A Medical Image Privacy Protection Method and System Based on Encrypted Image Restoration

Abstract

This invention discloses a method and system for protecting medical image privacy based on encrypted image restoration. The method includes: generating a mask image, an image to be restored, a texture image, a JL transform encrypted result image, and an encrypted medical image from a medical image; dividing the encrypted medical image into blocks based on the texture image to obtain multiple image blocks; determining the restricted source region of the block to be restored in each image block based on the mask image and the texture image; determining the optimal patch set within the restricted source region based on the JL transform encrypted result image; connecting adjacent patches in the optimal patch set to obtain a connected optimal patch set; and generating an encrypted image with embedded information based on the connected optimal patch set and a binary bitstream. This invention solves the problems of low security in existing medical image privacy protection methods and the inability to apply existing encrypted image restoration techniques to image privacy protection due to poor restoration results.

Description

Technical Field

[0001] This invention relates to the field of medical information security protection technology, and in particular to a method and system for protecting medical image privacy based on encrypted image restoration. Background Technology

[0002] With the increasing number of privacy breaches, people's awareness of privacy protection is gradually increasing. However, many medical institutions currently store patient diagnostic X-ray or ultrasound images, etc., directly and without protection in their separate databases. While this method of storing patient images is convenient for management and retrieval, it actually poses security risks. Although many medical institutions currently store their medical data on their own internal local area networks, largely isolated from the external internet, this seems secure. However, with the development of medical informatization, many medical institutions will need to implement network connectivity for medical information sharing and remote medical diagnosis, adapting to the changes in medical informatization. The seemingly secure medical data is due to its closed nature. Once medical institutions are connected to the network, this previously secure medical information is exposed to the public internet. On the internet, the security of this unprotected medical data cannot be guaranteed, and hackers may attack and illegally steal this valuable medical information. Medical images, as one of the important types of medical data, necessitate the implementation of secure protection measures.

[0003] Currently, the main methods for protecting medical image privacy include image encryption algorithms and plaintext-based image restoration techniques. Most image encryption algorithms are not effective against chosen-plaintext attacks, meaning hackers can potentially crack the encrypted images, thus their security is not very good. Plaintext-based image restoration techniques require an additional database to store cropped lesion images, and then the cropped image is directly transmitted to the image processor. Although the cropped image does not contain lesion information, the image processor can still infer some symptoms from it, posing a risk of leaking patient privacy; therefore, this method is not very secure. Ciphertext image restoration techniques reconstruct the damaged areas (the areas to be restored) in the ciphertext image (the image encrypted by the image encryption algorithm) from intact areas. After decrypting the ciphertext image restoration result, the restored area achieves a visually plausible effect. It has the advantage of completing the damaged image restoration work without the image processor's knowledge of the image content information, preventing dishonest image processors and overcoming the security shortcomings of existing methods. Existing ciphertext image restoration methods can be divided into two categories: diffusion-based and patch-based. Among them, diffusion-based encrypted image inpainting algorithms achieve this by propagating information inward from the neighborhood of the area to be repaired along the iso-illuminance direction. However, for thick areas to be repaired, the reliability decreases as the information propagates deeper inward, resulting in poor repair effects. Patch-based encrypted image inpainting algorithms find patches with good similarity to the block to be repaired within the image to fill the block. However, since existing algorithms do not consider the coordination between patches, there are obvious visual conflicts between adjacent patches in the repair results, which cannot achieve the goal of making the image modification imperceptible. Therefore, current encrypted image inpainting techniques cannot be applied to protect the privacy of medical images. Summary of the Invention

[0004] The purpose of this invention is to provide a medical image privacy protection method and system based on encrypted image restoration, in order to solve the problems of low security of existing medical image privacy protection methods and the inability to apply existing encrypted image restoration technology to image privacy protection due to poor restoration effect.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for protecting medical image privacy based on encrypted image restoration, comprising:

[0007] The lesion region in the medical image is segmented by a trained image segmentation neural network model, and a mask image is generated.

[0008] Based on the mask image, the lesion area in the copy image is cropped to generate the image to be repaired; the copy image is a copy of the medical image;

[0009] The image to be repaired is convolved and clustered using a linear spatial filter and a K-means clustering algorithm to obtain a texture image;

[0010] JL transform encryption is performed on each local image block in the image to be repaired to obtain the JL transform encrypted result image; the local image block is composed of each pixel in the image to be repaired and s-1 pixels adjacent to each pixel;

[0011] The medical image is encrypted using an image encryption algorithm to obtain an encrypted medical image;

[0012] The encrypted medical image is divided into multiple image blocks based on the texture image;

[0013] The restricted source region of the block to be repaired in each image block is determined based on the mask image and the texture image; the block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image;

[0014] Based on the JL transform encryption result image, determine the optimal patch set within the restricted source region;

[0015] Connect adjacent patches in the optimal patch set to obtain the connected optimal patch set;

[0016] Based on the best set of patches after concatenation and the binary bit stream, an encrypted image with embedded information is generated.

[0017] Optionally, the step of dividing the encrypted medical image into multiple image blocks based on the texture image specifically includes:

[0018] The encrypted medical image is divided into blocks to obtain multiple initial image blocks;

[0019] The orientation marker is determined based on the longest dimension of each initial image patch;

[0020] Based on the texture image, calculate the texture distance between each sub-image block in each initial image block of different orientation markers;

[0021] The initial image blocks are divided into blocks based on the texture distance.

[0022] Optionally, the restricted source region of the block to be repaired includes: a restricted source region of a trusted block to be repaired and a restricted source region of an untrusted block to be repaired.

[0023] Optionally, determining the restricted source region of the trusted block to be repaired includes:

[0024] The texture distance between each trusted block to be repaired and the other trusted blocks in the encrypted medical image is compared; the trusted block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image and whose number accounts for less than half of the total number of pixels in the image block; the other trusted blocks consist of all pixels in the image block whose value is 0 at the corresponding position in the mask image and whose number accounts for more than half of the total number of pixels in the image block.

[0025] All intact pixels within a trusted block whose texture distance to the trusted block to be repaired is less than a given difference threshold are grouped into a restricted source region for the trusted block to be repaired; the intact pixels are pixels with a value of 0 at the corresponding position in the mask image.

[0026] Optionally, the restricted source region of the untrusted block to be repaired is a region composed of intact pixels within the untrusted block to be repaired, the restricted source region of the adjacent trusted block to be repaired, and intact pixels within the adjacent untrusted block to be repaired; the untrusted block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image and whose number accounts for more than half of the total number of pixels in the block.

[0027] Optionally, determining the optimal patch set within the restricted source region based on the JL transform encryption result image specifically includes:

[0028] Calculate the priority of all nodes within the confined source region;

[0029] Each node is accessed according to the priority, and the texture weight overhead between the local image patch of each node and all patches in the restricted source region of the image patch where each node is located is calculated based on the JL transform encrypted result image.

[0030] The optimal patch set is determined based on the texture weight overhead.

[0031] Optionally, the step of connecting adjacent patches in the optimal patch set to obtain a connected optimal patch set specifically includes:

[0032] Based on the texture image, determine the boundary of the minimum texture error within the overlapping area of ​​adjacent nodes;

[0033] By connecting adjacent patches according to the boundary, the optimal set of patches after connection is obtained.

[0034] Optionally, generating an encrypted image with embedded information based on the best patch set after concatenation and the binary bitstream specifically includes:

[0035] Encode the ciphertext values ​​and location information of all pixels to be repaired in the encrypted medical image into a binary bit stream;

[0036] The encrypted repair result image is obtained by replacing the local image patches of all nodes in the encrypted medical image with the optimal patch set after merging.

[0037] By using a reversible information hiding technique with separable ciphertext fields, the binary bit stream is reversibly embedded into the encrypted repair result image to obtain an encrypted image with embedded information.

[0038] This invention also provides a medical image privacy protection system based on encrypted image restoration, comprising:

[0039] The segmentation module is used to segment lesion regions in medical images using a trained image segmentation neural network model and generate mask images.

[0040] The image to be repaired generation module is used to cut out the lesion area in the copy image based on the mask image to generate the image to be repaired; the copy image is a copy of the medical image;

[0041] The texture image generation module is used to perform convolution and clustering on the image to be repaired using a linear spatial filter and a K-means clustering algorithm to obtain a texture image;

[0042] The JL transform encryption result image generation module is used to perform JL transform encryption on each local image block in the image to be repaired, and obtain the JL transform encryption result image; the local image block is composed of each pixel in the image to be repaired and s-1 pixels adjacent to each pixel;

[0043] An encrypted medical image generation module is used to encrypt the medical image using an image encryption algorithm to obtain an encrypted medical image;

[0044] The segmentation module is used to segment the encrypted medical image into multiple image blocks based on the texture image;

[0045] The restricted source region determination module is used to determine the restricted source region of the block to be repaired in each image block based on the mask image and the texture image; the block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image;

[0046] The optimal patch set determination module is used to determine the optimal patch set within the restricted source region based on the JL transform encrypted result image.

[0047] The connection module is used to connect adjacent patches in the optimal patch set to obtain the connected optimal patch set.

[0048] An encrypted image generation module with embedded information is used to generate encrypted images with embedded information based on the best patch set after concatenation and the binary bit stream.

[0049] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0050] This invention employs encrypted image restoration techniques, enabling image processors to restore damaged images without knowing the image content, thus enhancing the method's security. This invention reliably restores lesion areas in encrypted medical images and subtly hides the lesion information within the encrypted restoration result, providing a double layer of privacy protection for medical images. It solves the problems of insufficient security in existing medical image privacy protection methods and the inability to apply existing encrypted image restoration techniques to image privacy protection due to poor restoration results. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of a medical image privacy protection method based on encrypted image restoration according to an embodiment of the present invention;

[0053] Figure 2 This is a flowchart of the medical image processing of the present invention;

[0054] Figure 3 This is a flowchart illustrating encrypted image restoration according to an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The purpose of this invention is to provide a medical image privacy protection method and system based on encrypted image restoration, in order to solve the problems of low security of existing medical image privacy protection methods and the inability to apply existing encrypted image restoration technology to image privacy protection due to poor restoration effect.

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] like Figure 1-3 As shown, the present invention provides a medical image privacy protection method based on encrypted image restoration, comprising the following steps:

[0059] Step 101: Segment the lesion region in the medical image using the trained image segmentation neural network model and generate a mask image.

[0060] In practical applications, image database administrators use a trained image segmentation neural network to automatically segment the lesion region from medical images (e.g., a 240×240 slice image of a brain tumor) taken for patient diagnosis, generating a mask image (I). M (In a medical image, all pixels belonging to the lesion area have a value of 1 at their corresponding positions in the mask image, while the remaining pixels in the medical image have a value of 0 at their corresponding positions in the mask image).

[0061] Step 102: Based on the mask image, cut out the lesion area in the copy image to generate the image to be repaired; the copy image is a copy of the medical image.

[0062] In practical applications, the image database administrator copies an image that is exactly the same as medical image I, and then uses the masked image I from step 101 as a reference. M The lesion region in the copy image is cropped out (for each pixel in the copy image, if the value of the corresponding position of that pixel in the mask image is 1, then the pixel value of that pixel is changed to 4095 (assuming a color depth of 12 bits); otherwise, the pixel value of that pixel remains unchanged), generating the image to be repaired, I. R .

[0063] Step 103: Perform convolution and clustering on the image to be repaired using a linear spatial filter and K-means clustering algorithm to obtain a texture image.

[0064] In practical applications, image database administrators convolve the image to be repaired in step 102 using Gabor filters with 18 different orientations (6) and sizes (3). R Then, the K-means clustering algorithm (K=16) is applied to the output amplitude to record the image I to be repaired. R The clustering results of each pixel in the image yield the texture image I. W .

[0065] Step 104: Perform JL transform encryption on each local image block in the image to be repaired to obtain the JL transform encrypted result image; the local image block is composed of each pixel in the image to be repaired and s-1 pixels adjacent to each pixel.

[0066] In practical applications, image database administrators use the encryption key κ of the JL transform algorithm. JGenerate a random matrix of size 7*25. And a randomly generated 7-dimensional noise vector σ, used to encrypt the image I to be repaired in step 102 according to formula (1). R A local image block G is formed by each pixel and its s-1 neighboring pixels. For pixels on the image boundary, the missing pixels in the image block can be filled by the surrounding pixels.

[0067]

[0068] In the formula, G(i,j) represents the matrix composed of all pixel values ​​in the local image block consisting of the pixel in the i-th row and j-th column of the image to be repaired and its s-1 adjacent pixels.

[0069] Collect all corresponding encryption results to form the JL transform encryption result image E J .

[0070] Step 105: Encrypt the medical image using an image encryption algorithm to obtain an encrypted medical image.

[0071] In practical applications, image database administrators use image encryption algorithms that do not change the pixel positions of the image, such as the RSA algorithm, to encrypt medical images I. The image owner first selects two large prime numbers p1 and p2 to generate the encryption public key κ = e and the decryption private key. Where e is a large integer satisfying gcd(e, (p1-1)(p2-1)) = , n is any positive integer. Then, each pixel in the medical image I is encrypted using the public key κ = e to obtain the encrypted medical image E, as shown in the following formula:

[0072] E(i,j)=I(i,j) e mod(p l ·p2)

[0073] Finally, the encrypted medical image E and JL are transformed into the encrypted result image E. J Texture Image I W and mask image I M All of them are transmitted to the image processor.

[0074] Step 106: Divide the encrypted medical image into blocks based on the texture image to obtain multiple image blocks.

[0075] In practical applications, the image processor adaptively divides the encrypted medical image E into four approximately equal and non-overlapping image blocks, and assigns a direction label δ = v or h to each image block based on its longer dimension.

[0076] If the orientation marker of the image block is v, then calculate the two equal and non-overlapping sub-blocks B above and below the image block according to formulas (2) and (3). 1v and B 2v Texture distance between:

[0077]

[0078]

[0079] In the formula, The texture representing an image patch is the statistical representation of the ηth type of texture within image patch B; (i, j) represents the position index of the pixel in the i-th row and j-th column of the image. W (i, j) represents the value in the i-th row and j-th column of the texture image, η represents the texture type number, Φ represents the set of position indices of pixels with a pixel value of 0 in the mask image, B represents the set of position indices of all pixels in image patch B, |.| represents the function that returns the number of elements in the set, and ξ represents the index function, which returns 1 when the parameter is true and 0 otherwise.

[0080] When texture distance When the difference exceeds the given difference threshold τ = 0.15, the image block is divided into upper and lower sub-blocks; otherwise, the left and rear equal sub-blocks B are calculated according to formulas (2) and (3). 1h and B 2h Texture distance between them, when texture distance If the difference is greater than the given difference threshold τ, divide the image block into left and right sub-blocks; otherwise, leave the image block unchanged.

[0081] If the orientation of the image block is h, the texture distance between the left and right sub-blocks is calculated first and it is determined whether it is greater than the given difference threshold. If it is, the image block is divided into two equal left and right sub-blocks. Otherwise, the texture distance between the upper and lower sub-blocks is calculated and it is determined whether it is greater than the given difference threshold. If it is, the image block is divided into two equal upper and lower sub-blocks. Otherwise, the image block remains unchanged.

[0082] After an image block is divided into two sub-blocks, each sub-block is assigned a direction flag opposite to that of the parent block. This process of dividing all image blocks, including sub-blocks, within the encrypted medical image E is repeated until all image blocks are indivisible (the texture distance between the upper and lower sub-blocks of that image block is not greater than a given difference threshold, or the size of that image block is the given minimum size).

[0083] Step 107: Determine the restricted source region of the block to be repaired in each image block based on the mask image and the texture image; the block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image.

[0084] Image processors compare the texture distance between each trusted block to be repaired (a block containing pixels with a value of 1 at the corresponding position in the mask image, and the number of such pixels is less than half of the total number of pixels in the block) and other trusted blocks (a block containing more than half of the total number of pixels with a value of 0 at the corresponding position in the mask image) in an encrypted medical image. All intact pixels (pixels with a value of 0 at the corresponding position in the mask image) within all trusted blocks whose texture distance to the trusted block to be repaired is less than a given difference threshold are grouped into a region; such a region is called a restricted source. For each untrusted block to be repaired (the image block contains pixels with a value of 1 at the corresponding position in the mask image, and the number of such pixels accounts for more than half of the total number of pixels in the block), its restricted source region is the region composed of the intact pixels in the block, the restricted source region of the adjacent trusted blocks to be repaired, and the intact pixels in the adjacent untrusted blocks (the ratio of the number of intact pixels in the image block to the total number of pixels in the block is less than 0.5). At this time, all blocks to be repaired in the encrypted medical image (the image block contains pixels with a value of 1 at the corresponding position in the mask image) have their own restricted source regions.

[0085] In practical applications, image processors narrow down the patch search scope by first calculating each trusted patch to be repaired block in the encrypted medical image E according to formulas (2) and (3). With the remaining trusted block B rel Texture distance between Will be with this trusted block to be repaired Texture distance between All trusted blocks B less than a given difference threshold τ rel The intact pixels within constitute the confined source region of the block to be repaired. Then for each untrusted repair block Its restricted source region It is the region composed of intact pixels within the block, the restricted source region of the adjacent trusted blocks to be repaired, and intact pixels within the adjacent untrusted blocks. In the final encrypted medical image E, each block to be repaired has its own restricted source region.

[0086] Step 108: Based on the JL transform encryption result image, determine the optimal patch set within the restricted source region.

[0087] In practical applications, image processors divide encrypted medical images into several lattices with dimensions x pixels (the endpoints of the lattice fall on pixels). Some lattices contain pixels to be repaired (the corresponding position of this pixel in the mask image has a value of 1). Each endpoint of these lattices is called a node. The priority P(λ) of each node λ is calculated according to formulas (4), (5), and (6), and each node is visited one by one in descending order of priority P(λ).

[0088]

[0089]

[0090]

[0091] In the formula, λ represents a node in the encrypted medical image, q represents a patch with a length and width of 2x-1 pixels (s = (2x-1)²), and Λ λ τ represents the set of all patches within the restricted source region of the image patch containing node λ. R This represents a relative threshold, which is given. (.) + This represents a step function; it returns 1 if the parameter is greater than 0, and 0 otherwise. V represents the relative cost, which is the relative cost of patch q within the constrained source region of the image patch containing node λ. λ (q) represents the cost, which is the cost between a local image patch centered at node λ with dimensions of 2x-1 pixels and patch q. Let N(λ) be the patch with the minimum cost among all patches within the confined source region of the image patch containing node λ, and let N(λ) be the set of position indices centered at node λ, with dimensions of 2x-1 pixels. J (i, j) represents the value in the i-th row and j-th column of the JL transform encryption result image. The relative position of (i2, j2) in the patch is the same as the relative position of (i1, j1) in the local image block.

[0092] When a node is accessed, the texture weight overhead between the local image patch of the node and all patches within the restricted source region of the image patch containing the node is calculated according to formulas (6) and (7). Only retain the texture weight overhead between the local image patch and the node. The smallest g patches are selected, and the rest are discarded. At this time, the node has only g patches for candidates. Then, the overhead is propagated to the adjacent nodes according to formulas (8) and (9).

[0093]

[0094]

[0095]

[0096] In the formula, B q B represents the image patch containing the center pixel of patch q. λ Let λ represent the image patch containing node λ. Only the g patches with the minimum texture weight cost among the local image patches of this node are retained, and the rest are discarded. At this point, this node has only g patches as candidates. This represents the node adjacent to the higher-priority node λ. Representative node set One of the patches, where t represents the update step. Representing paired potential, it is patch q and patch Assume that nodes λ and λ are covered respectively. Local image patches are overlapping regions within patch q and patches Overhead between overlapping regions This represents when patch q and patch Assume that nodes λ and λ are covered respectively. A local image patch is a set of pixel position indices for the overlapping regions within patch q. This represents when patch q and patch Assume that nodes λ and λ are covered respectively. Local image patch The set of pixel position indices (i2, j2) in the inner overlapping region in the patch The relative position of the inner overlapping region is consistent with the relative position of (i1, j1) in the inner overlapping region of patch q.

[0097] When all nodes have been visited, each node has only g candidate patches. The patch combination that minimizes the Markov energy is selected, leaving each node with only one patch for repair. The Markov energy can be represented as:

[0098]

[0099] Step 109: Connect adjacent patches in the optimal patch set to obtain the connected optimal patch set.

[0100] In practical applications, when a local image patch of each node is covered by a patch, there is an overlapping area between patches of adjacent nodes. In this case, the image processor needs to find the boundary of the minimum texture error S within the overlapping area to connect the patches of two adjacent nodes. The texture error of two pixels within the overlapping area of ​​two adjacent nodes can be expressed as...

[0101]

[0102] In the formula, I W (i1, j1) represents the value in the i1th row and j1st column of the texture image.

[0103] If the nodes are adjacent on the left and right, the cumulative texture error within the overlapping area can be expressed as:

[0104]

[0105] In the formula, (i, j) represents the index of the relative position within the overlapping region. If the nodes are adjacent vertically, the texture error is accumulated column by column within the overlapping region. The boundary that minimizes the total texture error is found within the overlapping region, and then two adjacent patches are connected according to the boundary.

[0106] Step 1010: Based on the best patch set after concatenation and the binary bit stream, generate an encrypted image with embedded information.

[0107] In practical applications, after all adjacent nodes have been patched, the image processor encodes the ciphertext values ​​and location information of all pixels to be repaired in the encrypted medical image E into a binary bit stream b. Then, the patched data from the previous step is used to replace the local image blocks of all nodes in the encrypted medical image E, resulting in an encrypted repair result E'. The binary bit stream b is then reversibly embedded into the encrypted repair result E' using the ciphertext field separable reversible information hiding technique based on the embedding key κ1, resulting in an encrypted image E' with embedded information, which is then transmitted to the image database administrator.

[0108] The image database administrator can store the received encrypted image E” in their own database. When a doctor or patient needs to view the original image I, the image database administrator transmits the encrypted image E” stored in the database to the doctor or patient. The doctor or patient can extract the key κ2 based on the information, extract the binary bit stream b, decode the binary bit stream to obtain the ciphertext value and position information, and then use these ciphertext values ​​to replace the ciphertext values ​​at the corresponding positions in the encrypted image E”. Finally, the decryption private key is used. Decrypting all the ciphertext in the encrypted image "E" yields the original image "I", as shown in the following formula:

[0109] I(i,j)=E"(i,j) d mod(p l ·p2))

[0110] The present invention has the following advantages:

[0111] 1. This invention employs encrypted image restoration techniques, thereby achieving the technical effect of restoring damaged images without the image processor having any knowledge of the image content information, thus improving the security of the method.

[0112] 2. This invention employs a technique for ciphertext image restoration based on a Markov model, thereby achieving a better technical effect in ciphertext image restoration.

[0113] 3. This invention employs a combination of encrypted image restoration and reversible information hiding in the encrypted domain, thereby achieving a dual-layer security protection for medical image privacy.

[0114] This invention also provides a medical image privacy protection system based on encrypted image restoration, comprising:

[0115] The segmentation module is used to segment lesion regions in medical images using a trained image segmentation neural network model and generate mask images.

[0116] The image to be repaired generation module is used to cut out the lesion area in the copy image based on the mask image to generate the image to be repaired; the copy image is a copy of the medical image;

[0117] The texture image generation module is used to perform convolution and clustering on the image to be repaired using a linear spatial filter and a K-means clustering algorithm to obtain a texture image;

[0118] The JL transform encryption result image generation module is used to perform JL transform encryption on each local image block in the image to be repaired, and obtain the JL transform encryption result image; the local image block is composed of each pixel in the image to be repaired and s-1 pixels adjacent to each pixel;

[0119] An encrypted medical image generation module is used to encrypt the medical image using an image encryption algorithm to obtain an encrypted medical image;

[0120] The segmentation module is used to segment the encrypted medical image into multiple image blocks based on the texture image;

[0121] The restricted source region determination module is used to determine the restricted source region of the block to be repaired in each image block based on the mask image and the texture image; the block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image;

[0122] The optimal patch set determination module is used to determine the optimal patch set within the restricted source region based on the JL transform encrypted result image.

[0123] The connection module is used to connect adjacent patches in the optimal patch set to obtain the connected optimal patch set.

[0124] An encrypted image generation module with embedded information is used to generate encrypted images with embedded information based on the best patch set after concatenation and the binary bit stream.

[0125] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0126] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for protecting the privacy of medical images based on encrypted image restoration, characterized in that, include: The lesion region in the medical image is segmented by a trained image segmentation neural network model, and a mask image is generated. Based on the mask image, the lesion area in the copy image is cropped to generate the image to be repaired; the copy image is a copy of the medical image; The image to be repaired is convolved and clustered using a linear spatial filter and a K-means clustering algorithm to obtain a texture image; JL transform encryption is performed on each local image block in the image to be repaired to obtain the JL transform encrypted result image; the local image block is composed of each pixel in the image to be repaired and s-1 pixels adjacent to each pixel; The medical image is encrypted using an image encryption algorithm to obtain an encrypted medical image; The encrypted medical image is divided into multiple image blocks based on the texture image; The restricted source region of the block to be repaired in each image block is determined based on the mask image and the texture image; the block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image; Based on the JL transform encryption result image, determine the optimal patch set within the restricted source region; Connect adjacent patches in the optimal patch set to obtain the connected optimal patch set; Based on the best set of patches after concatenation and the binary bit stream, an encrypted image with embedded information is generated.

2. The medical image privacy protection method based on encrypted image restoration according to claim 1, characterized in that, The step of dividing the encrypted medical image into multiple image blocks based on the texture image specifically includes: The encrypted medical image is divided into blocks to obtain multiple initial image blocks; The orientation marker is determined based on the longest dimension of each initial image patch; Based on the texture image, calculate the texture distance between each sub-image block in each initial image block of different orientation markers; The initial image blocks are divided into blocks based on the texture distance.

3. The medical image privacy protection method based on encrypted image restoration according to claim 1, characterized in that, The restricted source regions of the block to be repaired include: restricted source regions of trusted blocks to be repaired and restricted source regions of untrusted blocks to be repaired.

4. The medical image privacy protection method based on encrypted image restoration according to claim 3, characterized in that, The determination of the restricted source region of the trusted block to be repaired includes: The texture distance between each trusted block to be repaired and the other trusted blocks in the encrypted medical image is compared; the trusted block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image and whose number accounts for less than half of the total number of pixels in the image block; the other trusted blocks consist of all pixels in the image block whose value is 0 at the corresponding position in the mask image and whose number accounts for more than half of the total number of pixels in the image block. All intact pixels within a trusted block whose texture distance to the trusted block to be repaired is less than a given difference threshold are grouped into a restricted source region for the trusted block to be repaired; the intact pixels are pixels with a value of 0 at the corresponding position in the mask image.

5. The medical image privacy protection method based on encrypted image restoration according to claim 4, characterized in that, The restricted source region of the untrusted block to be repaired is the region composed of intact pixels within the untrusted block to be repaired, the restricted source region of the adjacent trusted block to be repaired, and intact pixels within the adjacent untrusted block to be repaired; the untrusted block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image and whose number accounts for more than half of the total number of pixels in the block.

6. The medical image privacy protection method based on encrypted image restoration according to claim 1, characterized in that, The step of determining the optimal patch set within the restricted source region based on the JL transform encryption result image specifically includes: Calculate the priority of all nodes within the confined source region; Each node is accessed according to the priority, and the texture weight overhead between the local image patch of each node and all patches in the restricted source region of the image patch where each node is located is calculated based on the JL transform encrypted result image. The optimal patch set is determined based on the texture weight overhead.

7. The medical image privacy protection method based on encrypted image restoration according to claim 6, characterized in that, The step of connecting adjacent patches in the optimal patch set to obtain the connected optimal patch set specifically includes: Based on the texture image, determine the boundary of the minimum texture error within the overlapping area of ​​adjacent nodes; By connecting adjacent patches according to the boundary, the optimal set of patches after connection is obtained.

8. The medical image privacy protection method based on encrypted image restoration according to claim 1, characterized in that, The process of generating an encrypted image with embedded information based on the best patch set after concatenation and the binary bitstream specifically includes: Encode the ciphertext values ​​and location information of all pixels to be repaired in the encrypted medical image into a binary bit stream; The encrypted repair result image is obtained by replacing the local image patches of all nodes in the encrypted medical image with the optimal patch set after merging. By using a reversible information hiding technique with separable ciphertext fields, the binary bit stream is reversibly embedded into the encrypted repair result image to obtain an encrypted image with embedded information.

9. A medical image privacy protection system based on encrypted image restoration, characterized in that, include: The segmentation module is used to segment lesion regions in medical images using a trained image segmentation neural network model and generate mask images. The image to be repaired generation module is used to cut out the lesion area in the copy image based on the mask image to generate the image to be repaired; the copy image is a copy of the medical image; The texture image generation module is used to perform convolution and clustering on the image to be repaired using a linear spatial filter and a K-means clustering algorithm to obtain a texture image; The JL transform encryption result image generation module is used to perform JL transform encryption on each local image block in the image to be repaired, and obtain the JL transform encryption result image; the local image block is composed of each pixel in the image to be repaired and s-1 pixels adjacent to each pixel; An encrypted medical image generation module is used to encrypt the medical image using an image encryption algorithm to obtain an encrypted medical image; The segmentation module is used to segment the encrypted medical image into multiple image blocks based on the texture image; The restricted source region determination module is used to determine the restricted source region of the block to be repaired in each image block based on the mask image and the texture image; the block to be repaired consists of all pixels in the image block whose value is 1 at the corresponding position in the mask image; The optimal patch set determination module is used to determine the optimal patch set within the restricted source region based on the JL transform encrypted result image. The connection module is used to connect adjacent patches in the optimal patch set to obtain the connected optimal patch set. An encrypted image generation module with embedded information is used to generate encrypted images with embedded information based on the best patch set after concatenation and the binary bit stream.

Images