Medical image secure transmission method based on multi-modal biological characteristics

Abstract

The invention relates to a medical image secure transmission method based on multi-modal biological characteristics, and belongs to the technical field of image security protection. According to the method, firstly, a medical image is segmented into lesion pixels and non-lesion pixels through a deep neural network, and then the lesion pixels are embedded into the non-lesion pixels by adopting a timestamp and least significant bit replacement method. Further, according to the method, the marked non-lesion part and the lesion part are combined to generate a marked image. Secondly, multi-modal biological features of a user are extracted through a deep neural network, and a user-defined secret key and features obtained after device MAC address coding extraction are adopted as secret keys; and finally, the marked image is encrypted by using the coded secret key and the chaotic system, so that the safety protection level of the medical image data is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of image security protection and relates to a medical image security transmission method based on multimodal biometrics. Background Art

[0002] In recent years, IoT technology has attracted much attention due to its excellent performance and has been widely used in the medical field. IoT connects various consumer electronic devices, sensors, networks, etc. to form an efficient data sharing system. In the digital medical system, the medical IoT is widely used to transmit medical images and related records for disease diagnosis and treatment. However, in the medical field, the use of digital medical images also brings the risk of data leakage, copyright infringement and identity theft, which seriously affects the lives of patients. Therefore, as an important information carrier, the security of medical images is very important for the realization of smart medical care. Encryption is an effective technology for protecting medical images, which converts medical images into ciphertext images by using public or private keys. However, the decrypted image information may be used without authorization, and image encryption alone cannot provide high security. Data hiding provides an additional layer of security to prove image ownership and protect them from unauthorized access and use. Data hiding is an important technology for copyright protection, which hides a type of information in digital media without being noticed. When the data with hidden tags is misused, the hidden tags can be extracted to verify ownership. In addition to encryption and data hiding, biometrics is widely used in many fields to identify a person based on his physiological or behavioral characteristics due to its excellent security performance. It is well known that unimodal biometrics have limited accuracy, and multimodal biometric systems have shown excellent performance in improving matching accuracy and solving the problems of non-universality and noisy data caused by obtaining information from multiple biometric features.

[0003] Based on the above motivations, the present invention proposes a medical image security transmission method based on multimodal biometrics to ensure the secure transmission of medical images and prevent data abuse. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method for secure transmission of medical images based on multimodal biometrics, which comprehensively utilizes multimodal biometric recognition technology, image segmentation technology, data hiding technology and image encryption technology, which not only ensures the security of medical images during transmission, but also effectively prevents the risk of data abuse, and builds a solid protection barrier for patients' personal privacy rights.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for secure transmission of medical images based on multimodal biometrics first segments the medical image into lesion pixels and non-lesion pixels through a deep neural network, and then embeds the lesion pixels into the non-lesion pixels using a timestamp and least significant bit replacement method. Furthermore, the method combines the marked non-lesion part with the lesion part to generate a marked image. Secondly, the user's multimodal biometrics are extracted through a deep neural network, and the extracted features are encoded using a user-defined key and the device MAC address as a key. Finally, the marked image is encrypted using the encoded key and a chaotic system.

[0007] The method specifically comprises the following steps:

[0008] S1: Feature extraction: construct multiple feature extraction network models to extract multimodal biometric features respectively; the multimodal biometric features include but are not limited to fingerprints, facial features, iris features and EEG signal features;

[0009] S2: Image segmentation: Build an image segmentation network model to segment medical images to obtain images of lesion areas and non-lesion areas;

[0010] S3: Pseudo-random sequence generation: construct a chaotic system as a pseudo-random sequence generator to obtain a pseudo-random sequence;

[0011] S4: Data hiding: obtaining a timestamp, obtaining a random number sequence according to the timestamp and a pseudo-random sequence generator, using the sequence to select appropriate pixels in the non-lesion image, and hiding the lesion image pixels into the selected non-lesion image pixels using the least significant bit (LSB) replacement method;

[0012] S5: Data extraction: combining the non-lesion image and the hidden data obtained in step S4 to obtain a secret image, extracting the hidden lesion image pixels using the same timestamp and pseudo-random sequence, and combining it with the non-lesion area image to obtain a complete medical image;

[0013] S6: Data encryption: The final key is generated using the extracted multimodal biometrics, user key and system MAC address, and the image is scrambled and pixel diffused using the final key and pseudo-random sequence to achieve data encryption;

[0014] S7: Data decryption: The encrypted image is subjected to the reverse process of scrambling and pixel diffusion using the final key and pseudo-random sequence to restore the original medical image.

[0015] Further, in step S1, the feature extraction specifically includes the following steps:

[0016] S11: Constructing feature extraction network model architecture based on deep learning technology;

[0017] S12: Use biological feature datasets to train multiple feature extraction networks;

[0018] S13: Utilize the trained multiple feature extraction networks to extract multimodal biological features respectively.

[0019] Further, in step S2, the image segmentation specifically comprises the following steps:

[0020] S21: Construct image segmentation network model architecture based on deep learning technology;

[0021] S22: Use image segmentation dataset to train image segmentation network;

[0022] S23: Segment the medical image using the trained image segmentation network to obtain images of the lesion area and images of the non-lesion area.

[0023] Furthermore, in step S3, the pseudo-random sequence generation specifically includes: firstly constructing a chaotic system; then selecting a suitable system initial value, and continuously iterating to generate a pseudo-random sequence.

[0024] Further, in step S6, the data encryption specifically includes the following steps:

[0025] S61: Obtain the multimodal biometrics according to step S1, obtain the system MAC address, and calculate the multimodal biometrics, the user key and the MAC address to obtain the final key;

[0026] S62: Obtain a pseudo-random sequence using the final key and the pseudo-random sequence generation process in step S3;

[0027] S63: converting the image into a plurality of N×N image blocks, and scrambling the image blocks using the acquired pseudo-random sequence;

[0028] S64: After scrambling the image block, further scramble the position of each pixel in the block using the acquired pseudo-random sequence;

[0029] S65: performing pixel diffusion on the scrambled image;

[0030] S66: Repeat the image scrambling and pixel diffusion process for several times.

[0031] Further, in step S7, the data decryption specifically includes the following steps:

[0032] S71: Obtain a key according to the key generation method in the encryption process of step S6;

[0033] S72: Obtain a pseudo-random sequence using the final key and the pseudo-random sequence generation process in step S3;

[0034] S73: using the acquired pseudo-random sequence, performing a reverse process of pixel diffusion on the encrypted image;

[0035] S74: performing the inverse process of pixel scrambling within the block on the position of each pixel in the image block;

[0036] S75: using the acquired pseudo-random sequence to perform the reverse process of image block scrambling;

[0037] S76: Repeat the image scrambling inverse process and the pixel diffusion inverse process for several times.

[0038] The beneficial effects of the present invention are as follows: the present invention provides a new type of medical image security transmission technology. In response to the security challenges faced by medical image transmission in open networks, the present invention utilizes multimodal biometrics to ensure the confidentiality and integrity of medical images during transmission through processes such as biological feature extraction, image segmentation, data hiding, and image encryption.

[0039] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0041] Figure 1 The overall architecture diagram of the medical image security transmission method based on multimodal biometrics of the present invention;

[0042] Figure 2 It is a schematic diagram of the flow chart of the sending end in the method of the present invention;

[0043] Figure 3 It is a flow chart of the receiving end in the method of the present invention. DETAILED DESCRIPTION

[0044] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0045] See also Figure 1 to Figure 3 ,The present invention provides a method for secure transmission of medical images based on multimodal biometrics, such as Figure 1 As shown, the method specifically comprises the following steps:

[0046] Step 1: Feature extraction: Use a deep learning-based architecture to build multiple feature extraction network models to extract multiple modal biometric features respectively.

[0047] Step 1 specifically includes the following steps:

[0048] Step 11: Build a feature extraction network model architecture based on deep learning technology;

[0049] Among them, firstly, a deep learning framework is selected, including but not limited to using TensorFlow or PyTorch framework. The network model structure used is any architecture that can realize the feature extraction function, such as convolutional neural network, autoencoder, residual network and Transformer network structures. The present invention does not limit the deep learning framework and feature extraction network architecture used.

[0050] Step 12: Using biological feature datasets to train multiple feature extraction networks;

[0051] The data set used is any data set that can realize the training of the corresponding feature extraction network. Taking iris feature extraction as an example, CASIA-IrisV3, UBIRIS, BioSecure or ND-Iris-0405 data sets can be used. The present invention does not limit the model training data set used. The method of training the feature extraction network is: divide the data set into training samples and test samples, use the training samples to train the model, use the test samples to evaluate the model, and when the model evaluation result is greater than a certain expected threshold, stop training and save the trained model file.

[0052] Step 13: Utilize the trained multiple feature extraction networks to extract multimodal biological features respectively;

[0053] The method of obtaining multimodal biological features is: reading a trained feature extraction model file, inputting a variety of biological data into a feature extraction network model, and obtaining multimodal biological features through calculation. The extraction process of multimodal biological features can be obtained by extracting different types of biological data in sequence, or by inputting multimodal biological data into a feature extraction network to obtain them at one time, or by a combination of the two. The present invention does not limit the extraction process of multimodal biological features.

[0054] Step 2: Image Segmentation: Build an image segmentation network model to segment medical images.

[0055] Step 2 specifically includes the following steps:

[0056] Step 21: Build an image segmentation network model architecture based on deep learning technology;

[0057] Among them, first select a deep learning framework for image segmentation tasks, including but not limited to using TensorFlow or PyTorch framework. Then design and build the image segmentation network model architecture, and the network model structure used is any architecture that can realize the image segmentation function, such as U-Net, Mask R-CNN or DeepLab network structure. The design of the network architecture should take into account the characteristics of medical images, such as resolution, contrast and complexity of the lesion area, to ensure that the model can capture the subtle features in the image. The present invention does not limit the deep learning framework and image segmentation network structure used.

[0058] Step 22: Use the image segmentation dataset to train the image segmentation network;

[0059] After the network model architecture is built, the image segmentation dataset is used for training. The dataset used is any dataset that can realize image segmentation network training. The dataset should contain a large number of medical images and their corresponding masks (Mask), where the mask is used to indicate the lesion area in the image. During the training process, the medical image is input into the network model, and the prediction result (i.e., segmentation map) is obtained through the forward propagation calculation of the network. The dataset is divided into training samples and test samples. The training samples are used to train the model, and the test samples are used to evaluate the model. The prediction result is compared with the true mask, the loss function (such as cross entropy loss, Dice loss, etc.) is calculated, and the network parameters are updated through the back propagation algorithm. When the model evaluation result is greater than a certain expected threshold, the training is stopped and the trained model file is saved.

[0060] Step 23: Segment the medical image using the trained image segmentation network to obtain images of the lesion area and images of the non-lesion area;

[0061] The method of obtaining the lesion area image and the non-lesion area image is: read the trained segmentation model file, input the medical image to be segmented into the segmentation network model, and obtain the segmentation result through the network forward propagation calculation. According to the result, the pixels in the medical image can be divided into two categories: lesion area and non-lesion area, and the lesion area image and non-lesion area image can be extracted respectively.

[0062] Step 3: Pseudo-random sequence generation: Construct a chaotic model as a pseudo-random sequence generator to obtain a pseudo-random sequence.

[0063] Step 3 specifically includes the following steps:

[0064] Step 31: Construct a chaotic system;

[0065] In the process of constructing a chaotic system as a pseudo-random sequence generator, a classic chaotic map is selected as an example of implementation. Taking the Logistic map as an example, it is a simple and widely used chaotic system, and its mathematical expression is:

[0066] x n+1 =rx n (1-x n )

[0067] Among them, x n represents the value of the nth iteration point, r is the control parameter of the chaotic system, and its value range is usually between (0,4]. When r is between about 3.57 and 4, the Logistic map shows obvious chaotic behavior, which means that the output sequence of the system is highly sensitive to the initial value and the sequence shows quasi-randomness. The present invention can also use other chaotic maps, such as Henon map, Lorenz system or Chen system, etc. The embodiment of the present invention does not limit the random sequence generation method used.

[0068] Step 32: Select appropriate system initial values ​​and iterate to generate pseudo-random sequences;

[0069] Select a floating point number in the interval (0,1) as the initial value x of the Logistic map 0 The choice of initial value is crucial to the unpredictability of the generated pseudo-random sequence. In order to ensure the diversity and non-repeatability of the sequence, high-precision floating-point numbers are usually used, and values ​​that are easy to guess or have special mathematical properties are avoided. Using the formula of the Logistic mapping, from the initial value x 0 Start iterating. Each iteration will be the current value x n Substitute into the formula to calculate the next value x n+ 1 Through continuous iteration, an infinitely long sequence of values ​​can be generated. Optionally, in order to meet the requirements of certain application scenarios for integers or values ​​within a specific range, the generated sequence can be further processed to meet specific pseudo-randomness requirements.

[0070] Step 4: Data hiding: A random number sequence is obtained using a timestamp and a pseudo-random sequence generator, and the sequence is used to select pixels in the non-lesion image that are suitable for embedding. The least significant bit (LSB) is used to hide the pixels of the lesion image into the non-lesion image.

[0071] Step 4 specifically includes the following steps:

[0072] Step 41: Obtain a timestamp, and obtain a pseudo-random sequence according to the timestamp and the pseudo-random sequence generation step;

[0073] First, the system obtains the current timestamp through the internal clock or external time synchronization device. The accuracy of the timestamp can be set as needed, for example, accurate to seconds, milliseconds or even microseconds, to ensure that a unique time starting point can be obtained each time it is executed. Then, using the timestamp as a seed, a pseudo-random sequence is generated through the above pseudo-random sequence generation process. The length of the sequence should be sufficient to cover all pixels of the lesion image to be embedded, or at least sufficient to select the embedding position in the non-lesion image. The generation of the pseudo-random sequence ensures that each embedding process is unpredictable, which enhances the security of the data.

[0074] Step 42: Select appropriate non-lesion area pixels using the generated sequence;

[0075] After obtaining the pseudo-random sequence, the system selects the pixel positions in the non-lesion image according to the values ​​in the sequence (or the values ​​after some mapping). These positions can be the corresponding pixel indexes when traversing the image according to some rules, or they can be directly mapped to the two-dimensional coordinates of the image. In order to ensure the quality of the image after embedding, important feature areas in the image should be avoided when selecting non-lesion areas to reduce the impact of the embedding process on the visual quality of the image. In addition, the selection strategy can be dynamically adjusted according to the local complexity of the image to further optimize the embedding effect.

[0076] Step 43: Embed the image pixels of the lesion area into the selected non-lesion area pixels by using the LSB replacement method;

[0077] After determining the embedding position, start embedding the pixel information of the lesion image into the corresponding position of the non-lesion image. The present invention adopts the least significant bit replacement method, that is, only the least significant bit of the pixel value of the non-lesion image is modified to store the pixel information of the lesion image. Specifically, for each lesion image pixel to be embedded, the system first converts its pixel value into binary form and extracts the least significant bit (or bits) as the information to be embedded. Then, find the corresponding non-lesion image pixel position and replace the least significant bit of its pixel value with the corresponding information of the lesion image. In addition, in order to accurately extract the embedded lesion image information later, it is also necessary to record the necessary metadata (such as embedding position, LSB bit number, etc.) during the embedding process and use it when extracting.

[0078] Step 5: Data extraction: Combine the non-lesion image and the lesion area image to obtain the secret image, and use the same timestamp to extract the hidden data.

[0079] Step 5 specifically includes the following steps:

[0080] Step 51: Obtain a timestamp, and obtain a pseudo-random sequence according to the timestamp and the pseudo-random sequence generation step;

[0081] First, in order to keep consistent with the data hiding process, the system obtains the same timestamp as when hiding the data through the internal clock or external time synchronization device to ensure that the same pseudo-random sequence can be generated when extracting data as when hiding the data, so as to accurately locate the embedded pixel position. Then, using the obtained timestamp as a seed, a pseudo-random sequence is generated through the same pseudo-random sequence generation method. The length and structure of this sequence are exactly the same as the sequence used when hiding the data to ensure that it can correspond one-to-one to the embedded pixel position.

[0082] Step 52: using the generated sequence to determine the pixels embedded in the non-lesion area;

[0083] After obtaining the pseudo-random sequence, the non-lesion area positions of the lesion image pixels embedded in the secret image are determined according to the values ​​in the sequence (or the values ​​after some mapping). These positions should be exactly the same as the positions selected during data hiding to ensure that the hidden lesion image information can be accurately extracted. If complex mapping strategies or dynamic adjustment selection strategies are used during embedding, these strategies also need to be applied accordingly during extraction.

[0084] Step 53: extracting the lesion area image pixels from the embedded non-lesion area pixels using the inverse process of the LSB replacement method;

[0085] After determining the embedding position, the system uses the inverse process of the LSB replacement method to extract the lesion image pixels. Specifically, for each embedded pixel position, the system first reads the pixel value of the corresponding pixel in the secret image and converts it into binary form. Then, the system extracts the least significant bit (or bits, depending on the number of bits used when embedding) of the pixel value, which is the hidden lesion image pixel information. By recombining these bits into complete pixel values, the system can gradually construct a hidden lesion area image.

[0086] Step 6: Data encryption: After data embedding, the labeled image is encrypted through the encoded biometrics and chaotic mapping.

[0087] Step 6 specifically includes the following steps:

[0088] Step 61: Obtain the final key;

[0089] First, the user's multimodal biometric data, including but not limited to fingerprints, facial images, iris images, etc., is obtained through relevant equipment, and multimodal biometric features are obtained through the above-mentioned multimodal biometric feature extraction process. Then, the MAC address of the system is obtained. The MAC address is the unique physical address of the network device, which is used to ensure the uniqueness and security of key generation. Then, the multimodal features, user keys and MAC addresses are calculated to ensure the complexity and security of the final key. Finally, a final key of a fixed length is obtained for the subsequent data encryption process.

[0090] Step 62: Obtain a pseudo-random sequence;

[0091] The final key generated in step 61 is used in combination with the above pseudo-random sequence generation process to obtain a pseudo-random sequence. The chaotic mapping algorithm is highly complex and unpredictable and can generate high-quality pseudo-random numbers, thereby ensuring the randomness and security of data encryption.

[0092] Step 63: scrambling the image blocks;

[0093] The marked image to be encrypted is converted into a number of N×N image blocks. The size of N can be adjusted according to actual needs to balance the encryption effect and computational complexity. Then, the image blocks are scrambled using the pseudo-random sequence generated in step 602. The scrambling process includes random exchange, rotation, flipping and other operations of the image blocks to ensure that the arrangement order of the image blocks is random and unpredictable.

[0094] Step 64: scrambling pixel positions;

[0095] After the image blocks are scrambled, the pixel positions in each image block are further scrambled using a pseudo-random sequence. This can be achieved by randomly swapping and shifting pixels to increase the difficulty and complexity of data extraction.

[0096] Step 65: Pixel diffusion;

[0097] After the pixels in the image block are scrambled, the pixels at the corresponding positions in the image are selected using a pseudo-random sequence, and the pixel diffusion is achieved by performing some operations (such as addition, XOR operation, etc.) on the values ​​of the selected pixels. Pixel diffusion can enhance the image's anti-attack ability, so that even if part of the image is cracked, the complete original image cannot be directly restored.

[0098] Step 66: Repeat the image scrambling and pixel diffusion process;

[0099] In order to improve the strength and security of encryption, the image scrambling and pixel diffusion process in steps 63 to 65 is repeated several times. Each time it is repeated, a pseudo-random sequence can be regenerated or the parameters of scrambling and diffusion can be adjusted to increase the complexity and unpredictability of encryption.

[0100] Step 7: Data decryption: After obtaining the ciphertext image, the ciphertext image is decrypted through the encoded biometrics and chaotic mapping.

[0101] Step 7 specifically includes the following steps:

[0102] Step 71: Obtain the key according to the steps in the above encryption process;

[0103] Similar to the steps of the encryption process, before decrypting the image, you need to obtain the key. First, obtain the multimodal biometric data used in encryption through relevant equipment, including but not limited to fingerprints, facial images, iris images, etc., and obtain the multimodal biometric features used in encryption through the above-mentioned multimodal biometric feature extraction process. Then, obtain the MAC address of the system. Then, the multimodal features, user keys and MAC addresses are calculated to ensure that the same key as the encryption key is used for the subsequent data decryption process.

[0104] Step 72: Obtain a pseudo-random sequence using the final key and the pseudo-random sequence generation process described above;

[0105] After obtaining the final decryption key, the key can be used to generate the same pseudo-random sequence as that used during encryption according to the pseudo-random sequence generation method defined during the encryption process. This step is a critical link in the decryption process because the pseudo-random sequence will be directly used to guide the subsequent pixel diffusion and image scrambling reverse process.

[0106] Step 73: using the acquired pseudo-random sequence, perform the reverse process of pixel diffusion on the encrypted image;

[0107] In the decryption stage, it is necessary to perform the reverse process of pixel diffusion on the encrypted image using a pseudo-random sequence in the opposite order and rules to those used during encryption, in order to restore the original value of each pixel in the original image.

[0108] Step 74: performing the inverse process of pixel scrambling within the block on the position of each pixel in the image block;

[0109] During the encryption process, the image is divided into several image blocks, and the pixel positions in each image block are scrambled. Therefore, in the decryption stage, the pixel positions in each image block need to be reversely scrambled according to the opposite rules of encryption to restore the original arrangement order of the pixels.

[0110] Step 75: using the acquired pseudo-random sequence to perform the reverse process of image block scrambling;

[0111] In addition to the scrambling of pixels within the block, the image block itself is scrambled during the encryption process. When decrypting, it is necessary to use a pseudo-random sequence to reversely scramble the image block in the opposite order and rules of encryption to restore the original arrangement order of the image block.

[0112] Step 76: Repeat the image scrambling inverse process and the pixel diffusion inverse process for several times;

[0113] Since the image scrambling and pixel diffusion operations are applied multiple times during the encryption process, the corresponding reverse process needs to be repeated several times during the decryption phase until the original image is completely restored. The number of repetitions of this step should be consistent with the number of operations during the encryption process.

[0114] The present invention proposes a method for secure transmission of medical images based on multimodal biometrics to address the security challenges faced by medical images during transmission. The present invention integrates deep learning technology, multimodal biometrics, data hiding and encryption functions to ensure the high security of medical information. First, a deep neural network is used to segment the lesion and non-lesion pixels in the medical image, and the least significant bit replacement technology is used to embed the lesion information into the non-lesion pixels to obtain an image carrying secret information. Furthermore, the present invention uses a biological feature extraction network to extract multimodal biometrics, and encodes these biometrics using the sender's private key and the device MAC address. Finally, the encoded multimodal biometrics and chaotic mapping are used to encrypt the medical image, thereby enhancing the security protection level of the medical image data.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for secure transmission of medical images based on multimodal biometrics, characterized in that: The method specifically comprises the following steps: S1: Feature extraction: construct multiple feature extraction network models to extract multimodal biological features respectively; S2: Image segmentation: Build an image segmentation network model to segment medical images to obtain images of lesion areas and non-lesion areas; S3: Pseudo-random sequence generation: construct a chaotic system as a pseudo-random sequence generator to obtain a pseudo-random sequence; S4: Data hiding: obtaining a timestamp, obtaining a random number sequence according to the timestamp and a pseudo-random sequence generator, using the sequence to select appropriate pixels in the non-lesion image, and hiding the lesion image pixels into the selected non-lesion image pixels using the least significant bit replacement method; S5: Data extraction: combining the non-lesion image and the hidden data obtained in step S4 to obtain a secret image, extracting the hidden lesion image pixels using the same timestamp and pseudo-random sequence, and combining it with the non-lesion area image to obtain a complete medical image; S6: Data encryption: The final key is generated using the extracted multimodal biometrics, user key and system MAC address, and the image is scrambled and pixel diffused using the final key and pseudo-random sequence to achieve data encryption; S7: Data decryption: The encrypted image is subjected to the reverse process of scrambling and pixel diffusion using the final key and pseudo-random sequence to restore the original medical image.

2. The method for secure transmission of medical images according to claim 1, characterized in that: In step S1, the feature extraction specifically includes the following steps: S11: Construct feature extraction network model architecture based on deep learning technology; S12: Use biological feature datasets to train multiple feature extraction networks; S13: Utilize the trained multiple feature extraction networks to extract multimodal biological features respectively.

3. The method for secure transmission of medical images according to claim 1, characterized in that: In step S2, the image segmentation specifically comprises the following steps: S21: Construct image segmentation network model architecture based on deep learning technology; S22: Use image segmentation dataset to train image segmentation network; S23: Segment the medical image using the trained image segmentation network to obtain images of the lesion area and images of the non-lesion area.

4. The method for secure transmission of medical images according to claim 1, characterized in that: In step S3, the pseudo-random sequence generation specifically includes: firstly constructing a chaotic system; then selecting a suitable system initial value, and continuously iterating to generate a pseudo-random sequence.

5. The method for secure transmission of medical images according to claim 1, characterized in that: In step S6, the data encryption specifically includes the following steps: S61: Obtain the multimodal biometrics according to step S1, obtain the system MAC address, and calculate the multimodal biometrics, the user key and the MAC address to obtain the final key; S62: Obtain a pseudo-random sequence using the final key and the pseudo-random sequence generation process in step S3; S63: converting the image into a plurality of N×N image blocks, and scrambling the image blocks using the acquired pseudo-random sequence; S64: After scrambling the image block, further scramble the position of each pixel in the block using the acquired pseudo-random sequence; S65: performing pixel diffusion on the scrambled image; S66: Repeat the image scrambling and pixel diffusion process for several times.

6. The method for secure transmission of medical images according to claim 1, characterized in that: In step S7, the data decryption specifically includes the following steps: S71: Obtain a key according to the key generation method in the encryption process of step S6; S72: Obtain a pseudo-random sequence using the final key and the pseudo-random sequence generation process in step S3; S73: using the acquired pseudo-random sequence, performing a reverse process of pixel diffusion on the encrypted image; S74: performing the inverse process of pixel scrambling within the block on the position of each pixel in the image block; S75: using the acquired pseudo-random sequence to perform the reverse process of image block scrambling; S76: Repeat the image scrambling inverse process and the pixel diffusion inverse process for several times.

Images