023] Furthermore, the auxiliary lossy encoding of the current image according to the reference image is specifically as follows:

[0024] C lossy =E lossy (I ref ,I cur )

[0025]

[0026] Among them, E lossy Represents a lossy coding network, which converts the current image I cur and reference image I ref As input, output lossy coded information C lossy , D lossy Represents a lossy decoding network, which combines the encoded information and the reference image I ref As input, get a lossy reconstruction of the current image

[0027] The auxiliary lossless encoding of the current image according to the original image is specifically as follows:

[0028]

[0029] Among them, E lossless Represents a lossless coding network, which transforms the current image I cur and lossy reconstruction The error between the two images, and the current image I cur As input, output lossless coded information C lossless .

[0030] Furthermore, during decompression, according to the image identifier id ref Obtain reference images from a database, and sort the reference images using a compressed sorting method;

[0031] Then, according to the reference image, the lossy coded information is losslessly decoded, and the formula is expressed as follows:

[0032]

[0033] The lossless coding information is then losslessly decoded, and the formula is expressed as follows:

[0034]

[0035] Among them, D lossless Represents a lossless decoding network, which converts the lossless coded information C lossless As input, obtain a lossless reconstruction of the current image

[0036] A massive medical image cold data joint compression system, comprising:

[0037] Feature extraction module: used to perform fine-grained feature extraction on large-scale medical images through a feature extraction network, thereby establishing a feature database of medical images and storing the extracted image features by category;

[0038] A similarity matching module, used for extracting features and performing similarity matching on the compressed medical image using the feature extraction network;

[0039] A sorting module, used for sorting the slices of the reference image and the image to be compressed based on the image brightness to optimize the compression order;

[0040] The compression module is used to assist the lossy encoding of the current image based on the reference image, and then assist the lossless encoding of the current image based on the original image.

[0041] Furthermore, it also includes a decompression module: used to obtain a reference image from a database according to an image identifier, and sort the reference image using a compressed sorting method; then, according to the reference image, lossy decoding is performed on the lossy coded information, and then lossless decoding is performed on the lossless coded information.

[0042] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned method for joint compression of massive medical image cold data are implemented.

[0043] A non-transitory computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for joint compression of massive medical image cold data as described above.

[0044] Compared with the prior art, the present invention and its preferred solution can effectively utilize the similarity and correlation between massive data, reduce the transmission code stream, and improve the compression ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0046] Figure 1 is an algorithm flow chart of an embodiment of the present invention;

[0047] Figure 2 This is a bpp comparison diagram of an embodiment of the present invention;

[0048] Figure 3 A compression ratio comparison diagram of an embodiment of the present invention;

[0049] Figure 4 This is a comparison diagram of bpp savings according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the features and advantages of this patent more obvious and easy to understand, the following embodiments are specifically described in detail as follows:

[0051] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0052] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0053] like Figure 1 As shown, an embodiment of the present invention provides a method for joint compression of massive medical image cold data, and its implementation process includes: first, fine-grained feature extraction is performed on large-scale medical images through a feature extraction network, thereby establishing a feature database of medical images, and the extracted image features are stored by category; then, the feature extraction network is used on the medical image to be compressed to extract features and perform similarity matching; then, based on the image brightness, the slices of the reference image and the image to be compressed are sorted to optimize the compression order; finally, according to the reference image, the lossy encoding of the current image is assisted, and then according to the original image, the lossless encoding of the current image is assisted.

[0054] The design of the algorithm includes the following steps:

[0055] Step S1: Prediction structure design for massive medical images;

[0056] Step S2: prediction-based lossless medical image encoder design;

[0057] Step S3: Corresponding medical image decoder design.

[0058] Wherein, step S1 is specifically implemented as follows:

[0059] Step S11: Design a feature extraction network to perform fine-grained feature extraction on large-scale medical images. The formula is as follows:

[0060] F i =ε(X i ,θ)

[0061] Among them, ε represents the feature extractor (Encoder), which is a function or model responsible for converting the image into a feature representation, X i represents the input image, θ represents the parameter set of the feature extractor, F i Represents the extracted feature vector or feature map. In particular, the present invention adopts the I frame encoding technology of the video encoder to extract features. By using the features extracted by I frame encoding, a reference image similar to the image to be compressed can be more accurately retrieved, which helps to improve compression efficiency and image quality.

[0062] Step S12: Establish a feature database of medical images, store the extracted image features by category, and form an efficient and searchable feature library, the formula of which is as follows:

[0063] DB={(F i ,C i )|i=1,2,…N}

[0064] Where N represents the total amount of medical imaging data, F i Represents the feature vector or feature map of the i-th image, C i Indicates the corresponding classification label. The data in the feature library is classified by C i Store for later retrieval.

[0065] Step S13: Use the same feature extraction network to extract features of the compressed medical image for similarity matching. The formula is as follows:

[0066] F cur =ε(I cur ,θ)

[0067]

[0068] Among them, ε represents the feature extractor (Encoder), I cur represents the image to be compressed, θ represents the parameter set of the feature extractor, F cur Represents the feature vector or feature map extracted from the image to be compressed, F * is the feature vector with the smallest distance to the current image feature F, I ref is the corresponding image data, id refis the corresponding image identifier, dist(·) is the difference between the query feature F and the feature F in the database i A function of the distance between two entities (such as Euclidean distance, cosine similarity).

[0069] Step S2 is specifically implemented as follows:

[0070] Step S21: Based on a certain index, the slices of the reference image and the image to be compressed are sorted to optimize the compression order, and the formula is as follows:

[0071] i ref =Sort(i ref ,metric)

[0072] I cur =Sort(I cur ,metric)

[0073] Wherein, Sort(·) is a sorting algorithm, and metric is a sorting index. In particular, the average brightness of the image is used in the present invention.

[0074] Step S22: assisting the lossy encoding of the current image according to the reference image, the formula is as follows:

[0075] C lossy =E lossy (I ref ,I cur )

[0076]

[0077] Among them, E lossy Represents a lossy coding network, which converts the current image I cur and reference image I ref As input, output lossy coded information C lossy , D lossy Represents a lossy decoding network, which combines the encoded information and the reference image I ref As input, get a lossy reconstruction of the current image

[0078] Step S23: assisting lossless coding of the current image according to the original image, and the formula is as follows:

[0079]

[0080] Among them, D lossless Represents a lossless coding network, which converts the current image I cur and lossy reconstruction The error between the two images, and the current image I cur As input, output lossless coded information Clossless .

[0081] Step S3 is specifically implemented as follows:

[0082] Step S31: According to the id of step S13 ref Reference images are obtained from the database, and the reference images are sorted using the method of step S21.

[0083] Step S32: According to the reference image, lossy decoding is performed on the lossy coded information, and the formula is expressed as follows:

[0084]

[0085] Step S33: Lossless decoding is performed on the lossless coding information. The formula is as follows:

[0086]

[0087] Among them, D lossless represents a lossless decoding network, which will losslessly encode the information C lossless As input, obtain a lossless reconstruction of the current image

[0088] The overall scheme of the joint compression method of the present invention is obtained through the above design. The trained model is further used to test the three data sets with the commonly used traditional and deep neural network-based methods. The bpp (bits per pixel) required for data compression by different models is as follows: Figure 2 As shown in the figure, bpp refers to the number of effective bits occupied by each pixel. The smaller its value, the higher the image compression efficiency, that is, the fewer bits required for storage or transmission. The bpp required for the proposed method to compress the Axial, Coronal, and Sagittal datasets are 4.47, 3.59, and 4.81, respectively. Figure 3 The compression ratio is the bpp before compression divided by the bpp after compression. The larger the value, the higher the compression efficiency of the image. The original data is represented by 8 bits per pixel, that is, bpp = 8. The compression ratios of the proposed method on the Axial, Coronal, and Sagittal datasets are 1.76, 4.66, and 2.23, respectively. The percentage of bit rate saved by various methods compared with the FLIF (Free Lossless Image Format) method is shown in Figure 4As shown, a positive value indicates that the bpp after compression is less than FLIF, and a negative value indicates that the bpp after compression is more than FLIF. The larger the value, the higher the compression efficiency of the image. The bpp savings of the proposed method on the Axial, Coronal, and Sagittal datasets are 8.0%, 10.1%, and 6.9%, respectively.

[0089] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0090] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0091] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0093] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0094] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

[0095] This patent is not limited to the above-mentioned optimal implementation mode. Anyone can derive other forms of a joint compression method for massive medical image cold data under the inspiration of this patent. All equal changes and modifications made according to the scope of the patent application of the present invention should be covered by this patent.

Claims

1. A method for joint compression of massive medical image cold data, characterized by: Through the feature extraction network, fine-grained feature extraction is performed on large-scale medical images, thereby establishing a feature database of medical images, and the extracted image features are stored by category; Then, the feature extraction network is used on the medical image to be compressed to extract features and perform similarity matching; Then, based on the image brightness, the slices of the reference image and the image to be compressed are sorted to optimize the compression order; Finally, the reference image is used to assist the lossy encoding of the current image, and then the original image is used to assist the lossless encoding of the current image.

2. The method for joint compression of massive medical image cold data according to claim 1 is characterized in that: The features are extracted using a feature extraction network, defined as: F i =ε(X i ,i) Among them, ε represents the feature extractor, which is used to convert the image into feature representation, X i represents the input image, θ represents the parameter set of the feature extractor, F i Represents the extracted feature vector or feature map.

3. The method for joint compression of massive medical image cold data according to claim 1, characterized in that: Establish a feature database of medical images, and store the extracted image features by category as follows: DB={(F i ,C i )|i=1,2,…N} Where N represents the total amount of medical imaging data, F i Represents the feature vector or feature map of the i-th image, C i Indicates the corresponding classification label; the data in the feature library is classified by C i storage.

4. The method for joint compression of massive medical image cold data according to claim 1, characterized in that: The feature extraction network is used for the medical image to be compressed to extract features and perform similarity matching as follows: F cur =ε(I cur ,i) Among them, I cur represents the image to be compressed, F cur Represents the feature vector or feature map extracted from the image to be compressed, F * is the feature vector with the smallest distance to the current image feature F, I ref is the corresponding image data, id ref is the corresponding image identifier, dist(·) is the difference between the query feature F and the feature F in the database i function of the distance between them.

5. The method for joint compression of massive medical image cold data according to claim 1, characterized in that: The ordering of the slices of the reference image and the image to be compressed based on a certain index is expressed as: I ref =Sort(I ref ,metric) I cur =Sort(I cur ,metric) Among them, Sort(·) is a sorting algorithm, and metric is a sorting indicator, which is one of image brightness, organ area, and image similarity.

6. The method for joint compression of massive medical image cold data according to claim 1, characterized in that: The auxiliary lossy coding of the current image according to the reference image is specifically as follows: C lossy =E lossy (I ref ,I cur ) Among them, E lossy Represents a lossy coding network, which converts the current image I cur and reference image I ref As input, output lossy coded information C lossy , D lossy Represents a lossy decoding network, which combines the encoded information and the reference image I ref As input, get a lossy reconstruction of the current image The auxiliary lossless encoding of the current image according to the original image is specifically as follows: Among them, E lossless Represents a lossless coding network, which converts the current image I cur and lossy reconstruction The error between the two images, and the current image I cur As input, output lossless coded information C lossless .

7. The method for joint compression of massive medical image cold data according to claim 6, characterized in that: During decompression, the image identifier id ref Obtain reference images from a database, and sort the reference images using a compressed sorting method; Then, according to the reference image, the lossy coded information is losslessly decoded, and the formula is expressed as follows: The lossless coding information is then losslessly decoded, and the formula is expressed as follows: Among them, D lossless Represents a lossless decoding network, which converts the lossless coded information C lossless As input, obtain a lossless reconstruction of the current image 8. A massive medical image cold data joint compression system, characterized in that: include: Feature extraction module: used to perform fine-grained feature extraction on large-scale medical images through a feature extraction network, thereby establishing a feature database of medical images and storing the extracted image features by category; A similarity matching module, used for extracting features and performing similarity matching on the compressed medical image using the feature extraction network; A sorting module, used for sorting the slices of the reference image and the image to be compressed based on the image brightness to optimize the compression order; The compression module is used to assist the lossy encoding of the current image based on the reference image, and then assist the lossless encoding of the current image based on the original image.

9. The massive medical image cold data joint compression system according to claim 8, characterized in that: It also includes a decompression module: used to obtain a reference image from a database according to an image identifier, and sort the reference image using a compressed sorting method; then, according to the reference image, lossy decoding is performed on the lossy coded information, and then lossless decoding is performed on the lossless coded information.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of a method for joint compression of massive medical image cold data as described in any one of claims 1 to 7 are implemented.

Images