PACS image desensitization processing method and device, electronic equipment and storage medium
By combining image segmentation and OCR, along with the first and second confirmation lists and contextual analysis, the problem of efficient and accurate desensitization of PACS images in complex scenarios was solved. This achieved precise protection of sensitive information and retention of non-sensitive information, while reducing costs.
Patent Information
- Application Number
- CN202511666945.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-14
- Publication Date
- 2026-03-03
AI Technical Summary
Existing PACS image desensitization methods cannot efficiently and accurately identify and protect sensitive information in complex scenarios, resulting in false desensitization and missed desensitization. Furthermore, they are costly and difficult to adapt to the differences between different equipment and hospitals.
The text region is separated by an image segmentation algorithm and the text content is recognized by an OCR engine. Through the analysis of the first confirmation list, the second confirmation list and contextual association, sensitive and non-sensitive information is accurately identified and distinguished to generate desensitized images.
It achieves efficient and accurate recognition and desensitization of various texts in PACS images in complex scenarios, with almost no false or missed desensitization, protecting privacy and security, and at a low cost.
Smart Images
Figure CN121601167A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical imaging technology, and in particular relates to a method, apparatus, electronic device and storage medium for desensitizing PACS images. Background Technology
[0002] To address the issues of incompatibility between hospitals' PACS images and the prohibition of internet access on computers, doctors can only manually capture screenshots of the PACS images on the screen using screenshot software on the imaging workstation, and then copy the captured images to mobile devices via USB for sharing. This method is extremely cumbersome, and hospitals may also prohibit the use of USB ports. Directly acquired PACS images contain sensitive information such as the patient's name, gender, and hospital, which inevitably leads to information leakage on users' mobile devices. This is highly detrimental to the hospital's information security management. Therefore, PACS images should be anonymized before being sent to users' mobile devices.
[0003] Most existing image anonymization methods directly identify and anonymize sensitive information in images according to preset rules, without considering the protection of non-sensitive information. This can lead to non-sensitive information being mistakenly anonymized as sensitive information, often containing crucial image information. Users do not want non-sensitive information to be mistakenly anonymized. Furthermore, existing anonymization methods suffer from low accuracy, resulting in the omission of some sensitive information. These shortcomings are particularly evident in PACS image anonymization, where users naturally desire PACS images where sensitive information (such as names and hospital names) is anonymized while non-sensitive information (such as PACS technical parameters or technical terms) is retained.
[0004] Existing image desensitization technologies can be mainly categorized as follows:
[0005] The first method is fixed-coordinate masking based on regular regions. This is the most primitive and simplest approach: predefine fixed regions (e.g., a rectangle in the upper left or lower right corner) where sensitive information (such as patient name or hospital name) typically appears in the PACS image. During desensitization, regardless of the actual content of this region, it is directly masked with a color block. However, this method has poor adaptability. PACS images from different devices, hospitals, and examination types can vary greatly in text information and layout. Fixed coordinates cannot cover all situations, leading to significant omissions (sensitive information is in other locations) or over-masking (masking key diagnostic information).
[0006] The second approach is basic OCR (Optical Character Recognition) combined with simple keyword matching. This is a common automated solution: an OCR engine is used to perform full-text recognition on the entire PACS image. A sensitive term, "second confirmation list" (e.g., containing specific names like "Zhang San" or "Peking Union Medical College Hospital"), is set. The OCR results are then matched against the second confirmation list. Successfully matched text is masked at its coordinates. However, maintaining the second confirmation list is difficult. It needs to include all possible patient and hospital names, a massive and dynamically changing set that is virtually inexhaustible, leading to a very high risk of missed desensitization. Furthermore, the overall accuracy is low, with a high risk of false desensitization, as it blindly masks all matched strings. If a parameter in the image unfortunately overlaps with a word in the second confirmation list, it will be falsely flagged.
[0007] The third approach is an end-to-end deep learning model. This doesn't rely on OCR but directly collects large amounts of labeled PACS image data, marking sensitive information at the pixel level. A deep learning model (such as one based on U-Net or Mask R-CNN segmentation networks) is trained, inputting the original image and directly outputting a "desensitization mask" indicating the pixel areas that need to be masked. However, this approach is highly data-dependent and costly. Model performance heavily depends on a large volume of high-quality, comprehensive labeled data. Collecting and labeling medical data is extremely expensive, and the model has poor adaptability to new layouts and fonts not present in the training set. Furthermore, this technology is difficult to update and maintain; adding a new type of sensitive information may require recollecting data and retraining the model, making the process cumbersome. Summary of the Invention
[0008] The purpose of this invention is to provide a method, apparatus, electronic device, and storage medium for desensitizing PACS images, so as to achieve safe and efficient acquisition of PACS images and protect the privacy of patients and hospitals.
[0009] To achieve the above objectives, the technical solution of the present invention is as follows:
[0010] In a first aspect, the present invention provides a method for desensitizing PACS images, comprising the following steps:
[0011] The original target image is acquired and cached on the terminal. The original target image is divided using an image segmentation algorithm to separate the text region from the background, locate the region of interest of the text, and output the image block of the region of interest corresponding to the region of interest of the text.
[0012] The region of interest image patch is input into the OCR engine for text content recognition, and the text type, text content and coordinate value in the original target image are output for each region of interest image patch;
[0013] Based on coordinate values, region-of-interest (ROI) image blocks are merged into logical text blocks. The logical text blocks inherit the text type and text content of the individual ROI image blocks before merging, and contain the semantic relationships between the merged ROI image blocks.
[0014] The logical text blocks are validated based on their text type, content, and semantic relationships. This validation process includes: validation of the first confirmation list, validation of the second confirmation list, and contextual analysis. Based on the validation results, all logical text blocks are input into the list to be desensitized or the list that can be displayed. The coordinate regions corresponding to the logical text blocks in the display list of the original target image are not processed. The coordinate regions corresponding to the logical text blocks in the list to be desensitized are masked. The final desensitized image is then output.
[0015] The text verification process includes:
[0016] Match the logical text block with the first confirmed list. If the first confirmed list matches successfully, add the logical text block to the displayable list.
[0017] If the first confirmation list fails to match, the logical text block is matched with the second confirmation list. If the second confirmation list matches successfully and the semantic relationship is patient identity feature, the logical text block is analyzed using context association. Logical text blocks with analysis results of sensitive information are added to the list to be anonymized, and logical text blocks with analysis results of non-sensitive information are added to the list that can be displayed. If the second confirmation list matches successfully and the semantic relationship is hospital name feature, the corresponding logical text block is added to the list to be anonymized.
[0018] If a logical text block does not match either the first confirmation list or the second confirmation list, then the logical text block is marked as irrelevant content and added to the displayable list.
[0019] Repeat the above text validation process until all logical text blocks have been processed.
[0020] The method for separating the text region from the background, locating the region of interest (ROI) of the text, and outputting the corresponding ROI image block is as follows: The Otsu binarization algorithm is used to perform adaptive threshold image segmentation on the grayscale image of the original target image to obtain the corresponding text region binarized image. The ROI image block is then obtained after secondary optimization of the binarized image. The algorithm process formula is as follows:
[0021]
[0022]
[0023] in, The inter-class variance is used to measure the degree of separation between the foreground and background. To make the inter-class variance The biggest one value, Indicates the candidate threshold. The function that takes the maximum value. Background class Average gray level, Foreground Average gray level, This represents the global average gray level of the entire image. This indicates that the pixel is classified as the background class. The probability (i.e., the proportion of background pixels). This indicates that the pixel is classified as a foreground pixel. The probability (i.e., the proportion of text pixels) is given by L, where L is the gray level of the grayscale image.
[0024] The method for merging image blocks of the region of interest into logical text blocks is as follows: Text block proximity and similarity measurement algorithm:
[0025]
[0026] in, and This represents two different regions of interest image patches. Image patch representing region of interest and Euclidean distance of the center point Indicates the distance threshold. This represents the vertical coordinate difference of the center. This indicates the height threshold.
[0027] The context association analysis uses a distance-based context association confidence algorithm:
[0028]
[0029] Where S represents the relevance score, K represents the keywords, and Keywords represents the keyword list. For candidate logical text blocks, Let D be the predefined context keyword logical text block, and let D be the spatial distance between the candidate logical text block and the predefined context keyword logical text block. For the preset threshold, when At that time, confirm that there is a contextual relationship.
[0030] The method further includes: after the terminal device sends the desensitized image, recording the image request, and creating a medical record folder locally or in the cloud, storing the original target image and the desensitized image in the medical record folder.
[0031] The text types include: Chinese characters, numbers, letters, and special symbols.
[0032] Secondly, the present invention discloses an apparatus for PACS image desensitization processing, comprising:
[0033] Image acquisition module, used to acquire raw target images;
[0034] The image preprocessing module is used to process the original target image, separate the text region from the background, locate the region of interest of the text, and output the region of interest image block corresponding to the region of interest of the text.
[0035] The text content recognition module inputs the region of interest (ROI) image block into the OCR engine for text content recognition and outputs the text type, text content, and coordinate values of each ROI image block in the original target image.
[0036] The text block aggregation module is used to merge image blocks of the region of interest into logical text blocks;
[0037] The text verification module performs first confirmation list verification, second confirmation list verification, and contextual analysis on all logical text blocks, and inputs the logical text blocks into the corresponding list to be de-identified or the list to be displayed.
[0038] The desensitization module performs masking on the coordinate areas corresponding to the logical text blocks in the desensitization list and outputs the final desensitized image.
[0039] The storage module is used to store the original target image and the desensitized image.
[0040] Thirdly, the present invention discloses an electronic device, comprising:
[0041] Memory, used to store computer programs / instructions;
[0042] A processor for executing the computer program / instructions to implement the steps of the above-described PACS image desensitization processing method.
[0043] Fourthly, the present invention discloses a computer-readable storage medium storing a computer program / instructions thereon, characterized in that, when the computer program / instructions are executed by a processor, they implement the steps of the above-described method for desensitizing PACS images.
[0044] Beneficial effects: The PACS image desensitization processing method of the present invention can achieve efficient and accurate identification and desensitization of various texts in PACS images in complex scenarios, with almost no false or missed desensitization, protecting privacy and security, and having a lower cost. Attached Figure Description
[0045] Figure 1 This is a flowchart of the PACS image desensitization processing method of the present invention. Detailed Implementation
[0046] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.
[0047] Example 1
[0048] refer to Figure 1 As shown, a method for desensitizing PACS images includes the following steps:
[0049] The original target image is acquired and cached on the terminal. The original target image is divided using an image segmentation algorithm to separate the text region from the background, locate the region of interest (ROI) of the text, and output the image block of the region of interest corresponding to the text ROI.
[0050] The region of interest image patch is input into the OCR engine for text content recognition, and the text type, text content and coordinate value in the original target image are output for each region of interest image patch;
[0051] Based on coordinate values, region-of-interest (ROI) image blocks are merged into logical text blocks. The logical text blocks inherit the text type and text content of the individual ROI image blocks before merging, and contain the semantic relationships between the merged ROI image blocks.
[0052] The logical text blocks are validated based on their text type, content, and semantic relationships. This validation process includes: validation of the first confirmation list, validation of the second confirmation list, and context validation. Based on the validation results, all logical text blocks are input into the list to be desensitized or the list that can be displayed. The coordinate regions corresponding to the logical text blocks in the display list of the original target image are not processed. The coordinate regions corresponding to the logical text blocks in the list to be desensitized are masked. The final desensitized image is then output.
[0053] First, the logical text block is matched against the first confirmation list. If the first confirmation list matches successfully, the logical text block is added to the displayable list. If the first confirmation list fails to match, the logical text block is matched against the second confirmation list. If the second confirmation list matches successfully and the type is "patient," contextual analysis is applied to the logical text block. Logical text blocks with sensitive information in the analysis result are added to the desensitization list, and logical text blocks with non-sensitive information in the analysis result are added to the displayable list. If the second confirmation list matches successfully and the type is "hospital," the corresponding logical text block is added to the desensitization list. If the logical text block matches neither the first nor the second confirmation list, it is marked as irrelevant content and added to the displayable list. This text validation process is repeated until all logical text blocks have been processed.
[0054] Before text verification, the content of the first and second confirmation lists needs to be updated. The second and first confirmation lists can be updated automatically on a regular basis, or they can be updated manually by the user.
[0055] Text types: Chinese characters, English characters (corresponding to name pinyin, words, medical parameter units), numbers, special symbols; semantic relationships include corresponding hospital name feature relationships, patient identity feature relationships, and medical parameter feature relationships. The information in the first confirmation list corresponds to the medical parameter feature relationships; the information in the second confirmation list corresponds to the hospital name feature relationships and patient identity feature relationships.
[0056] More specifically, a method for desensitizing PACS images includes the following steps:
[0057] Step S101: The user equipment and the terminal equipment establish a communication connection.
[0058] The user device connects to the terminal device via Bluetooth. User authentication is required before the connection begins. After a successful connection, the user sends an image acquisition request to the terminal device, which then connects to several image computers. The user device can be a mobile phone or a personal laptop.
[0059] Step S102: Acquire the original target image
[0060] In response to an image acquisition request, the terminal device controls the image retrieval hardware of the designated image computer to capture the current PACS medical image on the medical monitor as the original target image, rather than caching it to the terminal device.
[0061] Step S103: Perform image segmentation preprocessing on the original target image.
[0062] The original target image is preprocessed, which includes: using image segmentation algorithms (such as traditional algorithms based on edge and texture features) to locate the region of interest (ROI) of the text; automatically identifying and defining all regions containing text in the original target image (usually located at the four vertices and edges of the image); and performing secondary optimization on the segmented ROI (grayscale enhancement and noise removal) to improve character clarity and output the corresponding ROI image blocks. Preprocessing can effectively eliminate the interference of anatomical background, which accounts for most of the image, significantly improve the processing speed and recognition accuracy of OCR, and facilitate subsequent spatial relationship analysis of logical text blocks.
[0063] In this embodiment, to separate the text region from the background, the Otsu binarization algorithm is used to perform adaptive threshold segmentation on the grayscale image of the original target image. The algorithm steps are as follows:
[0064] Let the gray levels of the input grayscale image be L (usually 0~255), the number of pixels with gray level i be ni, and the total number of pixels in the image be N. Then the gray level probability distribution is:
[0065]
[0066] Where pi is the probability of gray level i appearing in the image, ni is the number of pixels with gray level i, and N is the total number of pixels in the image.
[0067] Set a candidate threshold Pixels are divided into background categories. (Gray level 0~) (usually corresponding to the background area), and foreground areas. (Grayscale) (+1~L-1, usually corresponding to text regions), then the probabilities of the two types of occurrence are respectively:
[0068]
[0069]
[0070] in, This indicates that the pixel is classified as the background class. The probability (i.e., the proportion of background pixels). This indicates that the pixel is classified as a foreground pixel. The probability (i.e., the percentage of text pixels). To represent the candidate threshold, the algorithm will iterate through all possible thresholds. value.
[0071] Calculate the background class separately Average gray level Foreground Average gray level Global average gray level of the entire image Then at the threshold Inter-class variance for:
[0072]
[0073] in The optimal threshold measures the degree of separation between the foreground and background classes. This makes the inter-class variance The biggest one value:
[0074]
[0075] in For the algorithm to finally determine the optimal threshold, This is the function that takes the maximum value.
[0076] The algorithm finds an optimal grayscale threshold through calculation. This makes the inter-class variance of the foreground (text) and background (anatomical image) segmented according to the threshold... Maximizing the inter-class variance ensures that the difference between the foreground and background is most pronounced, resulting in the clearest binarized image of the text region.
[0077] Step S104: Call the OCR engine to recognize the region of interest image patch.
[0078] The region of interest (ROI) image blocks obtained in step S103 are input into an open-source OCR engine (PaddleOCR) configured with multilingual recognition capabilities. The OCR engine is responsible for detecting and recognizing the specific text type and text content (e.g., "Chinese characters", "surname", "Zhang", "English characters", "n", "Hospital") within each ROI image block, and outputting the content of each ROI image block and its precise coordinates in the original target image.
[0079] Step S105: Merge image blocks of the region of interest to generate logical text blocks.
[0080] An OCR engine may break a line of text into multiple text blocks, each of which may contain only a single Chinese or English character. These blocks need to be merged. The scattered text blocks output by the OCR are merged and logically grouped according to their spatial position (such as horizontal and vertical alignment) to form more semantic logical text blocks. The logical text blocks inherit the text type and text content of the individual region of interest image blocks before merging, and contain the semantic relationships between the merged region of interest image blocks, which is convenient for subsequent analysis.
[0081] The merging method uses a text block proximity and similarity measurement algorithm:
[0082]
[0083] in, and This represents two different regions of interest image patches. Image patch representing region of interest and Euclidean distance of the center point Indicates the distance threshold. This represents the vertical coordinate difference of the center. This indicates the height threshold.
[0084] When aggregating scattered text blocks recognized by OCR, a spatial relationship-based merging strategy is adopted. For any two text blocks... and If and only if the Euclidean distance between their centers is equal to the distance between their centers, then the Euclidean distance between their centers is equal to the distance between their centers. Less than the distance threshold And the difference in their vertical coordinates is less than the height threshold. When the two text blocks are determined to belong to the same text line, they are merged into a single logical text block. This operation combines scattered characters (such as "surname" and "given name") into semantically meaningful words ("name"), which facilitates subsequent analysis.
[0085] Step S106: Text verification to determine the de-identification target.
[0086] The logical text blocks are validated based on their text type, text content, and semantic relationships. This validation process includes: validation of the first confirmation list, validation of the second confirmation list, and context validation. Based on the validation results, all logical text blocks are input into the list to be anonymized or the list to be displayed.
[0087] In this embodiment, the first confirmation list mechanism includes medical parameters and identifiers inherent in PACS images that must be retained, such as "kV" (kilovolt), "mA" (milliampere), "TE / TR" (magnetic resonance parameters), "FOV" (field of view), "L / R" (left / right orientation), "Slice Thickness", etc. The PACS image technical parameters in the first confirmation list do not involve sensitive information and contain necessary diagnostic information, so the characters in the first confirmation list cannot be desensitized.
[0088] The second confirmation list mechanism includes two types of sensitive word feature databases: a patient identity information feature database and a hospital name information feature database.
[0089] Patient identity information feature database: Surnames correspond to those in the Hundred Family Surnames, and given names correspond to those in the Xinhua Dictionary. It uses Chinese characters and pinyin characters and can be expanded to include common characters used in names, for quickly locating all possible name clues.
[0090] Hospital Name Information Feature Database: Contains the names, abbreviations, English names, and common abbreviations (such as "City First Hospital", "Huashan Hospital", "Hospital", "Hosp", etc.) of all hospitals in the target area.
[0091] Specifically, the decision-making process for text validation includes:
[0092] Step S1061: First confirmation list is verified first.
[0093] For each logical text block, it is first matched against the first confirmation list. If the match is successful (e.g., the text is "TE"), the logical text block is immediately marked as a non-sensitive object, and the process proceeds to add it to the displayable list.
[0094] The formula for the first confirmation list matching algorithm used in this step is:
[0095]
[0096] In the first confirmation list verification step, the system uses a protection function IsProtected(T) to evaluate each identified logical text block T. If the logical text block T contains any substring W (e.g., "kV", "mA") defined in the first confirmation list Whitelist, the function returns 1, indicating that this text is a non-sensitive medical parameter and is excluded from the desensitization area. This function definition ensures that all critical diagnostic information from PACS images is reliably preserved.
[0097] Step S1062: Quick screening of the second confirmation list
[0098] The logical text blocks that do not match the first confirmation list are matched with the second confirmation list. If a match is successful (e.g., the text contains "Li", "People's Hospital", "Hospital"), the attributes of the text content of the logical text block are determined. For logical text blocks containing hospital name attributes, there is no ambiguity, and they must be desensitized and directly added to the desensitization list. However, for logical text blocks containing patient identity attributes, they may overlap with medical parameter attributes in PACS images. For example, the character "circumference" in the labeled parameters may overlap with patients with the surname "Zhou" (there are many similar cases, which are also key factors leading to false desensitization). Therefore, text blocks containing suspected name characters will be marked as high-risk sensitive candidates and need to be finally confirmed by contextual association analysis.
[0099] Step S1063: Context Association Confirmation
[0100] For the patient name: Check whether there are strongly relevant context keywords such as "Name:", "Name:", "Patient:" etc. near the candidate words such as "Zhou" or "Zhang". As long as the text that simultaneously meets the "Match the Second Confirmation List" and "There is context association", it will be finally determined as sensitive information.
[0101] The context association analysis step uses a distance-based context association confidence algorithm:
[0102]
[0103] Among them, S is the association score, K is the keyword, Keywords is the keyword list, is the candidate logical text block, is the predefined context keyword logical text block, D is the spatial distance between the candidate logical text block and the predefined context keyword logical text block, is the preset threshold, when it is confirmed that there is context association.
[0104] For the candidate text corresponding to the patient identity information that matches the second confirmation list (such as: "Li"), perform context association analysis for final confirmation. First, calculate the candidate text block and all predefined context keywords (such as "Name:", "Name:") text blocks The spatial distance D between them is realized by calculating the spatial proximity of the candidate information and the predefined keywords. Then we define a distance-based association score function S. When there is at least one keyword such that its normalized distance score from the candidate block exceeds the preset threshold the system finally confirms that the candidate information is real sensitive information.
[0105] Step S1064: Processing of Unmatched Logical Text Blocks
[0106] For the logical text blocks that neither match the first confirmation list nor the second confirmation list, it is considered that the text does not need to be desensitized, and it is directly added to the displayable list.
[0107] Step S107: Loop and Judgment
[0108] The system processes in a loop to ensure that each logical text block is added to the list to be desensitized or the displayable list after going through the above complete text verification process.
[0109] Step S108: Perform Coding Desensitization Processing by Combining All Coordinates
[0110] Only the areas corresponding to the logical text blocks in the desensitization list are pixel-masked (mosaiced) to ensure that the key technical parameters of the target image, excluding the areas to be desensitized, remain intact.
[0111] Step S109: Generate desensitized images and send them to the user equipment.
[0112] Step S110: The terminal device records this image request and creates a medical record folder locally or in the cloud, storing the original target image and the desensitized image in the medical record folder for easy retrieval and traceability of the image in the future.
[0113] Example 2
[0114] An apparatus for desensitizing PACS images, comprising:
[0115] Image acquisition module, used to acquire raw target images;
[0116] The image preprocessing module is used to process the original target image, separate the text region from the background, locate the region of interest of the text, and output the region of interest image block corresponding to the region of interest of the text.
[0117] The text content recognition module inputs the region of interest (ROI) image block into the OCR engine for text content recognition and outputs the text type, text content, and coordinate values of each ROI image block in the original target image.
[0118] The text block aggregation module is used to merge image blocks of the region of interest into logical text blocks;
[0119] The text verification module performs first confirmation list verification, second confirmation list verification, and contextual analysis on all logical text blocks, and inputs the logical text blocks into the corresponding list to be de-identified or the list to be displayed.
[0120] The desensitization module performs masking on the coordinate areas corresponding to the logical text blocks in the desensitization list and outputs the final desensitized image.
[0121] The storage module is used to store the original target image and the desensitized image.
[0122] Example 3
[0123] An electronic device, comprising:
[0124] Memory, used to store computer programs / instructions;
[0125] A processor for executing the computer program / instructions to implement the steps of the above-described PACS image desensitization processing method.
[0126] Example 4
[0127] A computer-readable storage medium storing a computer program / instructions thereon, characterized in that, when the computer program / instructions are executed by a processor, they implement the steps of the above-described PACS image desensitization processing method.
[0128] In summary, the PACS image desensitization processing method of the present invention can achieve efficient and accurate identification and desensitization of various texts in PACS images in complex scenarios, with almost no false or missed desensitization, protecting privacy and security, and having a lower cost.
[0129] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0130] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.
[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.
[0133] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.
Claims
1. A method for desensitizing PACS images, characterized in that, Includes the following steps: The original target image is acquired and cached on the terminal. The original target image is divided using an image segmentation algorithm to separate the text region from the background, locate the region of interest of the text, and output the image block of the region of interest corresponding to the region of interest of the text. The region of interest image patch is input into the OCR engine for text content recognition, and the text type, text content and coordinate value in the original target image are output for each region of interest image patch; Based on coordinate values, region-of-interest (ROI) image blocks are merged into logical text blocks. The logical text blocks inherit the text type and text content of the individual ROI image blocks before merging, and contain the semantic relationships between the merged ROI image blocks. The logical text blocks are validated based on their text type, content, and semantic relationships. This validation process includes: validation of the first confirmation list, validation of the second confirmation list, and contextual analysis. Based on the validation results, all logical text blocks are input into the list to be desensitized or the list that can be displayed. The coordinate regions corresponding to the logical text blocks in the display list of the original target image are not processed. The coordinate regions corresponding to the logical text blocks in the list to be desensitized are masked. The final desensitized image is then output.
2. The method for desensitizing PACS images according to claim 1, characterized in that, The text verification process includes: Match the logical text block with the first confirmed list. If the first confirmed list matches successfully, add the logical text block to the displayable list. If the first confirmation list fails to match, the logical text block is matched with the second confirmation list. If the second confirmation list matches successfully and the semantic relationship is patient identity feature, the logical text block is analyzed using context association. Logical text blocks with analysis results of sensitive information are added to the list to be anonymized, and logical text blocks with analysis results of non-sensitive information are added to the list that can be displayed. If the second confirmation list matches successfully and the semantic relationship is hospital name feature, the corresponding logical text block is added to the list to be anonymized. If a logical text block does not match either the first confirmation list or the second confirmation list, then the logical text block is marked as irrelevant content and added to the displayable list. Repeat the above text validation process until all logical text blocks have been processed.
3. The method for desensitizing PACS images according to claim 1, characterized in that, The method for separating the text region from the background, locating the region of interest (ROI) of the text, and outputting the corresponding ROI image block is as follows: The Otsu binarization algorithm is used to perform adaptive threshold image segmentation on the grayscale image of the original target image to obtain the corresponding text region binarized image. The ROI image block is then obtained after secondary optimization of the binarized image. The algorithm process formula is as follows: in, The inter-class variance is used to measure the degree of separation between the foreground and background. To make the inter-class variance The biggest one value, Indicates the candidate threshold. The function that takes the maximum value. Background class Average gray level, Foreground Average gray level, This represents the global average gray level of the entire image. This indicates that the pixel is classified as the background class. The probability (i.e., the proportion of background pixels). This indicates that the pixel is classified as a foreground pixel. The probability (i.e., the proportion of text pixels) is given by L, where L is the gray level of the grayscale image.
4. The method for desensitizing PACS images according to claim 1, characterized in that, The method for merging image blocks of the region of interest into logical text blocks is as follows: Text block proximity and similarity measurement algorithm: in, and This represents two different regions of interest image patches. Image patch representing region of interest and Euclidean distance of the center point Indicates the distance threshold. This represents the vertical coordinate difference of the center. This indicates the height threshold.
5. The method for desensitizing PACS images according to claim 1, characterized in that, Context association analysis uses a distance-based context association confidence algorithm: Where S represents the relevance score, K represents the keywords, and Keywords represents the keyword list. Candidate logical text blocks, Let D be the predefined context keyword logical text block, and let D be the spatial distance between the candidate logical text block and the predefined context keyword logical text block. For the preset threshold, when At that time, confirm that there is a contextual relationship.
6. The method for desensitizing PACS images according to claim 1, characterized in that, The method further includes: after the terminal device sends the desensitized image, recording the image request, and creating a medical record folder locally or in the cloud, storing the original target image and the desensitized image in the medical record folder.
7. The method for desensitizing PACS images according to claim 1, characterized in that, The text types include: Chinese characters, numbers, letters, and special symbols.
8. A device for desensitizing PACS images, characterized in that, include: Image acquisition module, used to acquire raw target images; The image preprocessing module is used to process the original target image, separate the text region from the background, locate the region of interest of the text, and output the region of interest image block corresponding to the region of interest of the text. The text content recognition module inputs the region of interest (ROI) image block into the OCR engine for text content recognition and outputs the text type, text content, and coordinate values of each ROI image block in the original target image. The text block aggregation module is used to merge image blocks of the region of interest into logical text blocks; The text verification module performs first confirmation list verification, second confirmation list verification, and contextual analysis on all logical text blocks, and inputs the logical text blocks into the corresponding list to be de-identified or the list to be displayed. The desensitization module performs masking on the coordinate areas corresponding to the logical text blocks in the desensitization list and outputs the final desensitized image. The storage module is used to store the original target image and the desensitized image.
9. An electronic device, characterized in that, include: Memory, used to store computer programs / instructions; A processor for executing the computer program / instructions to implement the steps of the PACS image desensitization processing method according to any one of claims 1-7.
10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the PACS image desensitization processing method as described in any one of claims 1-7.