Distributed processing-based AML chemotherapy regimen sharing method and system

Abstract

The invention discloses an AML chemotherapy regimen sharing method and system based on distributed processing, and belongs to the technical field of medical information. The method comprises the following steps: acquiring clinical data related to an AML chemotherapy regimen, including patient information, pathological indexes and iconography data, performing preprocessing and semantic annotation, and generating a semantic clinical data unit with a distributed identifier; registering the identifier, the metadata and the digital signature to a distributed account book network maintained by multiple mechanisms; receiving a query request based on the semantic tag and the metadata, and matching a data unit in an account book network through a semantic engine; and after verifying the authority of the requester, transmitting a data load or executing a distributed computing task of which the data cannot be out of the domain through the secure channel. The problems of privacy security, cross-mechanism interoperation and credible authentication in clinical data sharing are effectively solved, and an efficient and reliable data collaboration platform is provided for AML diagnosis and treatment research.

Description

Technical Field

[0001] This invention belongs to the field of medical information technology, and more specifically, relates to a method and system for sharing AML chemotherapy regimens based on distributed processing. Background Technology

[0002] In today's rapidly evolving healthcare landscape, clinical data has become a core driving force for advancing medical research, optimizing treatment pathways, and achieving precision medicine. This is especially true for complex diseases such as acute myeloid leukemia (AML), where the accuracy of diagnosis, the personalization of treatment plans, and the evaluation of efficacy all heavily rely on massive amounts of high-quality, reliable clinical data. Therefore, building an efficient, secure, and deeply shared clinical data ecosystem is of immeasurable strategic and social value for improving the quality of healthcare services, accelerating new drug development, and even saving more lives. The core challenge in this technological field is how to maximize the potential of clinical data while ensuring data privacy and security, enabling its seamless flow and intelligent application across diverse healthcare scenarios.

[0003] For a long time, the industry has relied on Uniform Resource Locators (URLs) as a fundamental technology for locating and accessing network resources. The core design concept of URLs lies in their address-driven nature, that is, uniquely identifying the physical or logical location of data within the network, thereby allowing users or systems to retrieve the corresponding resources. Its working principle involves organically combining network protocols, server addresses, file paths, and other information to form a structured string that indicates the precise location of data storage. In the early stages of internet development, this location method, which focused on data address space, greatly simplified the information retrieval process, enabling users to easily access massive amounts of data resources distributed globally. At that time, the introduction of URLs successfully solved the widespread problem of scattered and difficult-to-find resources on the network, laying the technological foundation for modern information interconnection. Its contribution at the time was undeniable and epoch-making, effectively supporting the needs of basic information exchange and document sharing.

[0004] Traditional URL mechanisms, as an address-driven resource location paradigm, while solving the information accessibility problem in a specific historical period, lack the inherent ability to describe the semantic, pragmatic, and management information of the data content itself. This has led to fundamental limitations and technical bottlenecks in supporting the deep exchange, intelligent sharing, knowledge extraction, content governance, and trusted authentication of modern clinical data, especially core medical data with extremely high requirements for accuracy, timeliness, and security, such as AML chemotherapy regimens. This deep contradiction lies in the fact that as data usage patterns evolve from simple acquisition to complex understanding, evaluation, and collaboration, the original location methods can no longer meet the urgent needs for data intrinsic value and trust mechanisms. Summary of the Invention

[0005] According to a first aspect of the present invention, the present invention claims protection for a method for sharing AML chemotherapy regimens based on distributed processing, comprising the following steps: S1, acquire clinical data associated with chemotherapy regimens for acute myeloid leukemia (AML), preprocess and semantically annotate the clinical data, and generate semantic clinical data units (SCDUs) with distributed identifiers. S2, register the distributed identifier, partial metadata, and digital signature of the semantic clinical data unit (SCDU) to a distributed ledger network maintained by multiple participating institutions; S3, receive a query request from a user or system, the query request containing search conditions based on semantic tags and metadata fields, the semantic query engine matches the distributed identifier and partial metadata of the semantic clinical data unit (SCDU) in the distributed ledger network according to the search conditions; S4. After identifying a matching semantic clinical data unit (SCDU), the access rights of the requester are verified according to the access control block of the SCDU and the smart contract recorded in the distributed ledger network. If the access rights are verified, the data payload of the SCDU is transmitted to the requester through a secure channel, or a distributed computing task is executed without leaving the domain.

[0006] Further, step S1 specifically includes: The clinical data obtained includes basic patient information, pathophysiological indicators, treatment history data, and imaging data related to the AML chemotherapy regimen. The acquired clinical data is anonymized, standardized, and subjected to semantic tags based on medical ontology. Generate the semantic clinical data unit (SCDU) with a distributed identifier, wherein the distributed identifier is obtained by performing a cryptographic hash calculation on the semantically annotated clinical data content and serves as a unique and tamper-proof identifier for the SCDU, and the digital signature is obtained by signing the content hash value with the private key of the creator of the SCDU.

[0007] Furthermore, acquiring the imaging data specifically includes: Bone marrow cell images are acquired using devices such as ultrasound, MRI, or digital slide scanning. The acquired bone marrow cell images are then transferred to an image storage device, which is located within the image processing unit. The preprocessing and semantic annotation of the bone marrow cell images include the following sub-steps: S11, the bone marrow cell image is calculated by RGB three-color averaging to form the image histogram of the bone marrow cell image after three-color averaging, and then saved to the image memory; S12, count the number of pixels at each gray level in the image histogram, calculate the gray level distribution density of the image, calculate the cumulative histogram distribution, perform a rounding operation on the cumulative distribution result, perform equalization processing on the gray level values, and save the equalized image histogram of the bone marrow cell image to the image memory.

[0008] S13, calculate the local threshold of the bone marrow cell image, delete the pixels in the image histogram whose gray value is less than the local threshold to form a filtered bone marrow cell image, and save the filtered bone marrow cell image to the image memory; S14, a threshold segmentation method is used to make adjacent pixels in the target area or background area have equal gray values, and the pixels in different target areas or background areas have different gray values. Different peaks are set on the image histogram to correspond to different target areas and background areas. A threshold located at the valley between the two peaks is selected to separate the peaks, forming the bone marrow cell image after image threshold segmentation, and the bone marrow cell image after image threshold segmentation is saved to the image memory.

[0009] Furthermore, the preprocessing and semantic annotation of the bone marrow cell images also include the following sub-steps: Access coordinates and marker coordinates are set for pixels in the bone marrow cell image. When the gray value of a pixel is greater than a preset threshold, the marker coordinates are set. When the gray value of a pixel is less than a preset threshold, the image histogram recognition device also sets a marker coordinate until all non-hole pixels in the marker area are set with marker coordinates. Once all non-hole pixels have been visited and all points within the threshold have been marked, calculate the area of ​​all connected non-marked regions. A gradient extraction method is used to extract the boundaries of fat vesicles until non-internal, non-connected regions are distinguished and the boundaries of the non-internal, non-connected regions are delineated. Non-connected regions smaller than a preset threshold are stored in an array, and pixels in the array with the marker coordinates are restored to the background color; The first current point is determined on the bone marrow cell image after it has been restored to the background color. The Sobel template is used to search the bone marrow cell image after it has been restored to the background color starting from the first current point. The Sobel template uses two templates in the horizontal and vertical directions to perform weighted calculations on at least 9 points around the current point, and the sum of the absolute values ​​of the gradients in the x and y directions of the current point is obtained. The sum of the absolute values ​​obtained will be... Compare the results and select those greater than... The current point of the result value is determined as the boundary point; otherwise, the average value of the surrounding pixels is taken. For small non-connected regions in the array, if the gradient value of a pixel is greater than a preset threshold, the flag coordinates fail and the background color value is returned, thereby forming the bone marrow cell image after hole seed filling, and the bone marrow cell image after hole seed filling is saved to the image storage.

[0010] Furthermore, the preprocessing and semantic annotation of the bone marrow cell images also include the following sub-steps: Find pixels with marked coordinates but no access coordinates on the bone marrow cell image after the holes are filled with seeds, set them as points to be counted, and call a recursive function to process the points to be counted. Determine whether the point to be counted is a boundary point. If the point to be counted is not a boundary point, then increment the fat bubble area by 1 and set the point to be counted as a flag access point. The number of fat vesicles is incremented by 1 when the point to be counted is processed using a recursive function in the four directions of up, down, left, and right until a boundary point is encountered. Repeat until the analysis of bone marrow cell images after seed filling of the entire cavity is completed, calculate the diameter of the fat vesicles, and complete the statistical analysis of the fat vesicles; The results of the statistical analysis of fat vesicles, including the number of fat vesicles, total area, and average diameter, are used as part of the metadata of the semantic clinical data unit.

[0011] According to a second aspect of the present invention, the present invention claims protection for an AML chemotherapy regimen sharing system based on distributed processing, comprising: A semantic clinical data unit construction module is used to acquire clinical data related to chemotherapy regimens for acute myeloid leukemia (AML), preprocess and semantically annotate the clinical data, and generate semantic clinical data units (SCDUs) with distributed identifiers. The structure of the semantic clinical data unit (SCDU) includes a content hash value, a data payload, a metadata block, a traceability information block, an access control block, a set of semantic tags, and a digital signature. A distributed network registration and management module, connected to the semantic clinical data unit construction module, is used to register the distributed identifier, partial metadata, and digital signature of the semantic clinical data unit (SCDU) to a distributed ledger network maintained by multiple participating institutions, wherein the distributed ledger network adopts permissioned chain technology. The semantic clinical data unit query and discovery module is connected to the distributed network registration and management module. It is used to receive query requests from users or the system. The query request includes search conditions based on semantic tags and metadata fields, and matches the distributed identifier and partial metadata of the semantic clinical data unit (SCDU) in the distributed ledger network according to the search conditions. The secure access and distributed processing module, connected to the semantic clinical data unit query and discovery module, is used to verify the requester's access rights based on the access control block of the semantic clinical data unit SCDU and the smart contract recorded in the distributed ledger network after identifying a matching semantic clinical data unit SCDU. If the access rights are verified, the data payload of the semantic clinical data unit SCDU is transmitted to the requester through a secure channel, or a distributed computing task is executed without leaving the domain.

[0012] Furthermore, the semantic clinical data unit construction module includes: The data acquisition unit is used to acquire the clinical data, which includes basic patient information, pathophysiological indicators, treatment history data, and imaging data related to the AML chemotherapy regimen. The data acquisition unit is configured with an interface for receiving bone marrow cell images from an ultrasound device, an MRI device, or a digital slide scanning device, and transmitting the bone marrow cell images to an image storage device. An image processing unit, connected to the data acquisition unit, is used to process the acquired bone marrow cell image; The preprocessing and semantic annotation unit, connected to the data acquisition unit and the image processing unit, is used to anonymize, standardize, and attach semantic tags based on medical ontology to the acquired clinical data. A semantic data unit generation unit, connected to the preprocessing and semantic annotation unit, is used to generate semantic clinical data units (SCDUs) with distributed identifiers based on the clinical data content after anonymization, normalization transformation, and semantic tag appending. The distributed identifier is obtained by performing a cryptographic hash calculation on the semantically annotated clinical data content and serves as a unique and tamper-proof identifier for the SCDU. The semantic data unit generation unit is also used to digitally sign the content hash value using the private key of the creator of the SCDU to ensure data authenticity and integrity.

[0013] Furthermore, the image processing unit includes: Image storage device for storing the acquired bone marrow cell images; An image histogram recognition device, connected to the image memory, is used for: Receive the bone marrow cell image and perform image histogram conversion to form an image histogram of the bone marrow cell image; The image histogram is calculated using RGB three-color averaging to form the image histogram of the bone marrow cell image after three-color averaging, and the image histogram after three-color averaging is saved to the image memory.

[0014] Furthermore, the image histogram recognition device is also used for: The image histogram of the bone marrow cell image after tri-color averaging is obtained from the image memory. By counting the number of pixels at each gray level and setting 256 gray levels, the number of pixels corresponding to each gray value is increased by 1. The gray level distribution density is calculated, and the cumulative histogram distribution is calculated. The cumulative distribution result is rounded, and the gray values ​​are equalized. The equalized image histogram of the bone marrow cell image is saved to the image memory. The image histogram recognition device is also used for: The image histogram of the equalized bone marrow cell image is obtained from the image memory, a local threshold of the bone marrow cell image is calculated, pixels in the image histogram that are smaller than the local threshold are deleted to form a filtered bone marrow cell image, and the filtered bone marrow cell image is saved to the image memory. The filtered bone marrow cell image is obtained from the image memory. A threshold segmentation method is used to make adjacent pixels in the target region or background region have equal gray values, and the pixels in different target regions or background regions have different gray values. Different peaks are set on the image histogram to correspond to different target regions and background regions. A threshold located at the valley between the two peaks is selected to separate the peaks, forming the bone marrow cell image after image threshold segmentation. The bone marrow cell image after image threshold segmentation is then saved to the image memory. The bone marrow cell image after threshold segmentation is obtained from the image memory. Access coordinates and marker coordinates are set for the pixels in the bone marrow cell image. When the gray value of a pixel is greater than a preset threshold, the marker coordinates are set. When the gray value of a pixel is less than a preset threshold, the marker coordinates are also set until all non-hole pixels in the marker area are marked with marker coordinates. After all non-hole pixels have been visited and all points within the threshold have been marked, the area of ​​all connected non-marked regions is calculated; a gradient extraction method is used to extract fat bubble boundaries until non-connected regions outside the cell are distinguished, and the boundaries of the non-connected regions outside the cell are delineated. Non-connected regions smaller than a preset threshold are stored in an array, and pixels in the array with the marker coordinates are restored to the background color; On the bone marrow cell image restored to the background color, the first current point is determined. The Sobel template is used to search the bone marrow cell image restored to the background color starting from the first current point. The two templates in the horizontal and vertical directions are used to perform weighted calculations on at least 9 points around the current point. Finally, the sum of the absolute values ​​of the gradients in the x and y directions of the current point is obtained. The sum of the absolute values ​​obtained will be... Compare the results and select those greater than... The current point of the result value is determined as a boundary point; otherwise, the average value of the surrounding pixels is taken. For small non-connected regions in the array, if the gradient value of a pixel is greater than a preset threshold, the marker coordinates fail and its background color value is returned, forming the bone marrow cell image after hole seed filling, and the bone marrow cell image after hole seed filling is saved to the image storage.

[0015] Furthermore, the image processing unit further includes: The fat vesicle statistical analysis unit, connected to the image storage device, is used for: The image of bone marrow cells after hole seed filling is retrieved from the image memory. Pixels with set marker coordinates but no access coordinates are found on the bone marrow cell image after hole seed filling and set as points to be counted. A recursive function is then called to process the points to be counted. It is determined whether the points to be counted are boundary points. If the points to be counted are not boundary points, the area of ​​the fat vesicle is incremented by 1, and the points to be counted are set as marker access points. The points to be counted are processed using recursive functions in the four directions of up, down, left, and right until a boundary point is encountered. The number of fat vesicles is incremented by 1; this process continues until the analysis of the bone marrow cell image after the entire cavity is seeded is completed, the diameter of the fat vesicles is calculated, the statistical analysis of the fat vesicles is completed, and the result of the statistical analysis of the fat vesicles is used as the metadata part of the semantic clinical data unit (SCDU).

[0016] This invention discloses a method and system for sharing AML chemotherapy regimens based on distributed processing, belonging to the field of medical information technology. It involves acquiring clinical data related to AML chemotherapy regimens, including patient information, pathological indicators, and imaging data; preprocessing and semantically annotating these data to generate semantically labeled clinical data units with distributed identifiers; registering the identifiers, metadata, and digital signatures to a distributed ledger network maintained by multiple institutions; receiving query requests based on semantic tags and metadata; matching data units in the ledger network using a semantic engine; verifying the requester's permissions; and transmitting the data payload or executing distributed computing tasks that prevent data from leaving the domain through a secure channel. This invention effectively solves the problems of privacy and security, cross-institutional interoperability, and trusted authentication in clinical data sharing, providing an efficient and reliable data collaboration platform for AML diagnosis and treatment research. Attached Figure Description

[0017] Figure 1 The flowchart is for a distributed processing-based AML chemotherapy regimen sharing method claimed in this invention. Figure 2 This is a second flowchart of a distributed processing-based AML chemotherapy regimen sharing method claimed in this invention; Figure 3 This is a structural block diagram of an AML chemotherapy regimen sharing system based on distributed processing, for which protection is sought in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0019] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications in the embodiments of this application, such as up, down, left, right, front, back, etc., are only used to explain the relative positional relationships and movements between components in a specific orientation as shown in the accompanying drawings. If the specific orientation changes, the directional indication will change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] As healthcare informatization deepens and the demand for extracting value from clinical data continues to rise, the inherent limitations of traditional data location mechanisms, represented by URLs, are becoming increasingly apparent when dealing with increasingly complex application scenarios. This has led to a series of deeper, less obvious technical contradictions. Fundamentally, the design philosophy of URLs is "where you are," not "what you are" or "what you mean." This overemphasis on address space results in an inherent deficiency in semantic description capabilities, almost completely lacking the ability to effectively represent content space. Specifically, when clinical data needs to be exchanged and shared across institutions, systems, and contexts, simply knowing the data's storage location (URL) is far from sufficient to support reliable and intelligent applications.

[0022] This inherent characteristic at the principle level creates numerous insurmountable obstacles for traditional URL mechanisms when dealing with modern clinical data management, especially highly sensitive, complex, and valuable data such as AML chemotherapy regimens. First, the problems of low data collection efficiency and the inability to guarantee authenticity and security stem not merely from a lack of technical means, but more importantly from the fact that URLs themselves cannot inherently carry and verify the metadata of the data. A URL points only to an address; it does not contain the semantic content of the data itself, the context of its creation, modification history, or key information such as access rights. Therefore, any verification of data authenticity, security assurance, and granular control of permissions must be implemented through external systems independent of URLs and complex business logic. This undoubtedly increases system complexity and maintenance costs, and is highly susceptible to vulnerabilities and inconsistencies when integrating multiple systems, making it difficult to break down data silos and effectively trace the credibility of data sources during cross-institutional flows. Second, the lack of effective devices and methods for data screening, analysis, and organization is also a direct consequence of the lack of descriptive capabilities of URL content. Traditional URLs lack rich semantic context, making the intelligent, multi-dimensional retrieval, filtering, and aggregation of massive amounts of clinical data an extremely cumbersome and inefficient task. For example, to find patient data for all specific age groups, specific gene mutation types, and those who have received a certain AML chemotherapy regimen, URL-based systems can only traverse addresses, download content, and then process it using external semantic parsing tools. This is not only time-consuming and labor-intensive, but the accuracy and completeness of semantic matching also heavily depends on the coverage of external tools and the quality of manual annotation. This post-processing, plug-in semantic processing mechanism is essentially a compensation for the blind spots in URL content, rather than an endogenous solution. Its inherent indirectness and fragmentation greatly hinder the effective extraction of knowledge and the efficient governance of content. Finally, the inadequacy of URLs is even more apparent for higher-level needs such as copyright protection and trusted authentication. Since URLs only indicate location, they cannot embed copyright information at the data level or establish a direct link between data owners and users, nor can they provide a native, encrypted trust chain for the data. This means that once data is obtained through a URL, its subsequent flow and use are no longer subject to the original control system, creating potential risks of data abuse, infringement, and a crisis of trust.

[0023] According to the first embodiment of the present invention, referring to Figure 1 This invention claims protection for a method for sharing AML chemotherapy regimens based on distributed processing, comprising the following steps: S1, acquire clinical data associated with chemotherapy regimens for acute myeloid leukemia (AML), preprocess and semantically annotate the clinical data, and generate semantic clinical data units (SCDUs) with distributed identifiers. S2, register the distributed identifier, partial metadata, and digital signature of the semantic clinical data unit (SCDU) to a distributed ledger network maintained by multiple participating institutions; S3, receive a query request from a user or system, the query request containing search conditions based on semantic tags and metadata fields, the semantic query engine matches the distributed identifier and partial metadata of the semantic clinical data unit (SCDU) in the distributed ledger network according to the search conditions; S4. After identifying a matching semantic clinical data unit (SCDU), the access rights of the requester are verified according to the access control block of the SCDU and the smart contract recorded in the distributed ledger network. If the access rights are verified, the data payload of the SCDU is transmitted to the requester through a secure channel, or a distributed computing task is executed without leaving the domain.

[0024] In this embodiment, it further includes: In S2, the distributed identifier, partial metadata, and digital signature of the semantic clinical data unit are registered to the distributed ledger network.

[0025] The distributed ledger network consists of distributed nodes maintained by multiple participating institutions, and it employs permissioned blockchain technology. The distributed ledger network performs consensus verification on the registration information and records it on the blockchain, forming an immutable transaction record.

[0026] The distributed ledger network is also used to manage access permissions and usage logs for semantic clinical data units. Specifically, the access permissions are programmed and executed through smart contracts, and the usage logs record all queries and access operations on the semantic clinical data units, which are then uploaded to the blockchain after consensus verification. S3 receives a query request from a user or the system, the query request containing search conditions based on semantic tags and metadata fields.

[0027] The query and discovery further includes: the semantic query engine matching distributed identifiers and partial metadata of semantic clinical data units in the distributed ledger network according to the search conditions.

[0028] The semantic query engine efficiently retrieves the semantic tag set and metadata block by constructing a semantic index.

[0029] The query request may include a combination of queries for specific AML subtypes, gene mutations, treatment efficacy, or specific imaging features, such as the range of statistical analysis results for fat vesicles.

[0030] The semantic query engine performs multi-dimensional, ontology-based semantic matching to overcome the limitations of traditional keyword search. In S4, after identifying a matching semantic clinical data unit, the requester's access rights are verified based on its access control block and the smart contract recorded in the distributed ledger network.

[0031] If the access permission is verified, the secure access and distributed processing further includes: transmitting the data payload of the semantic clinical data unit to the requester through a secure channel, or performing a distributed computing task without leaving the domain.

[0032] The secure channel uses an end-to-end encryption protocol.

[0033] The distributed computing task includes, but is not limited to, privacy-preserving technologies such as federated learning and secure multi-party computation, which are used to aggregate and analyze semantic clinical data units from multiple institutions, train models, or predict trends without exposing the original clinical data, thereby generating insight reports on AML chemotherapy regimens.

[0034] The result of the distributed processing is registered as a new semantic clinical data unit to the distributed ledger network, and its traceability information block clearly records its generation process and the original semantic clinical data unit on which it depends.

[0035] Further, step S1 specifically includes: The clinical data obtained includes basic patient information, pathophysiological indicators, treatment history data, and imaging data related to the AML chemotherapy regimen. The acquired clinical data is anonymized, standardized, and subjected to semantic tags based on medical ontology. Generate the semantic clinical data unit (SCDU) with a distributed identifier, wherein the distributed identifier is obtained by performing a cryptographic hash calculation on the semantically annotated clinical data content and serves as a unique and tamper-proof identifier for the SCDU, and the digital signature is obtained by signing the content hash value with the private key of the creator of the SCDU.

[0036] Furthermore, acquiring the imaging data specifically includes: Bone marrow cell images are acquired using devices such as ultrasound, MRI, or digital slide scanning. The acquired bone marrow cell images are then transferred to an image storage device, which is located within the image processing unit. Reference Figure 2The preprocessing and semantic annotation of the bone marrow cell images include the following sub-steps: S11, the bone marrow cell image is calculated by RGB three-color averaging to form the image histogram of the bone marrow cell image after three-color averaging, and then saved to the image memory; S12, count the number of pixels at each gray level in the image histogram, calculate the gray level distribution density of the image, calculate the cumulative histogram distribution, perform a rounding operation on the cumulative distribution result, perform equalization processing on the gray level value, and save the equalized image histogram of the bone marrow cell image to the image memory; S13, calculate the local threshold of the bone marrow cell image, delete the pixels in the image histogram whose gray value is less than the local threshold to form a filtered bone marrow cell image, and save the filtered bone marrow cell image to the image memory; S14, a threshold segmentation method is used to make adjacent pixels in the target area or background area have equal gray values, and the pixels in different target areas or background areas have different gray values. Different peaks are set on the image histogram to correspond to different target areas and background areas. A threshold located at the valley between the two peaks is selected to separate the peaks, forming the bone marrow cell image after image threshold segmentation, and the bone marrow cell image after image threshold segmentation is saved to the image memory.

[0037] In this embodiment, it further includes: S11 transmits the original bone marrow cell image from the image storage device in the form of a digital image file, such as TIFF, PNG, or DICOM, to the image histogram recognition device. After receiving the image data, the image histogram recognition device first performs image histogram conversion. This conversion process generates a grayscale distribution map of the image by statistically analyzing the frequency of occurrence of the grayscale value of each pixel in the image. Specifically, for a color bone marrow cell image, the image histogram recognition device calculates the grayscale histograms of the three color channels: red (R), green (G), and blue (B). Further, to simplify processing and extract overall brightness information, the image histogram recognition device uses RGB three-color averaging to calculate the image histogram. This process performs an arithmetic mean of the R, G, and B component values ​​of each pixel to obtain an average grayscale value, and then recalculates the grayscale histogram of the image based on these average grayscale values, forming the image histogram of the bone marrow cell image after three-color averaging. The image histogram after three-color averaging is typically represented as an array of 256 bins, each bin corresponding to a gray level 0-255, and storing the number of pixels that gray level appears in the image. It is then saved to the image memory for subsequent processing.

[0038] S12 involves histogram equalization. The image histogram recognition device retrieves the image histogram of the bone marrow cell image after tri-color averaging from the image memory. To enhance image contrast, especially for low-contrast areas that may exist in the bone marrow cell image, the image histogram recognition device counts the number of pixels at each gray level in the image histogram. Using a standard set of 256 gray levels, the image histogram recognition device initializes a counter array and increments the number of pixels corresponding to each gray value by 1, thus obtaining the precise number of pixels at each gray level. Next, the image histogram recognition device calculates the gray-level distribution density of the image, i.e., the proportion of pixels at each gray level to the total number of pixels. Based on the PDF, a cumulative histogram distribution is calculated, i.e., the cumulative proportion of all pixels from gray level 0 to the current gray level. To equalize the gray values, the image histogram recognition device performs a rounding operation on the cumulative distribution result, typically using a linear transformation or non-linear mapping, mapping the gray values ​​of the original image to the new equalized gray values. This mapping process makes the grayscale distribution of the image more uniform, thereby improving the overall contrast of the image. The image histogram of the equalized bone marrow cell image is then saved to the image memory.

[0039] S13 involves image filtering. The image histogram recognition device retrieves the histogram-equalized bone marrow cell image from the image memory. To remove noise and smooth the image, the image histogram recognition device first calculates a local threshold for the bone marrow cell image. The local threshold calculation method can employ adaptive thresholding algorithms such as Niblack, Sauvola, or Otsu, which determine the optimal threshold for each pixel based on the statistical characteristics of local image regions.

[0040] After obtaining a local threshold, the image histogram recognition device identifies pixels in the image histogram with gray values ​​less than the local threshold as noise or background, and deletes or sets them as the background color, thereby forming a filtered bone marrow cell image. This filtering process effectively reduces artifacts and unnecessary details in the image, providing a clearer foundation for subsequent image segmentation. The filtered bone marrow cell image is then saved to the image memory.

[0041] S14 is image thresholding. The image histogram recognition device acquires a filtered bone marrow cell image from the image memory. The goal of this sub-step is to divide the image into different regions, such as cell regions, fat vesicle regions, and background regions. The image histogram recognition device employs a thresholding method, the core idea of ​​which is to distinguish between the target and the background based on the grayscale values ​​of pixels. By selecting one or more appropriate thresholds, adjacent pixels within the target or background regions have equal grayscale values ​​or are in similar grayscale ranges, and pixels in different target or background regions exhibit significant differences in grayscale. Specifically, the image histogram recognition device analyzes the peak-valley structure of the image histogram, setting different peaks to correspond to different target and background regions. For example, the cell nucleus may correspond to one grayscale peak, the cytoplasm to another, the fat vesicle region to a third peak, and the background region to the lowest grayscale peak. The image histogram recognition device selects the grayscale values ​​located at the valleys of two peaks as thresholds to separate the peaks. For example, a bimodal or iterative method can be used to automatically determine these thresholds. This results in a thresholded image of the bone marrow cells, in which different regions have been clearly defined. The thresholded image of the bone marrow cells is then saved to the image storage.

[0042] Furthermore, the preprocessing and semantic annotation of the bone marrow cell images also include the following sub-steps: Access coordinates and marker coordinates are set for pixels in the bone marrow cell image. When the gray value of a pixel is greater than a preset threshold, the marker coordinates are set. When the gray value of a pixel is less than a preset threshold, the image histogram recognition device also sets a marker coordinate until all non-hole pixels in the marker area are set with marker coordinates. Once all non-hole pixels have been visited and all points within the threshold have been marked, calculate the area of ​​all connected non-marked regions. A gradient extraction method is used to extract the boundaries of fat vesicles until non-internal, non-connected regions are distinguished and the boundaries of the non-internal, non-connected regions are delineated. Non-connected regions smaller than a preset threshold are stored in an array, and pixels in the array with the marker coordinates are restored to the background color; The first current point is determined on the bone marrow cell image after it has been restored to the background color. The Sobel template is used to search the bone marrow cell image after it has been restored to the background color starting from the first current point. The Sobel template uses two templates in the horizontal and vertical directions to perform weighted calculations on at least 9 points around the current point, and the sum of the absolute values ​​of the gradients in the x and y directions of the current point is obtained. The sum of the absolute values ​​obtained will be... Compare the results and select those greater than... The current point of the result value is determined as the boundary point; otherwise, the average value of the surrounding pixels is taken. For small non-connected regions in the array, if the gradient value of a pixel is greater than a preset threshold, the flag coordinates fail and the background color value is returned, thereby forming the bone marrow cell image after hole seed filling, and the bone marrow cell image after hole seed filling is saved to the image storage.

[0043] Furthermore, the preprocessing and semantic annotation of the bone marrow cell images also include the following sub-steps: Find pixels with marked coordinates but no access coordinates on the bone marrow cell image after the holes are filled with seeds, set them as points to be counted, and call a recursive function to process the points to be counted. Determine whether the point to be counted is a boundary point. If the point to be counted is not a boundary point, then increment the fat bubble area by 1 and set the point to be counted as a flag access point. The number of fat vesicles is incremented by 1 when the point to be counted is processed using a recursive function in the four directions of up, down, left, and right until a boundary point is encountered. Repeat until the analysis of bone marrow cell images after seed filling of the entire cavity is completed, calculate the diameter of the fat vesicles, and complete the statistical analysis of the fat vesicles; The results of the statistical analysis of fat vesicles, including the number of fat vesicles, total area, and average diameter, are used as part of the metadata of the semantic clinical data unit.

[0044] According to a second embodiment of the present invention, referring to Figure 3 This invention claims protection for a distributed processing-based AML chemotherapy regimen sharing system, comprising: A semantic clinical data unit construction module is used to acquire clinical data related to chemotherapy regimens for acute myeloid leukemia (AML), preprocess and semantically annotate the clinical data, and generate semantic clinical data units (SCDUs) with distributed identifiers. The structure of the semantic clinical data unit (SCDU) includes a content hash value, a data payload, a metadata block, a traceability information block, an access control block, a set of semantic tags, and a digital signature. A distributed network registration and management module, connected to the semantic clinical data unit construction module, is used to register the distributed identifier, partial metadata, and digital signature of the semantic clinical data unit (SCDU) to a distributed ledger network maintained by multiple participating institutions, wherein the distributed ledger network adopts permissioned chain technology. The semantic clinical data unit query and discovery module is connected to the distributed network registration and management module. It is used to receive query requests from users or the system. The query request includes search conditions based on semantic tags and metadata fields, and matches the distributed identifier and partial metadata of the semantic clinical data unit (SCDU) in the distributed ledger network according to the search conditions. The secure access and distributed processing module, connected to the semantic clinical data unit query and discovery module, is used to verify the requester's access rights based on the access control block of the semantic clinical data unit SCDU and the smart contract recorded in the distributed ledger network after identifying a matching semantic clinical data unit SCDU. If the access rights are verified, the data payload of the semantic clinical data unit SCDU is transmitted to the requester through a secure channel, or a distributed computing task is executed without leaving the domain.

[0045] Furthermore, the semantic clinical data unit construction module includes: The data acquisition unit is used to acquire the clinical data, which includes basic patient information, pathophysiological indicators, treatment history data, and imaging data related to the AML chemotherapy regimen. The data acquisition unit is configured with an interface for receiving bone marrow cell images from an ultrasound device, an MRI device, or a digital slide scanning device, and transmitting the bone marrow cell images to an image storage device. An image processing unit, connected to the data acquisition unit, is used to process the acquired bone marrow cell image; The preprocessing and semantic annotation unit, connected to the data acquisition unit and the image processing unit, is used to anonymize, standardize, and attach semantic tags based on medical ontology to the acquired clinical data. A semantic data unit generation unit, connected to the preprocessing and semantic annotation unit, is used to generate semantic clinical data units (SCDUs) with distributed identifiers based on the clinical data content after anonymization, normalization transformation, and semantic tag appending. The distributed identifier is obtained by performing a cryptographic hash calculation on the semantically annotated clinical data content and serves as a unique and tamper-proof identifier for the SCDU. The semantic data unit generation unit is also used to digitally sign the content hash value using the private key of the creator of the SCDU to ensure data authenticity and integrity.

[0046] Furthermore, the image processing unit includes: Image storage device for storing the acquired bone marrow cell images; An image histogram recognition device, connected to the image memory, is used for: Receive the bone marrow cell image and perform image histogram conversion to form an image histogram of the bone marrow cell image; The image histogram is calculated using RGB three-color averaging to form the image histogram of the bone marrow cell image after three-color averaging, and the image histogram after three-color averaging is saved to the image memory.

[0047] Furthermore, the image histogram recognition device is also used for: The image histogram of the bone marrow cell image after tri-color averaging is obtained from the image memory. By counting the number of pixels at each gray level and setting 256 gray levels, the number of pixels corresponding to each gray value is increased by 1. The gray level distribution density is calculated, and the cumulative histogram distribution is calculated. The cumulative distribution result is rounded, and the gray values ​​are equalized. The equalized image histogram of the bone marrow cell image is saved to the image memory. The image histogram recognition device is also used for: The image histogram of the equalized bone marrow cell image is obtained from the image memory, a local threshold of the bone marrow cell image is calculated, pixels in the image histogram that are smaller than the local threshold are deleted to form a filtered bone marrow cell image, and the filtered bone marrow cell image is saved to the image memory. The filtered bone marrow cell image is obtained from the image memory. A threshold segmentation method is used to make adjacent pixels in the target region or background region have equal gray values, and the pixels in different target regions or background regions have different gray values. Different peaks are set on the image histogram to correspond to different target regions and background regions. A threshold located at the valley between the two peaks is selected to separate the peaks, forming the bone marrow cell image after image threshold segmentation. The bone marrow cell image after image threshold segmentation is then saved to the image memory. The bone marrow cell image after threshold segmentation is obtained from the image memory. Access coordinates and marker coordinates are set for the pixels in the bone marrow cell image. When the gray value of a pixel is greater than a preset threshold, the marker coordinates are set. When the gray value of a pixel is less than a preset threshold, the marker coordinates are also set until all non-hole pixels in the marker area are marked with marker coordinates. After all non-hole pixels have been visited and all points within the threshold have been marked, the area of ​​all connected non-marked regions is calculated; a gradient extraction method is used to extract fat bubble boundaries until non-connected regions outside the cell are distinguished, and the boundaries of the non-connected regions outside the cell are delineated. Non-connected regions smaller than a preset threshold are stored in an array, and pixels in the array with the marker coordinates are restored to the background color; On the bone marrow cell image restored to the background color, the first current point is determined. The Sobel template is used to search the bone marrow cell image restored to the background color starting from the first current point. The two templates in the horizontal and vertical directions are used to perform weighted calculations on at least 9 points around the current point. Finally, the sum of the absolute values ​​of the gradients in the x and y directions of the current point is obtained. The sum of the absolute values ​​obtained will be... Compare the results and select those greater than... The current point of the result value is determined as a boundary point; otherwise, the average value of the surrounding pixels is taken. For small non-connected regions in the array, if the gradient value of a pixel is greater than a preset threshold, the marker coordinates fail and its background color value is returned, forming the bone marrow cell image after hole seed filling, and the bone marrow cell image after hole seed filling is saved to the image storage.

[0048] Furthermore, the image processing unit further includes: The fat vesicle statistical analysis unit, connected to the image storage device, is used for: The image of bone marrow cells after hole seed filling is retrieved from the image memory. Pixels with set marker coordinates but no access coordinates are found on the bone marrow cell image after hole seed filling and set as points to be counted. A recursive function is then called to process the points to be counted. It is determined whether the points to be counted are boundary points. If the points to be counted are not boundary points, the area of ​​the fat vesicle is incremented by 1, and the points to be counted are set as marker access points. The points to be counted are processed using recursive functions in the four directions of up, down, left, and right until a boundary point is encountered. The number of fat vesicles is incremented by 1; this process continues until the analysis of the bone marrow cell image after the entire cavity is seeded is completed, the diameter of the fat vesicles is calculated, the statistical analysis of the fat vesicles is completed, and the result of the statistical analysis of the fat vesicles is used as the metadata part of the semantic clinical data unit (SCDU).

[0049] In one specific embodiment, the present invention is applied to a project to build a predictive model for the treatment efficacy of AML chemotherapy regimens, involving three large hospitals: Hospital A, Hospital B, and Hospital C. The project aims to construct a federated learning model based on real-world data that can predict whether AML patients will achieve complete or incomplete remission with a specific chemotherapy regimen, such as daunorubicin + cytarabine (DA), while ensuring strict privacy protection for patient data from each hospital.

[0050] First, regarding the construction of SCDUs, each of the three hospitals deployed the semantic clinical data unit construction module of this invention. Taking Hospital A as an example, its data acquisition unit obtains patient clinical data from the Electronic Medical Record System (EHR), Laboratory Information System (LIS), and Pathology Image Management System. For bone marrow cell images, such as patient bone marrow biopsy slide images obtained through a digital slide scanning device, the images are transmitted to an image storage device. The image processing unit performs the following precise operations on these images: Image histogram calculation: The original DICOM image is converted into an 8-bit RGB image. The image histogram recognition device calculates the RGB three-channel histogram, and then converts the color image into a grayscale image using the formula Gray=0.299R+0.587G+0.114B, and calculates its 256-level grayscale histogram.

[0051] For example, the grayscale histogram of a particular image shows that there are fewer pixels between grayscale levels [0,50] and [200,255], while the number of pixels is concentrated between [80,150].

[0052] The cumulative distribution function (CDF) is calculated based on the gray-level histogram. This maps the image's gray values ​​to a new uniform distribution, enhancing the overall contrast of the image. After equalization, pixels that were originally concentrated in the [80, 150] range are expanded to a wider gray-level range, such as [40, 200].

[0053] An adaptive Gaussian filter is employed, using a 3x3 Gaussian kernel to smooth the equalized image, removing high-frequency noise while preserving edge information. Local thresholding is calculated using the Sauvola method, with a window size of 51x51 pixels, a k value of 0.3, and an R value of 128, to determine the binarization threshold for each pixel.

[0054] Based on the filtered image, a multi-threshold segmentation strategy based on the maximum entropy method is used to identify bone marrow cell regions, fat vesicle regions, and background regions. For example, by setting two thresholds T1=60 and T2=180, gray values ​​[0,T1] are identified as background, [T1+1,T2] as cell regions, and [T2+1,255] as fat vesicles or bright areas.

[0055] The fat vesicle statistical analysis unit identifies all connected fat vesicle regions, calculates the area of ​​each fat vesicle, and converts it to micrometers based on pixel size. 2 Ultimately, metadata is generated for each patient, including quantitative indicators such as the number of fat vesicles, the total area of ​​fat vesicles, and the average diameter of fat vesicles. For example, the number of fat vesicles is 120, and the total area is 35,000 micrometers. 2 Average diameter: 10.5 micrometers.

[0056] The preprocessing and semantic annotation unit anonymizes and standardizes patient basic information, pathophysiological indicators, treatment plans, and treatment effects, and adds SNOMEDCT and LOINC semantic tags.

[0057] The semantic data unit generation unit encapsulates the above data and metadata into an SCDU, calculates the SHA-256 hash as a distributed identifier, and digitally signs it using the private key of Hospital A.

[0058] For distributed network registration and management, the three hospitals registered the distributed identifiers, partial metadata (such as patient anonymous IDs, AML subtypes, major gene mutations, treatment plan types, treatment effects, and digital signatures) of their constructed SCDUs into a permissioned blockchain network built on Hyperledger Fabric. Orderers in the network sort transactions, while peers verify and submit them to the ledger. A smart contract named 'access_control_contract' is deployed on-chain, defining data access permissions: federated learning model trainers can only access aggregated metadata such as the patient's anonymized ID, AML subtype, gene mutations, fat vesicle statistics, treatment plan, and treatment effects; they cannot access raw images or personally identifiable information. Any query and computation request for the SCDU is logged in the usage log by 'access_control_contract' and uploaded to the blockchain.

[0059] For the querying and discovery of SCDUs, project researchers submit query requests through a query request receiving unit: query all SCDUs of AMLM2 patients who are FLT3-ITD positive, have received DA regimen chemotherapy, and have an average bone marrow fat vesicle diameter of less than 12 micrometers. The semantic query engine parses this request, using medical ontology for semantic expansion and matching, and efficiently retrieves matching SCDU distributed identifiers and metadata on a distributed ledger. For example, if FO-AMO defines the association between high bone marrow fat vesicles and an average diameter greater than 12 micrometers, the query can be automatically expanded to include more relevant SCDUs.

[0060] Regarding secure access and distributed processing, the permission verification unit verifies the researcher's permissions based on "access_control_contract" to confirm that they can only access anonymized aggregate metadata used for federated learning. Subsequently, a federated learning task is initiated. Distributed computing units in each hospital load the matching SCDU locally, extract the required features and labels, and train a prediction model based on logistic regression or a neural network. After training, each hospital only transmits the model's weight parameters to a central aggregation server via a secure data transmission unit. The aggregation server uses a federated averaging algorithm to aggregate these model weights, generating a global model. This process is iterated until the model converges.

[0061] The final global model parameters are generated as a new SCDU, whose traceability information block records the identifiers of all original SCDUs that participated in federated learning and is registered to the distributed ledger network.

[0062] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0063] Furthermore, the functional units in the various embodiments of this 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 integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0064] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for sharing AML chemotherapy regimens based on distributed processing, characterized by, The method comprises the following steps: S1, obtaining clinical data associated with acute myeloid leukemia (AML) chemotherapy regimens, preprocessing and semantically annotating the clinical data, and generating a semantic clinical data unit (SCDU) with a distributed identifier; S2, registering the distributed identifier, part of the metadata, and the digital signature of the semantic clinical data unit (SCDU) to a distributed ledger network maintained by multiple participating institutions; S3, receiving a query request from a user or a system, the query request containing retrieval conditions based on semantic tags and metadata fields, and a semantic query engine matching the distributed identifier and part of the metadata of the semantic clinical data unit (SCDU) in the distributed ledger network according to the retrieval conditions; S4, after identifying the matched semantic clinical data unit (SCDU), verifying the access rights of the requester according to the access control block of the semantic clinical data unit (SCDU) and the smart contract recorded in the distributed ledger network, and if the access rights are verified, transmitting the data payload of the semantic clinical data unit (SCDU) to the requester through a secure channel, or performing a distributed computing task under the premise that the data does not leave the domain.

2. The method of claim 1, wherein, The step S1 specifically comprises: Obtaining the clinical data, which includes patient basic information, pathophysiological indicators, treatment history data, and imaging data related to AML chemotherapy regimens; Anonymizing the obtained clinical data, standardizing the conversion, and adding semantic tags based on medical ontology; Generating the semantic clinical data unit (SCDU) with a distributed identifier, wherein the distributed identifier is obtained by encrypting and hashing the semantically annotated clinical data content, and serves as the unique and tamper-proof identifier of the semantic clinical data unit (SCDU), and the digital signature is obtained by signing the content hash value with the creator's private key of the semantic clinical data unit (SCDU).

3. The method of claim 2, wherein, Obtaining the imaging data specifically comprises: Obtaining bone marrow cell images, which are obtained by devices such as B-ultrasound equipment, nuclear magnetic resonance equipment, or digital slice scanning equipment, and are transmitted to an image storage configured inside an image processing unit; The preprocessing and semantic annotation of the bone marrow cell images comprise the following sub-steps: S11, performing image histogram calculation on the bone marrow cell images by RGB three-color averaging method, forming the image histogram of the three-color averaged bone marrow cell images, and saving them to the image storage; S12, counting the number of pixels at each gray level in the image histogram, calculating the gray distribution density of the image, and calculating the cumulative histogram distribution, performing rounding operation on the cumulative distribution result, and performing equalization processing on the gray value, saving the equalized image histogram of the bone marrow cell images to the image storage; S13, calculate a local threshold value of the bone marrow cell image, delete pixel points with a gray value less than the local threshold value in the image histogram, form a filtered bone marrow cell image, and save the filtered bone marrow cell image to the image storage; S14, adopt a threshold segmentation method, so that adjacent pixel points in a target region or a background region have equal gray values, and pixel points in different target regions or background regions are not equal in gray, set different peaks on the image histogram to correspond to different target regions and background regions, select a threshold value located at a valley between two peaks to separate each peak, form a bone marrow cell image after threshold segmentation of the image, and save the bone marrow cell image after threshold segmentation of the image to the image storage.

4. The method of claim 3, wherein, For the preprocessing and semantic labeling of the bone marrow cell image, the following sub-steps are further included: Set an access coordinate and a mark coordinate for a pixel point in the bone marrow cell image, and set the mark coordinate when the pixel point gray value is greater than a preset threshold value; When the pixel point gray value is less than the preset threshold value, the image histogram recognition device also sets the mark coordinate, until all non-hole pixel points in the mark region are set with the mark coordinate; When all non-hole pixel points are accessed and all points within the threshold value are marked, the area of all connected non-mark regions is counted; A gradient is used to extract the boundary of a fat bubble, until the non-connected region inside the non-cell is distinguished, and the boundary of the non-connected region inside the non-cell is divided out; The non-connected region less than the preset threshold value is stored in an array, and the pixel with the mark coordinate in the array is restored to a background color; Determine a first current point on the bone marrow cell image restored to the background color, and search the bone marrow cell image restored to the background color from the first current point using a Sobel template, the Sobel template uses two templates in the horizontal and vertical directions to perform weighted calculation on at least 9 points around the current point, and the sum of the absolute values of the x and y direction gradients of the current point is accumulated; The sum of the absolute values is compared with the result value of The current point is determined as a boundary point if it is greater than the average value of the surrounding pixels otherwise. For the small non-connected region in the array, if the gradient value of the pixel is greater than the preset threshold value, the mark coordinate fails, and the background color value is returned, thereby forming a bone marrow cell image after hole seed filling, and saving the bone marrow cell image after hole seed filling to the image storage.

5. The method of claim 4, wherein, For the preprocessing and semantic labeling of the bone marrow cell image, the following sub-steps are further included: Find a pixel point that has been set with a mark coordinate but has not been set with an access coordinate on the bone marrow cell image after hole seed filling, and set it as a point to be counted, and call a recursive function to process the point to be counted; Determine whether the point to be counted is a boundary point, if the point to be counted is not a boundary point, the fat bubble area is increased by 1, and the point to be counted is set as a mark access point; Use a recursive function to process the point to be counted in the up, down, left and right four directions until a boundary point is encountered, and the fat bubble number is increased by 1; The analysis of the marrow cell images after filling the entire hole seed is repeated until the diameter of the fat bubble is calculated and the fat bubble statistical analysis is completed; The results of the fat bubble statistical analysis include the number of fat bubbles, total area, and average diameter, which are part of the metadata of the semantic clinical data unit.

6. A distributed processing based AML chemotherapy regimen sharing system, characterized by, Comprise: A semantic clinical data unit construction module is configured to obtain clinical data related to an acute myeloid leukemia (AML) chemotherapy regimen, pre-process and semantically label the clinical data, and generate a semantic clinical data unit (SCDU) with a distributed identifier, wherein the structure of the semantic clinical data unit (SCDU) includes a content hash value, a data payload, a metadata block, a provenance information block, an access control block, a set of semantic tags, and a digital signature; A distributed network registration and management module is connected to the semantic clinical data unit construction module and is configured to register the distributed identifier, part of the metadata, and the digital signature of the semantic clinical data unit (SCDU) to a distributed ledger network maintained by multiple participating institutions, wherein the distributed ledger network adopts a permissioned chain technology; A semantic clinical data unit query and discovery module is connected to the distributed network registration and management module and is configured to receive a query request from a user or a system, wherein the query request includes a search condition based on semantic tags and metadata fields, and to match the distributed identifier and part of the metadata of the semantic clinical data unit (SCDU) in the distributed ledger network according to the search condition; A secure access and distributed processing module is connected to the semantic clinical data unit query and discovery module and is configured to verify the access authority of a requestor according to the access control block of the semantic clinical data unit (SCDU) and a smart contract recorded in the distributed ledger network after identifying a matched semantic clinical data unit (SCDU); if the access authority is verified, the data payload of the semantic clinical data unit (SCDU) is transmitted to the requestor through a secure channel, or a distributed computing task is executed under the premise that the data does not leave the domain.

7. The system of claim 6, wherein, The semantic clinical data unit construction module comprises: A data acquisition unit is configured to acquire the clinical data, wherein the clinical data includes patient basic information, pathophysiological indicators, treatment history data, and imaging data related to the AML chemotherapy regimen, and the data acquisition unit is configured with an interface for receiving marrow cell images from a B-ultrasound device, a nuclear magnetic resonance device, or a digital slice scanning device and transmitting the marrow cell images to an image storage; An image processing unit is connected to the data acquisition unit and is configured to process the acquired marrow cell images; A pre-processing and semantic labeling unit is connected to the data acquisition unit and the image processing unit and is configured to anonymize, standardize, and add semantic tags based on medical ontology to the acquired clinical data. A semantic data unit generation unit, connected with the preprocessing and semantic labeling unit, is configured to generate a semantic clinical data unit SCDU with a distributed identifier according to the clinical data content after the anonymization processing, the standardization conversion and the semantic label addition, wherein the distributed identifier is obtained by performing an encryption hash calculation on the clinical data content after the semantic labeling, and serves as a unique and tamper-proof identifier of the semantic clinical data unit SCDU; the semantic data unit generation unit is further configured to perform a digital signature on the content hash value by using a private key of a creator of the semantic clinical data unit SCDU, so as to ensure data authenticity and integrity.

8. The system of claim 7, wherein, The image processing unit comprises: An image storage unit configured to store the obtained bone marrow cell image; An image histogram identification device connected with the image storage unit and configured to: receive the bone marrow cell image and perform image histogram conversion to form an image histogram of the bone marrow cell image; calculate the image histogram by using an RGB three-color averaging method to form an image histogram of the bone marrow cell image after three-color averaging, and save the image histogram after three-color averaging to the image storage unit.

9. The system of claim 8, wherein, The image histogram identification device is further configured to: obtain the image histogram of the bone marrow cell image after three-color averaging from the image storage unit, count the number of pixels of each gray level and set 256 gray levels, increase the number of pixel points corresponding to each gray value by 1, calculate the gray distribution density, and calculate the cumulative histogram distribution; perform an integer operation on the cumulative distribution result and perform equalization processing on the gray value, and save the image histogram of the equalized bone marrow cell image to the image storage unit; The image histogram identification device is further configured to: obtain the image histogram of the equalized bone marrow cell image from the image storage unit, calculate the local threshold value of the bone marrow cell image, delete the pixel points of the image histogram smaller than the local threshold value to form a filtered bone marrow cell image, and save the filtered bone marrow cell image to the image storage unit; obtain the filtered bone marrow cell image from the image storage unit, use a threshold segmentation method to make adjacent pixel points in a target region or a background region have equal gray values, and different pixels in different target regions or background regions are not equal in gray, set different peaks on the image histogram to correspond to different target regions and background regions, select a threshold value located at the valley between two peaks to separate each peak, form the bone marrow cell image after image threshold segmentation, and save the bone marrow cell image after image threshold segmentation to the image storage unit; obtain the bone marrow cell image after image threshold segmentation from the image storage unit, set access coordinates and flag coordinates for the pixel points in the bone marrow cell image, set the flag coordinates when the pixel point gray value is greater than a preset threshold value, and also set the flag coordinates when the pixel point gray value is less than the preset threshold value, until all non-hole pixel points in the flag region are set with flag coordinates; When all non-hole pixel points are accessed and all points within the threshold are marked, the area of all connected non-marked regions is counted; The method of extracting the boundary of fat bubble by gradient is adopted until the non-connected region inside the non-cell is distinguished, and the boundary of the non-connected region inside the non-cell is divided out; The non-connected region smaller than the preset threshold is stored in an array, and the pixel with the marked coordinate in the array is restored to the background color; A first current point is determined on the marrow cell image restored to the background color, the marrow cell image restored to the background color is searched from the first current point using a Sobel template, two templates in horizontal and vertical directions are used to perform weighted calculation on at least 9 points around the current point, and finally the sum of the gradient absolute values of the current point in x and y directions is obtained by accumulation; The sum of the absolute values is compared with the result value of The current point is determined as a boundary point if the result value is greater than Otherwise, the average value of the surrounding pixels is taken; for small non-connected regions in the array, if the gradient value of a pixel is greater than a preset threshold, the flag coordinates fail, and the background color value is returned, forming the bone marrow cell image after hole seed filling, and saving the bone marrow cell image after hole seed filling to the image storage.

10. The system of claim 9, wherein, The image processing unit further comprises: The fat bubble statistical analysis unit is connected with the image storage and used for: The marrow cell image after hole seed filling in the image storage is called, pixel points with marked coordinates but without access coordinates on the marrow cell image after hole seed filling are found and set as to-be-counted points, and a recursive function is called to process the to-be-counted points; whether the to-be-counted points are boundary points is judged, if the to-be-counted points are not boundary points, the fat bubble area is added by 1, and the to-be-counted points are set as marked access points; Recursive functions are used to process the to-be-counted points in up, down, left and right directions until boundary points are encountered; The fat bubble number is added by 1; this process continues until the analysis of the whole marrow cell image after hole seed filling is completed, the diameter of the fat bubble is calculated, the fat bubble statistical analysis is completed, and the result of the fat bubble statistical analysis is taken as the metadata part of the semantic clinical data unit SCDU.

Images