A method and system for constructing and analyzing medical records that cannot be detected

Abstract

The present invention proposes a method and system for constructing and analyzing undetectable medical records, the method comprising: generating a cover image based on a face mask image through a collaborative steganographic generation model combined with a text description; fusing the low-frequency features and high-frequency features of the secret image to obtain a fused image; based on the fused image and the cover image, performing multi-layer space-channel joint attention mechanism processing through a reversible hidden module to generate a stego image; extracting the fused image from the stego image through a decoding function for analysis and diagnosis by medical professionals. The method of the present invention effectively combines steganography technology and medical image processing, ensures the security of patient privacy information during transmission and storage, and improves the integration and diagnostic performance of medical data.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a construction and analysis method and system for unobservable medical records. Background Art

[0002] With the rapid development of medical informatization, telemedicine has become a key way to improve the accessibility and efficiency of medical services. In telemedicine scenarios, the transmission and management of patient medical data is particularly important. By building a patient-centered medical archive, various data such as patient demographic information and medical images can be effectively integrated to achieve integrated information management. This archive management method based on remote transmission significantly improves the interoperability and utilization efficiency of medical data, reduces the need for frequent data uploads, and further improves the overall efficiency of data transmission. At the same time, archiving management can also centrally store patient medical information, facilitate data sharing and collaboration across organizations, and promote efficient diagnosis and treatment. In addition, this centralized management method not only improves the uniformity and accuracy of medical services, but also provides patients with a more convenient and high-quality medical experience.

[0003] On the other hand, the inherent high sensitivity of medical archives is also a core issue. The medical data in the archives include patient medical records and image data, which are closely related to the individual's psychological and social status and his or her career.

[0004] Therefore, the use of encryption technology to protect medical records can effectively ensure the integrity and independence of patient information management. This technology can protect the security, privacy and integrity of sensitive data during storage and transmission. In addition, encryption technology also helps healthcare organizations meet regulatory requirements and prevent data leakage and unauthorized operations, thereby enhancing patient trust and improving information security. However, although encryption has significant advantages in protecting information, it may also attract the attention of attackers, thereby increasing the risk of privacy violations for patients. Attackers can bypass encryption by cracking encryption algorithms or stealing encryption keys to gain access to sensitive data in the archives. In addition, the encryption of medical records makes it more difficult for doctors to classify them and accurately match each record with the corresponding patient. Summary of the invention

[0005] In view of the above situation, the main purpose of the present invention is to propose an unobservable medical record construction and analysis method and system to solve the above technical problems.

[0006] The present invention proposes a method for constructing and analyzing an unobservable medical record, the method comprising the following steps:

[0007] Step 1: Generate a cover image based on the face mask image by using a collaborative steganography generation model combined with text description;

[0008] Step 2: Fuse the low-frequency features and high-frequency features of the secret image to obtain a fused image;

[0009] Step 3: Based on the fused image and the cover image, a multi-layer space-channel joint attention mechanism is processed through a reversible hidden module to generate a stego image, where the stego image is used to construct an undetectable medical file;

[0010] Step 4: extract the fused image from the stego-image through a decoding function for analysis and diagnosis by medical professionals.

[0011] The present invention also provides a system for constructing and analyzing an unobservable medical record, the system comprising:

[0012] Image generation module for:

[0013] Based on the face mask image, a cover image is generated by combining the collaborative steganographic generation model with the text description;

[0014] Fusion modules for:

[0015] The low-frequency features and high-frequency features of the secret image are fused to obtain a fused image;

[0016] Reversible hidden modules for:

[0017] Based on the fusion image and the cover image, a multi-layer spatial-channel joint attention mechanism is used to generate a stego image, and the stego image is used to construct an undetectable medical file;

[0018] Image extraction module for:

[0019] The fused image is extracted from the stego-image by a decoding function for analysis and diagnosis by medical professionals.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] 1. The present invention has finely optimized the low-frequency components and noise tolerance of the generated image, ensuring that the cover image has a smooth visual effect and maintains high fidelity of details when displayed over a large area. In addition, the optimization process significantly improves the robustness of the model to noise interference;

[0022] 2. The present invention optimizes the frequency distribution of the cover image and increases the proportion of low-frequency areas, which allows embedding more information while maintaining high visual quality. The capacity and robustness of the steganographic data are improved, and the difficulty of the detection algorithm to identify the hidden data is reduced by increasing the low-frequency components, thereby effectively reducing the visual difference and enhancing the concealment and anti-interference ability of steganography;

[0023] 3. The present invention introduces noise tolerance loss, injects a small amount of noise into the image, and minimizes the impact of the noise on the image structure and content, thereby ensuring that the generated image can maintain a high visual quality when disturbed by noise.

[0024] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description or learned through embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flowchart of an unobservable medical record construction and analysis method proposed by the present invention;

[0026] Figure 2 This is a framework diagram of an unobservable medical record construction and analysis system proposed by the present invention. DETAILED DESCRIPTION

[0027] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0028] These and other aspects of the embodiments of the present invention will be apparent with reference to the following description and accompanying drawings. In these descriptions and accompanying drawings, some specific implementations of the embodiments of the present invention are specifically disclosed to represent some ways of implementing the principles of the embodiments of the present invention, but it should be understood that the scope of the embodiments of the present invention is not limited thereto.

[0029] See also Figure 1 The embodiment of the present invention provides a method for constructing and analyzing an unobservable medical record, the method comprising the following steps:

[0030] Step 1: Generate a cover image based on the face mask image by using a collaborative steganography generation model combined with text description;

[0031] In this step, based on the face mask image, the cover image is generated by the collaborative steganography generation model combined with the text description. The relationship between the corresponding process is:

[0032] ;

[0033] in, Represents the cover image, represents the mask image, Represents a text description, It means that it has been processed by the collaborative steganalysis generation model;

[0034] The collaborative steganalysis generation model works as follows:

[0035] The collaborative steganography generation model is a diffusion model, and the forward diffusion process of the diffusion model is expressed by the following relationship:

[0036] ;

[0037] in, represents the conditional probability distribution, All represent data, Represents the initial data, represents the total number of iteration steps, represents the single-step conditional probability, Indicates The data of the iteration steps, Indicates Data for each iteration step;

[0038] The reverse process of the diffusion model is expressed by the following relationship:

[0039] ;

[0040] in, represents the conditional probability in the reverse process, It means that the Gaussian distribution of covariance is processed. It means that after being processed by the mean function, Indicates the result of the data of the previous iteration being processed by the mean function;

[0041] In each diffusion step, the dynamic diffuser calculates the influence function of each mode according to the conditions of the current mode. The relationship between the corresponding process is:

[0042] ;

[0043] in, Indicates The mode is in The influence function for each iteration step is represents a dynamic diffuser, Indicates Conditional input of a modal;

[0044] The influence of each pixel position is weighted and synthesized to obtain the weighted influence intensity. The relationship between the corresponding process is:

[0045] ;

[0046] in, Indicates The weighted influence strength of each mode at the pixel position, represents the pixel position, represents the total number of modes, Indicates The influence strength of each mode at the pixel position, Indicates The influence strength of each modality at the pixel position;

[0047] In the reverse diffusion process of the diffusion model, the weighted influence intensity of all modes at the pixel position is denoised and predicted to obtain the denoising result. The relationship between the corresponding process is:

[0048] ;

[0049] in, Indicated in The denoising result of the iterative steps is represents the weighted influence strength of the modality at the pixel position, represents the denoising result obtained after the conditional input of given data, iteration steps and mode, Represents the dot product operation;

[0050] In the process of generating cover images based on face mask images, through collaborative steganography generation model and text description, noise prediction loss is introduced. The expression of noise prediction loss is:

[0051] ;

[0052] in, represents the noise prediction loss function, represents the parameters of the diffusion model, represents the time step, represents the real sample of training data, represents the sample after noise is added in the time step, represents the covariance matrix used to describe the Gaussian distribution, represents noise sampled from a multidimensional Gaussian distribution with zero mean and unit variance, represents Gaussian noise, represents a Gaussian distribution, represents the noise predicted by the neural network controlled by the parameters of the diffusion model;

[0053] In the process of generating cover images based on face mask images, through collaborative steganography generation model and text description, low-frequency preference loss is also introduced. The expression of low-frequency preference loss is:

[0054] ;

[0055] in, represents the low-frequency preference loss function, represents the high frequency region, represents the frequency coordinate in the frequency domain representation of the image, It means after Fourier transform processing;

[0056] In the process of generating the cover image based on the face mask image, through the collaborative steganography generation model and combining the text description, the noise tolerance loss is also introduced. The expression of the noise tolerance loss is:

[0057] ;

[0058] in, represents the noise tolerance loss function, represents the injected noise;

[0059] In the process of generating cover images based on face mask images, through collaborative steganography generation model and text description, total variation loss is also introduced. The expression of total variation loss is:

[0060] ;

[0061] in, represents the total variation loss function, Represents pixel coordinates.

[0062] Furthermore, in this step, a low-frequency preference loss is introduced to ensure the smoothness and consistency of the cover image when it is displayed on a large area;

[0063] The noise tolerance loss injects a small amount of noise into the image and minimizes the impact of the noise on the image structure and content, thereby ensuring that the generated image maintains high visual quality even when it is disturbed by noise;

[0064] The total variation loss reduces the high-frequency variations in the image, making the image more suitable for data hiding.

[0065] Step 2: Fuse the low-frequency features and high-frequency features of the secret image to obtain a fused image;

[0066] In step 2, the low-frequency features and high-frequency features of the secret image are fused to obtain a fused image. The relationship between the corresponding process is:

[0067] ;

[0068] in, and Both represent low-frequency features. and Both represent high-frequency features. represents the fused image, Indicates that it has been processed by the decoder module. It means that after being processed by the low-frequency feature fusion function, Indicates that it has been processed by the high-frequency feature fusion function.

[0069] Specifically, the decoder module in this step is composed of a fully connected layer, which can map the input features to a high-dimensional space, or restore the input features to a target state; the decoder generates the final output by learning the mapping from the feature space to the output space.

[0070] Step 3: Based on the fused image and the cover image, a multi-layer space-channel joint attention mechanism is processed through a reversible hidden module to generate a stego image, where the stego image is used to construct an undetectable medical file;

[0071] In step 3, based on the fused image and the cover image, a multi-layer space-channel joint attention mechanism is processed through a reversible hidden module to generate a stego image. The relationship between the corresponding process is:

[0072] ;

[0073] in, represents a stego-image, The representation is processed by a reversible hidden module containing a multi-layer spatial-channel joint attention mechanism;

[0074] The reversible hidden module with multi-layer spatial-channel joint attention mechanism works as follows:

[0075] The fused image and the cover image are processed by discrete wavelet transform to obtain the low-frequency subband of the fused image and the high-frequency subband of the cover image. The relationship between the corresponding processes is:

[0076] ;

[0077] in, represents discrete wavelet transform, represents the high frequency subband of the cover image, represents the low frequency subband of the fused image;

[0078] The inverse neural network is used to embed the fused image information into the high-frequency area of ​​the cover image. The corresponding relationship is:

[0079] ;

[0080] in, Represents the cover image output by the current processing layer. Represents the cover image output by the previous processing layer, represents the constant factor, It means that after nonlinear transformation, Represents the fused image output by the current processing layer, represents the fused image output by the previous processing layer, It means that after nonlinear transformation, Indicates that it has been mapped;

[0081] The image processed by the inverse neural network is then processed by inverse wavelet transform to restore the stego image. The relationship between the corresponding process is:

[0082] ;

[0083] in, represents the image processed by the reversible neural network, Indicates the extra information that does not need to be retained during the steganography process. Indicates the redundant information during the recovery process. Indicates that it has been processed by inverse wavelet transform.

[0084] Step 4: extracting the fused image from the stego-image by a decoding function for analysis and diagnosis by medical professionals;

[0085] In step 4, the fused image is extracted from the stego-image by a decoding function, and the corresponding process has the following relation:

[0086] ;

[0087] in, Indicates that it has been processed by the decoding function;

[0088] The decoding function works as follows:

[0089] By performing discrete wavelet transform on the stego image, we can obtain the low-frequency and high-frequency sub-bands of the stego image. The corresponding relationship is:

[0090] ;

[0091] in, It means that after discrete wavelet transform, represents a random variable, represents a random variable after discrete wavelet transformation;

[0092] The fused image is recovered from the stego-image, and the corresponding relationship is:

[0093] ;

[0094] in, Represents the stego-image output by the current processing layer, represents the stego-image output by the previous processing layer, represents the adjustment parameters, Represents the redundant information related to the stego-image in the current processing layer.

[0095] See also Figure 2 The embodiment of the present invention further provides a system for constructing and analyzing an unobservable medical record, the system comprising:

[0096] Image generation module for:

[0097] Based on the face mask image, a cover image is generated by combining the collaborative steganographic generation model with the text description;

[0098] Fusion modules for:

[0099] The low-frequency features and high-frequency features of the secret image are fused to obtain a fused image;

[0100] Reversible hidden modules for:

[0101] Based on the fusion image and the cover image, a multi-layer spatial-channel joint attention mechanism is used to generate a stego image, and the stego image is used to construct an undetectable medical file;

[0102] Image extraction module for:

[0103] The fused image is extracted from the stego-image by a decoding function for analysis and diagnosis by medical professionals.

[0104] Further, please refer to Table 1 and Table 2, where the average PSNR and SSIM values ​​of the stego images generated by different methods are compared when multiple secret images are hidden, where the first column is the number of hidden secret images and the first row is the method. It can be intuitively seen from the table that the method proposed in the present invention is superior to other methods in both indicators, which shows that the present invention can not only better maintain the visual quality of the image, but also effectively reduce distortion in the process of information hiding, further verifying its superiority in steganographic performance.

[0105] Among them, the existing technology 1: HINET: J. Jing, X. Deng, M. Xu, J. Wang, and Z. Guan, "Hinet: Deep image hiding by invertible network," in Proc. IEEE / CVF Int.Conf. Comput.Vis., 2021, pp. 4733–4742. An image hiding framework based on invertible neural network (INN) HiNet hides the secret image in the high-frequency subband of the wavelet domain through an inverse learning mechanism;

[0106] Prior Art 2: Deepmih: ZG et al., “Deepmih: Deep invertible network for multiple image hiding,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 45, no.1, pp.372–390, Jan 2023. A multi-image hiding method based on a reversible neural network, which realizes image hiding and restoration through forward and reverse processes, and uses importance map module and low-frequency wavelet loss to improve hiding performance;

[0107] Prior Art 3: F. Li, Y. Sheng, X. Zhang, and C. Qin, “iscmis: Spatial-channel attention based deep invertible network for multi-image steganography,” IEEE Trans. Multimedia, vol. 26, pp. 3137–3152, 2024 A deep reversible network based on spatial-channel joint attention mechanism for multi-image steganography, which achieves high-capacity hiding while improving visual quality and anti-detection ability through reversible network structure and attention guidance.

[0108] Table 1: Average PSNR of steganatomical image quality for multiple secret images

[0109]

[0110] Table 2: Average SSIM of steganatomical image quality for multiple secret images

[0111]

[0112] In summary, according to the construction and analysis method of undetectable medical records provided in this embodiment, the proposed collaborative steganographic generation model can dynamically optimize the cover image and effectively embed medical data into it to form a safe and reliable medical record. This method significantly improves data embedding capability, undetectability and security while ensuring privacy transmission. The experimental results show that compared with existing research methods, this method has obvious advantages in data capacity, image quality and security, and demonstrates excellent performance and wide application potential.

[0113] This embodiment also discloses a terminal device, which includes a memory and a processor. When the processor executes a computer program stored in the memory, the above-mentioned unobservable medical record construction and analysis method is implemented.

[0114] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0115] The memory may be an internal storage unit of the terminal device, such as a hard disk or memory of the terminal device. The memory may also be an external storage device of the terminal device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device. Furthermore, the memory may include both an internal storage unit and an external storage device of the terminal device. The memory is used to store the computer program and other programs and data required by the terminal device. The memory may also be used to temporarily store data that has been output or is to be output.

[0116] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional unit.

[0117] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0118] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0119] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for constructing and analyzing an unobservable medical record, characterized in that: The method comprises the following steps: Step 1: Generate a cover image based on the face mask image by using a collaborative steganography generation model combined with text description; Step 2: Fuse the low-frequency features and high-frequency features of the secret image to obtain a fused image; Step 3: Based on the fused image and the cover image, a multi-layer space-channel joint attention mechanism is processed through a reversible hidden module to generate a stego image, where the stego image is used to construct an undetectable medical file; Step 4: extracting the fused image from the stego-image by a decoding function for analysis and diagnosis by medical professionals; Among them, in the step 1, based on the face mask image, the cover image is generated by the collaborative steganography generation model combined with the text description, and the relationship between the corresponding process is: ； in, Represents the cover image, represents the mask image, Represents a text description, Indicates that it has been processed by the collaborative steganalysis generation model.

2. The method for constructing and analyzing an unobservable medical record according to claim 1, characterized in that: In step 1, based on the face mask image, a cover image is generated by a collaborative steganography generation model combined with a text description, wherein the working principle of the collaborative steganography generation model is as follows: The collaborative steganography generation model is a diffusion model, and the forward diffusion process of the diffusion model is expressed by the following relationship: ； in, represents the conditional probability distribution, All represent data, Represents the initial data, represents the total number of iteration steps, represents the single-step conditional probability, Indicates The data of the iteration steps, Indicates Data for each iteration step; The reverse process of the diffusion model is expressed by the following relationship: ； in, represents the conditional probability in the reverse process, It means that the Gaussian distribution of covariance is processed. It means that after being processed by the mean function, Indicates the result of the data of the previous iteration being processed by the mean function; In each diffusion step, the dynamic diffuser calculates the influence function of each mode according to the conditions of the current mode. The relationship between the corresponding process is: ； in, Indicates The mode is in The influence function for each iteration step is represents a dynamic diffuser, Indicates Conditional input of a modal; The influence of each pixel position is weighted and synthesized to obtain the weighted influence intensity. The relationship between the corresponding process is: ； in, Indicates The weighted influence strength of each mode at the pixel position, represents the pixel position, represents the total number of modes, Indicates The influence strength of each mode at the pixel position, Indicates The influence strength of each modality at the pixel position; In the reverse diffusion process of the diffusion model, the weighted influence intensity of all modes at the pixel position is denoised and predicted to obtain the denoising result. The relationship between the corresponding process is: ； in, Indicated in The denoising result of the iterative steps is represents the weighted influence strength of the modality at the pixel position, represents the denoising result obtained after the conditional input of given data, iteration steps and mode, Represents the dot product operation.

3. The method for constructing and analyzing an unobservable medical record according to claim 2, characterized in that: In the process of generating cover images based on face mask images, through collaborative steganography generation model and text description, noise prediction loss is introduced. The expression of noise prediction loss is: ； in, represents the noise prediction loss function, represents the parameters of the diffusion model, represents the time step, represents the real sample of training data, represents the sample after noise is added in the time step, represents the covariance matrix used to describe the Gaussian distribution, represents noise sampled from a multidimensional Gaussian distribution with zero mean and unit variance, represents Gaussian noise, represents a Gaussian distribution, represents the noise in the neural network predictions controlled by the parameters of the diffusion model.

4. The method for constructing and analyzing an unobservable medical record according to claim 3, characterized in that: In the process of generating cover images based on face mask images, through collaborative steganography generation model and text description, low-frequency preference loss is also introduced. The expression of low-frequency preference loss is: ； in, represents the low-frequency preference loss function, represents the high frequency region, represents the frequency coordinate in the frequency domain representation of the image, It means after Fourier transform processing.

5. The method for constructing and analyzing an unobservable medical record according to claim 4, characterized in that: In the process of generating the cover image based on the face mask image, through the collaborative steganography generation model and combining the text description, the noise tolerance loss is also introduced. The expression of the noise tolerance loss is: ； in, represents the noise tolerance loss function, represents the injected noise.

6. The method for constructing and analyzing an unobservable medical record according to claim 5, characterized in that: In the process of generating cover images based on face mask images, through collaborative steganography generation model and text description, total variation loss is also introduced. The expression of total variation loss is: ； in, represents the total variation loss function, Represents pixel coordinates.

7. The method for constructing and analyzing an unobservable medical record according to claim 6, characterized in that: In step 2, the low-frequency features and high-frequency features of the secret image are fused to obtain a fused image. The relationship between the corresponding process is: ； in, and Both represent low-frequency features. and Both represent high-frequency features. represents the fused image, Indicates that it has been processed by the decoder module. It means that after being processed by the low-frequency feature fusion function, Indicates that it has been processed by the high-frequency feature fusion function.

8. The method for constructing and analyzing an unobservable medical record according to claim 7, characterized in that: In step 3, based on the fused image and the cover image, a multi-layer space-channel joint attention mechanism is processed through a reversible hidden module to generate a stego image. The relationship between the corresponding process is: ； in, represents a stego-image, The representation is processed by a reversible hidden module containing a multi-layer spatial-channel joint attention mechanism; The reversible hidden module with multi-layer spatial-channel joint attention mechanism works as follows: The fused image and the cover image are processed by discrete wavelet transform to obtain the low-frequency subband of the fused image and the high-frequency subband of the cover image. The relationship between the corresponding processes is: ； in, represents discrete wavelet transform, represents the high frequency subband of the cover image, represents the low frequency subband of the fused image; The inverse neural network is used to embed the fused image information into the high-frequency area of ​​the cover image. The corresponding relationship is: ； in, Represents the cover image output by the current processing layer. Represents the cover image output by the previous processing layer, represents the constant factor, It means that after nonlinear transformation, Represents the fused image output by the current processing layer, represents the fused image output by the previous processing layer, It means that after nonlinear transformation, Indicates that it has been mapped; The image processed by the inverse neural network is then processed by inverse wavelet transform to restore the stego image. The relationship between the corresponding process is: ； in, represents the low-frequency and high-frequency subbands of the stego-image, Indicates the extra information that does not need to be retained during the steganography process. Indicates the redundant information during the recovery process. Indicates that it has been processed by inverse wavelet transform.

9. The method for constructing and analyzing an unobservable medical record according to claim 8, characterized in that: In step 4, the fused image is extracted from the stego image by a decoding function, and the relationship between the corresponding process is: ； in, Indicates that it has been processed by the decoding function; The decoding function works as follows: By performing discrete wavelet transform on the stego image, we can obtain the low-frequency and high-frequency sub-bands of the stego image. The corresponding relationship is: ； in, It means that after discrete wavelet transform, represents a random variable, represents a random variable after discrete wavelet transformation; The fused image is recovered from the stego-image, and the corresponding relationship is: ； in, Represents the stego-image output by the current processing layer, represents the stego-image output by the previous processing layer, represents the adjustment parameters, Represents the redundant information related to the stego-image in the current processing layer.

10. A system for constructing and analyzing medical records that is not observable, characterized in that: The system applies an unobservable medical record construction and analysis method as described in any one of claims 1 to 9 above, and the system comprises: Image generation module for: Based on the face mask image, a cover image is generated by combining the collaborative steganographic generation model with the text description; Fusion modules for: The low-frequency features and high-frequency features of the secret image are fused to obtain a fused image; Reversible hidden modules for: Based on the fusion image and the cover image, a multi-layer spatial-channel joint attention mechanism is used to generate a stego image, and the stego image is used to construct an undetectable medical file; Image extraction module for: The fused image is extracted from the stego-image by a decoding function for analysis and diagnosis by medical professionals.

Images