Intelligent blood disease diagnosis and treatment data processing system

Abstract

The invention relates to the technical field of medical informatization, and particularly discloses a blood disease intelligent diagnosis and treatment data processing system which adopts an image processing technology based on deep learning to perform cell detection on a bone marrow morphological image of a patient object so as to focus on a cell target in the image. The method comprises the following steps: acquiring a cell target, performing multi-level feature extraction on the cell target to capture superficial detail features and trunk features of the cell, and performing significant sparse interactive coding on the superficial detail features and the trunk features to obtain cell feature representation with higher distinction degree, thereby realizing intelligent identification of abnormal cells. Through the mode, the cell characteristics in the blood sample can be automatically identified and analyzed, and rapid detection and diagnosis of abnormal cells are realized, so that a doctor can be assisted to carry out faster and more accurate blood disease diagnosis, and the workload of the doctor is relieved.

Description

Technical Field

[0001] The present application relates to the field of medical information technology, and more specifically, to a blood disease intelligent diagnosis and treatment data processing system. Background Art

[0002] With the development of information technology, the medical and health field is undergoing unprecedented changes. Especially in the diagnosis of blood diseases, although the traditional manual microscope examination method can provide detailed cell morphology information, doctors need to make summary judgments based on the collection of multiple types of knowledge and data, which makes clinicians spend a lot of time and energy in finding cases, reviewing medical history, clarifying treatment processes, and formulating the best treatment plan. This leads to a large workload for doctors, increased repetitive work, and the inability to accurately recommend a matching treatment plan.

[0003] In this regard, the invention patent with publication number CN115295150A discloses an intelligent blood disease diagnosis and treatment assistance system, which has the function of storing and querying patient information. By establishing a database and cloud platform, it supports doctors to quickly obtain patients' medical history, examination data, past treatment plans and other related information through their effective identity authentication, thereby reducing the time and energy spent by clinicians in finding cases, reviewing medical history, and clarifying treatment processes.

[0004] However, the above-mentioned intelligent blood disease diagnosis and treatment assistance system mainly focuses on the integration and query of patients' medical information. For the core link of blood disease diagnosis - cell morphology analysis, it still needs to rely on the doctor's professional knowledge and experience, and cannot achieve fully automated data processing, thus limiting the efficiency of diagnosis.

[0005] Therefore, we look forward to an optimized intelligent blood disease diagnosis and treatment data processing system that can automate cell morphology analysis to reduce the workload of doctors and improve the accuracy and efficiency of diagnosis. Summary of the invention

[0006] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides a data processing system for intelligent diagnosis and treatment of blood diseases, which uses deep learning-based image processing technology to perform cell detection on the bone marrow morphology image of the patient subject to focus on the cell target in the image, and extracts multi-level features of the cell target to capture the shallow detail features and trunk features of the cell, and then performs significant sparse interactive encoding on the two to obtain a more discriminative cell feature representation, thereby realizing intelligent identification of abnormal cells. In this way, the cell features in the blood sample can be automatically identified and analyzed, and the rapid detection and diagnosis of abnormal cells can be realized, thereby assisting doctors in diagnosing blood diseases more quickly and accurately, reducing the workload of doctors.

[0007] Accordingly, according to one aspect of the present application, a blood disease intelligent diagnosis and treatment data processing system is provided, which includes: a database and a cloud platform, the data in the database can be uploaded to the cloud platform, the data in the cloud platform is accessible to qualified hospital terminals, the database includes a patient information module, the patient information module includes a basic information module and an examination information module, the basic information module is used to store the patient's previous medical records in hospitals at all levels, the examination information module includes a routine examination module and a blood disease examination module, the routine examination module is used to store routine examination documents, the blood disease examination module is used to store patients' blood disease examination documents, doctors can access the patient information module through the patient's account number and password, wherein the cloud platform also includes a diagnosis and treatment data processing module, which is used to analyze the patient's bone marrow morphology image to automatically identify abnormal cells.

[0008] In the above-mentioned intelligent diagnosis and treatment data processing system for hematological diseases, the diagnosis and treatment data processing module includes: a bone marrow morphology image acquisition unit, which is used to extract the bone marrow morphology image of the patient object from the hematological disease examination module; a cell target focusing unit, which is used to perform cell target detection on the bone marrow morphology image to obtain a cell target focused image; a multi-level feature extraction unit, which is used to perform multi-level feature extraction on the cell target focused image to obtain a cell target shallow feature map and a cell target trunk feature map; a feature sparse interaction unit, which is used to perform significant feature sparse interactive fusion on the cell target shallow feature map and the cell target trunk feature map to obtain a cell target shallow-trunk significant sparse interactive fusion feature map; a cell recognition unit, which is used to determine the cell recognition result based on the cell target shallow-trunk significant sparse interactive fusion feature map.

[0009] In the above-mentioned intelligent diagnosis and treatment data processing system for hematological diseases, the cell target focusing unit includes: an image preprocessing subunit, which is used to perform image denoising and contrast enhancement on the bone marrow morphology image to obtain a preprocessed bone marrow morphology image; and a target detection subunit, which is used to input the preprocessed bone marrow morphology image into a cell target detection network based on the Yolo model to obtain the cell target focused image.

[0010] In the above-mentioned intelligent diagnosis and treatment data processing system for blood diseases, the multi-level feature extraction unit is used to: input the cell target focused image into the MBCNet network including a backbone network and a boundary feature extraction branch to obtain the cell target shallow feature map and the cell target backbone feature map.

[0011] In the above-mentioned intelligent diagnosis and treatment data processing system for hematological diseases, the feature sparse interaction unit includes: a core feature extraction subunit, which is used to input the cell target shallow feature map and the cell target trunk feature map into the core feature distillation network to obtain the cell target shallow core feature vector and the cell target trunk core feature vector; a core feature global association encoding subunit, which is used to perform global association encoding on the cell target shallow core feature vector and the cell target trunk core feature vector to obtain the global encoding matrix between the cell target shallow and trunk core features; a feature interaction sparse modulation subunit, which is used to perform association interaction feature sparse modulation on the cell target shallow feature map and the cell target trunk feature map based on the global encoding matrix between the cell target shallow and trunk core features to obtain the cell target shallow-trunk significant sparse interaction fusion feature map.

[0012] In the above-mentioned intelligent diagnosis and treatment data processing system for hematological diseases, the core feature extraction subunit is used to: perform feature decomposition on the shallow feature map of the cell target to obtain a set of shallow local feature vectors of the cell target; calculate the semantic difference coefficient of each shallow local feature vector of the cell target in the set of shallow local feature vectors of the cell target relative to other shallow local feature vectors of the cell target to obtain a set of semantic difference coefficients of shallow local features of the cell target; take the negation of the semantic difference coefficient of the shallow local features of each cell target in the set of semantic difference coefficients of shallow local features of the cell target and input it into the Softmax function for normalization processing to obtain a set of significant description factors of shallow local features of the cell target; use the set of significant description factors of shallow local features of the cell target as the weight distribution, and perform weighted aggregation on the set of shallow local feature vectors of the cell target to obtain the shallow core feature vector of the cell target.

[0013] In the above-mentioned intelligent diagnosis and treatment data processing system for blood diseases, the core feature global association coding subunit is used to: calculate the matrix product of the transposed vector of the shallow core feature vector of the cell target and the trunk core feature vector of the cell target to obtain the global coding matrix between the shallow and trunk core features of the cell target.

[0014] In the above-mentioned intelligent diagnosis and treatment data processing system for hematological diseases, the feature interaction sparse modulation subunit includes: a feature sparsification secondary subunit, which is used to input the global coding matrix between the shallow and trunk core features of the cell target into a sparse mask unit to obtain a correlation sparse weight matrix between the shallow and trunk core features of the cell target; a position-by-position correlation interaction secondary subunit, which is used to input the shallow feature map of the cell target and the trunk feature map of the cell target into a position-by-position correlation interaction network to obtain a shallow-trunk position-by-position interaction fusion feature map of the cell target; and a sparse modulation secondary subunit, which is used to perform feature significant sparse modulation on the shallow-trunk position-by-position interaction fusion feature map of the cell target based on the correlation sparse weight matrix between the shallow and trunk core features of the cell target to obtain a significant sparse interaction fusion feature map of the shallow-trunk of the cell target.

[0015] In the above-mentioned intelligent diagnosis and treatment data processing system for blood diseases, the sparse modulation secondary subunit is used to: calculate the product of the sparse weight matrix associated between the shallow and trunk core features of the cell target and the various feature matrices of the shallow and trunk position-by-position interactive fusion feature map of the cell target along the channel dimension to obtain the cell target shallow and trunk significant sparse interactive fusion feature map.

[0016] In the above-mentioned intelligent diagnosis and treatment data processing system for blood diseases, the cell recognition unit is used to: input the shallow layer-trunk significant sparse interactive fusion feature map of the cell target into the classifier-based cell identifier to obtain the recognition result, and the recognition result is used to indicate whether it is an abnormal cell.

[0017] Compared with the prior art, the intelligent diagnosis and treatment data processing system for blood diseases provided by this application uses deep learning-based image processing technology to perform cell detection on the bone marrow morphology images of patients to focus on the cell targets in the image, and extracts multi-level features of the cell targets to capture the shallow detail features and trunk features of the cells, and then performs significant sparse interactive encoding of the two to obtain a more distinguishable cell feature representation, thereby realizing intelligent identification of abnormal cells. In this way, the cell features in blood samples can be automatically identified and analyzed, and the rapid detection and diagnosis of abnormal cells can be realized, thereby assisting doctors in making faster and more accurate diagnosis of blood diseases and reducing the workload of doctors. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 It is a block diagram of the diagnosis and treatment data processing module in the intelligent diagnosis and treatment data processing system for blood diseases according to an embodiment of the present application.

[0020] Figure 2 This is a schematic diagram of data flow of the diagnosis and treatment data processing module in the intelligent diagnosis and treatment data processing system for blood diseases according to an embodiment of the present application.

[0021] Figure 3 It is a block diagram of a cell target focusing unit in a blood disease intelligent diagnosis and treatment data processing system according to an embodiment of the present application.

[0022] Figure 4 It is a block diagram of a feature sparse interaction unit in a data processing system for intelligent diagnosis and treatment of hematological diseases according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0024] As mentioned in the above background technology, patent CN115295150A proposes an intelligent blood disease diagnosis and treatment assistance system, which includes: a database and a cloud platform. The data in the database can be uploaded to the cloud platform. The data in the cloud platform is accessible to qualified hospital terminals. The database includes a patient information module. The patient information module includes a basic information module and an examination information module. The basic information module is used to store the patient's previous medical records in hospitals at all levels. The examination information module includes a routine examination module and a blood disease examination module. The routine examination module is used to store routine examination documents, and the blood disease examination module is used to store patient blood disease examination documents. Doctors can access the patient information module through the patient's account and password.

[0025] Since the above-mentioned intelligent diagnosis and treatment auxiliary system for hematological diseases mainly focuses on the integration and query of patient medical information, the cell morphological analysis in the diagnosis process of hematological diseases still needs to rely on the professional knowledge and experience of doctors, and it is impossible to achieve fully automated data processing, which further limits the efficiency of diagnosis. In order to solve this problem, the present application proposes an optimized intelligent diagnosis and treatment data processing system for hematological diseases, which further deploys a diagnosis and treatment data processing module in the cloud platform to analyze the patient's bone marrow morphological image to automatically identify abnormal cells. Specifically, the diagnosis and treatment data processing module uses deep learning-based image processing technology to perform cell detection on the bone marrow morphological image of the patient object to focus on the cell target in the image, and extracts multi-level features of the cell target to capture the shallow detail features and trunk features of the cell, and then performs significant sparse interactive encoding on the two to obtain a more distinguishable cell feature representation, thereby realizing intelligent identification of abnormal cells. In this way, the cell features in the blood sample can be automatically identified and analyzed, and the rapid detection and diagnosis of abnormal cells can be achieved, so that doctors can be assisted in faster and more accurate diagnosis of hematological diseases, reducing the workload of doctors.

[0026] Figure 1 It is a block diagram of the diagnosis and treatment data processing module in the intelligent diagnosis and treatment data processing system for blood diseases according to an embodiment of the present application. Figure 2 This is a data flow diagram of the diagnosis and treatment data processing module in the intelligent diagnosis and treatment data processing system for blood diseases according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the diagnosis and treatment data processing module 100 of the intelligent diagnosis and treatment data processing system for hematological diseases according to the embodiment of the present application includes: a bone marrow morphology image acquisition unit 110, used to extract the bone marrow morphology image of the patient object from the blood disease examination module; a cell target focusing unit 120, used to perform cell target detection on the bone marrow morphology image to obtain a cell target focused image; a multi-level feature extraction unit 130, used to perform multi-level feature extraction on the cell target focused image to obtain a cell target shallow feature map and a cell target trunk feature map; a feature sparse interaction unit 140, used to perform significant feature sparse interactive fusion on the cell target shallow feature map and the cell target trunk feature map to obtain a cell target shallow-trunk significant sparse interactive fusion feature map; a cell recognition unit 150, used to determine the cell recognition result based on the cell target shallow-trunk significant sparse interactive fusion feature map.

[0027] In the above-mentioned hematological disease intelligent diagnosis and treatment data processing system 100, the bone marrow morphology image acquisition unit 110 is used to extract the bone marrow morphology image of the patient object from the hematological disease examination module. It should be understood that the bone marrow is the place where blood cells (red blood cells, white blood cells and platelets) are produced. For many blood diseases, the cell morphological characteristics in the bone marrow can provide important diagnostic information. For example, in certain types of leukemia, a large number of immature or abnormal white blood cells appear in the bone marrow. Therefore, in the technical solution of the present application, by analyzing the bone marrow morphology image of the patient object, abnormal cells can be effectively identified based on the cell morphology characteristics in the image, thereby revealing the hematopoietic function state of the bone marrow and potential signs of blood diseases.

[0028] In the above-mentioned intelligent diagnosis and treatment data processing system 100 for blood diseases, the cell target focusing unit 120 is used to perform cell target detection on the bone marrow morphology image to obtain a cell target focused image. Figure 3 FIG. 1 is a block diagram of a cell target focusing unit in a blood disease intelligent diagnosis and treatment data processing system according to an embodiment of the present application. Figure 3 As shown, the cell target focusing unit 120 includes: an image preprocessing subunit 121, used for performing image denoising and contrast enhancement on the bone marrow morphology image to obtain a preprocessed bone marrow morphology image; and a target detection subunit 122, used for inputting the preprocessed bone marrow morphology image into a cell target detection network based on the Yolo model to obtain the cell target focused image.

[0029] Specifically, the image preprocessing subunit 121 is used for performing image noise reduction and contrast enhancement on the bone marrow morphology image to obtain the preprocessed bone marrow morphology image. It should be understood that, considering that in practical applications, the collected bone marrow morphology image may be affected by various factors, such as equipment noise, ambient light interference, etc., resulting in unnecessary noise in the image, thereby interfering with the clarity of the cell boundary and affecting the subsequent cell detection accuracy. Therefore, in order to improve the image quality, the application performs image noise reduction and contrast enhancement on the bone marrow morphology image to improve the clarity of the cell boundary and the overall contrast of the image, thereby reducing the false detection or missed detection caused by image quality problems, and improving the reliability and accuracy of the detection result. In an embodiment of the present application, a Gaussian filter is used to perform image noise reduction on the bone marrow morphology image to smooth the image and remove noise. Then, the contrast of the bone marrow morphology image after noise reduction is enhanced by using histogram equalization technology, so that the outline and internal structure of the cell are more clearly visible.

[0030] Specifically, the target detection subunit 122 is used for inputting the pre-processed bone marrow morphology image into a cell target detection network based on the Yolo model to obtain the cell target focused image. That is, considering that the cell target morphology is diverse, densely arranged and the background is complex in the bone marrow morphology image. Therefore, in order to avoid the interference of background information to the cell morphology analysis and reduce the amount of calculation of the invalid area, the present application further uses the cell target detection network based on the Yolo model to process the pre-processed bone marrow morphology image, so as to separate cells from the complex background by utilizing the powerful target detection ability of the Yolo model and obtain the cell target focused image. Those of ordinary skill in the art should know that the Yolo model can realize the rapid positioning and identification of the cell target by dividing the pre-processed bone marrow morphology image into grids and predicting the bounding box and probability in each grid.

[0031] In the above-mentioned intelligent diagnosis and treatment data processing system 100 for blood diseases, the multi-level feature extraction unit 130 is used to perform multi-level feature extraction on the cell target focused image to obtain a cell target shallow feature map and a cell target trunk feature map. It should be understood that due to the diversity of cell morphological features, including detailed information such as cell edges and textures, as well as global information such as overall morphology, structure and size, the traditional neural network can capture deeper features as the network depth increases, but it will also cause the loss and blurring of shallow detail information. Therefore, in order to capture the local details and global information of the cell target at the same time, in a specific example of the present application, the cell target focused image is input into the MBCNet network including the backbone network and the boundary feature extraction branch to obtain the cell target shallow feature map and the cell target trunk feature map. Specifically, the MBCNet network can process shallow features such as edge details of the cell target and trunk features such as the overall morphological structure of the cell target through the collaborative work of the backbone network and the boundary feature extraction branch, so as to generate a cell target shallow feature map and a cell target trunk feature map, thereby realizing a comprehensive analysis of the cell target.

[0032] In the above-mentioned hematological disease intelligent diagnosis and treatment data processing system 100, the feature sparse interaction unit 140 is used to perform significant feature sparse interaction fusion on the cell target shallow feature map and the cell target trunk feature map to obtain a cell target shallow-trunk significant sparse interaction fusion feature map. It should be understood that in order to make full use of the complementarity of the shallow features and trunk features of the cell target to improve the comprehensiveness of the feature representation, it is necessary to further perform feature fusion on the shallow feature map of the cell target and the trunk feature map of the cell target. In particular, considering that traditional fusion methods such as feature splicing or position addition cannot fully explore the correlation between shallow features and trunk features, it may cause information loss or redundancy, affecting the final analysis results. In this regard, the present application proposes a fusion method based on feature sparse interaction coding, which can effectively identify the significant areas in the feature map and extract the most representative fusion feature representation through feature distillation and sparsity constraint technology, thereby achieving a more accurate description of cell features.

[0033] Figure 4 FIG. 1 is a block diagram of a feature sparse interaction unit in a blood disease intelligent diagnosis and treatment data processing system according to an embodiment of the present application. Figure 4 As shown, the feature sparse interaction unit 140 includes: a core feature extraction subunit 141, which is used to input the cell target shallow feature map and the cell target trunk feature map into the core feature distillation network to obtain a cell target shallow core feature vector and a cell target trunk core feature vector; a core feature global correlation encoding subunit 142, which is used to perform global correlation encoding on the cell target shallow core feature vector and the cell target trunk core feature vector to obtain a global coding matrix between cell target shallow and trunk core features; a feature interaction sparse modulation subunit 143, which is used to perform correlation interaction feature sparse modulation on the cell target shallow feature map and the cell target trunk feature map based on the global coding matrix between the cell target shallow and trunk core features to obtain the cell target shallow-trunk significant sparse interaction fusion feature map.

[0034] Specifically, the core feature extraction subunit 141 is used to: perform feature decomposition on the shallow feature map of the cell target to obtain a set of shallow local feature vectors of the cell target; calculate the semantic difference coefficient of each shallow local feature vector of the cell target in the set of shallow local feature vectors of the cell target relative to other shallow local feature vectors of the cell target to obtain a set of semantic difference coefficients of shallow local features of the cell target; take the negation of each semantic difference coefficient of shallow local features of the cell target in the set of semantic difference coefficients of shallow local features of the cell target and input the result into the Softmax function for normalization to obtain a set of significant description factors of shallow local features of the cell target; use the set of significant description factors of shallow local features of the cell target as the weight distribution, and perform weighted aggregation on the set of shallow local feature vectors of the cell target to obtain the shallow core feature vector of the cell target. In a specific example of the present application, the semantic difference coefficient of the shallow local feature vector of each cell target relative to the shallow local feature vectors of other cell targets is calculated, including: subtracting the i-th cell target shallow local feature vector in the set of the cell target shallow local feature vectors from the other cell target shallow local feature vectors at the position point to obtain a set of cell target shallow local semantic difference feature vectors; calculating the average value of the one-norm of the shallow local semantic difference feature vectors of each cell target to obtain the semantic difference coefficient of the i-th cell target shallow local feature vector.

[0035] Specifically, the core feature global association coding subunit 142 is used to calculate the matrix product of the transposed vector of the cell target shallow core feature vector and the cell target trunk core feature vector to obtain the global coding matrix between the cell target shallow and trunk core features.

[0036] Specifically, the feature interaction sparse modulation subunit 143 includes: a feature sparsification secondary subunit, which is used to input the global coding matrix between the shallow and trunk core features of the cell target into a sparse mask unit to obtain a correlation sparse weight matrix between the shallow and trunk core features of the cell target; a position-by-position correlation interaction secondary subunit, which is used to input the shallow feature map of the cell target and the trunk feature map of the cell target into a position-by-position correlation interaction network to obtain a shallow-trunk position-by-position interaction fusion feature map of the cell target; and a sparse modulation secondary subunit, which is used to perform feature significant sparse modulation on the shallow-trunk position-by-position interaction fusion feature map of the cell target based on the correlation sparse weight matrix between the shallow and trunk core features of the cell target to obtain a significant sparse interaction fusion feature map of the cell target shallow-trunk.

[0037] In a specific example of the present application, the feature sparsification secondary subunit is used to: set a first mask threshold and a second mask threshold, wherein the second mask threshold is twice the first mask threshold; if the eigenvalue in the global coding matrix between the shallow layer and the trunk core features of the cell target is greater than the second mask threshold, it is tripled; if the eigenvalue in the global coding matrix between the shallow layer and the trunk core features of the cell target is greater than the first mask threshold and is less than or equal to the second mask threshold, it remains unchanged; if the eigenvalue in the global coding matrix between the shallow layer and the trunk core features of the cell target is less than or equal to the first mask threshold, it is tripled, thereby obtaining the associated sparse weight matrix between the shallow layer and the trunk core features of the cell target.

[0038] In a specific example of the present application, the sparse modulation secondary subunit is used to: calculate the product of the sparse weight matrix associated between the shallow and trunk core features of the cell target and the various feature matrices of the shallow and trunk position-by-position interactive fusion feature map of the cell target along the channel dimension to obtain the cell target shallow and trunk significant sparse interactive fusion feature map.

[0039] Accordingly, the feature sparse interaction unit 140 is used to perform feature interactive fusion on the cell target shallow feature map and the cell target trunk feature map according to the following feature interactive fusion formula to obtain the cell target shallow-trunk significant sparse interactive fusion feature map, wherein the feature interactive fusion formula is:

[0040] v 1 =KFDN(F 1 )

[0041] v 2 =KFDN(F 2 )

[0042]

[0043] F i =F 1 ⊙F 2

[0044]

[0045] Among them, F 1 represents the shallow feature map of the cell target, F 2 represents the main feature map of the cell target, KFDN(·) is the core feature distillation network, decouple(·) represents the feature decomposition operation, X represents the set of shallow local feature vectors of the cell target, x 1 、x 2 、x i 、x jand x n Respectively represent the first, second, i-th, j-th and n-th shallow local feature vectors of the cell target in the set of shallow local feature vectors of the cell target, n is the number of shallow local feature vectors of the cell target, ‖·‖ 1 Indicates the calculation of the norm of the eigenvector, D i represents the semantic difference coefficient of the shallow local feature of the ith cell target, exp(·) represents the exponential function with e as the base, and v 1 represents the shallow core feature vector of the cell target, v 2 represents the core feature vector of the cell target backbone, (·) T represents the transpose of a vector, represents matrix multiplication operation, ⊙ represents Hadamard product, M represents the global encoding matrix between the shallow layer and the backbone core features of the cell target, M ij represents the eigenvalue of the position (i, j) in the global encoding matrix between the shallow layer and the backbone core features of the cell target, θ is the first mask threshold, mask(·) represents the mask processing, M a Represents the sparse weight matrix of the correlation between the shallow layer and the backbone core features of the cell target, F i Represents the shallow layer-trunk position-by-position interactive fusion feature map of the cell target, F f Represents the shallow layer-trunk significant sparse interaction fusion feature map of the cell target.

[0046] That is, first, a core feature distillation network is constructed to process the shallow feature map of the cell target and the trunk feature map of the cell target respectively, and the importance of each local area is evaluated based on the semantic difference of each local area in the feature map relative to other feature areas, and the corresponding weight is generated. By weighted aggregation of the features of each local area, the core features of the shallow feature map of the cell target and the trunk feature map of the cell target are extracted, and the shallow core feature vector of the cell target and the trunk core feature vector of the cell target are obtained to retain the core attribute information of the feature map. Then, the shallow core feature vector of the cell target and the trunk core feature vector of the cell target are globally associated and encoded, and a global encoding matrix between the shallow and trunk core features of the cell target is constructed to reveal the strength and pattern of association between the two. Then, a sparsity constraint is introduced, and a sparse masking process is performed on the global encoding matrix between the shallow and trunk core features of the cell target to suppress redundant information, strengthen the effective association between features, and generate a sparse weight matrix of association between the shallow and trunk core features of the cell target. At the same time, in order to focus on the feature interaction between the shallow feature map of the cell target and the trunk feature map of the cell target at the local detail level, on another data processing link, the two are associated and interacted position by position to achieve information exchange and integration at each corresponding position between the two, and form a shallow-trunk position-by-position interactive fusion feature map of the cell target. Finally, based on the importance level indicated by the sparse weight matrix of the association between the shallow-trunk core features of the cell target, the shallow-trunk position-by-position interactive fusion feature map of the cell target is significantly sparsely modulated to obtain a shallow-trunk significantly sparse interactive fusion feature map of the cell target. In this way, the potential correlation structure between the shallow features and trunk features of the cell target that are difficult to capture by traditional methods can be effectively revealed, thereby achieving a deeper understanding of the features of the cell target.

[0047] In the above-mentioned intelligent diagnosis and treatment data processing system 100 for blood diseases, the cell recognition unit 150 is used to determine the recognition result of the cell based on the shallow-trunk significant sparse interactive fusion feature map of the cell target. In a specific example of the present application, the cell recognition unit 150 is used to: input the shallow-trunk significant sparse interactive fusion feature map of the cell target into a cell identifier based on a classifier to obtain the recognition result, and the recognition result is used to indicate whether it is an abnormal cell. That is, the classification algorithm is used to classify and judge the shallow-trunk significant sparse interactive fusion feature map of the cell target. Specifically, the classifier can effectively identify and distinguish normal cells from abnormal cells by learning and analyzing the cell target morphological features contained in the shallow-trunk significant sparse interactive fusion feature map of the cell target, and map the shallow-trunk significant sparse interactive fusion feature map of the cell target to the corresponding category label to achieve accurate classification of the cell state, thereby providing effective auxiliary information for the clinical diagnosis of blood diseases.

[0048] In summary, according to the embodiment of the present application, the intelligent diagnosis and treatment data processing system for blood diseases is explained, which uses deep learning-based image processing technology to perform cell detection on the bone marrow morphology image of the patient object to focus on the cell target in the image, and extracts multi-level features of the cell target to capture the shallow detail features and trunk features of the cell, and then performs significant sparse interactive encoding on the two to obtain a more distinguishing cell feature representation, thereby realizing intelligent identification of abnormal cells. In this way, the cell features in the blood sample can be automatically identified and analyzed, and the rapid detection and diagnosis of abnormal cells can be realized, thereby assisting doctors in diagnosing blood diseases more quickly and accurately, reducing the workload of doctors.

[0049] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0050] In the above embodiments, the description of each embodiment has its own emphasis. For the parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0051] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0052] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A blood disease intelligent diagnosis and treatment data processing system, comprising: A database and a cloud platform, wherein the data in the database can be uploaded to the cloud platform, and the data in the cloud platform is accessible to qualified hospital terminals, wherein the database includes a patient information module, wherein the patient information module includes a basic information module and an examination information module, wherein the basic information module is used to store the patient's previous medical records in hospitals at all levels, wherein the examination information module includes a routine examination module and a blood disease examination module, wherein the routine examination module is used to store routine examination receipts, and the blood disease examination module is used to store the patient's blood disease examination receipts, and the doctor can access the patient information module through the patient's account number and password, wherein the cloud platform also includes a diagnosis and treatment data processing module, which is used to analyze the patient's bone marrow morphology image to automatically identify abnormal cells; Wherein, the diagnosis and treatment data processing module includes: A bone marrow morphology image acquisition unit, used for extracting a bone marrow morphology image of a patient object from the blood disease examination module; A cell target focusing unit, used for performing cell target detection on the bone marrow morphology image to obtain a cell target focused image; A multi-level feature extraction unit, used for performing multi-level feature extraction on the cell target focused image to obtain a cell target shallow feature map and a cell target trunk feature map; A feature sparse interaction unit, used for performing significant feature sparse interaction fusion on the cell target shallow feature map and the cell target trunk feature map to obtain a cell target shallow layer-trunk significant sparse interaction fusion feature map; The cell recognition unit is used to determine the recognition result of the cell based on the shallow layer-trunk significant sparse interactive fusion feature map of the cell target.

2. The intelligent blood disease diagnosis and treatment data processing system according to claim 1 is characterized in that: The cell target focusing unit comprises: An image preprocessing subunit, used for performing image noise reduction and contrast enhancement on the bone marrow morphology image to obtain a preprocessed bone marrow morphology image; The target detection subunit is used to input the pre-processed bone marrow morphology image into a cell target detection network based on the Yolo model to obtain the cell target focused image.

3. The intelligent blood disease diagnosis and treatment data processing system according to claim 2 is characterized in that: The multi-level feature extraction unit is used to: The cell target focused image is input into an MBCNet network including a backbone network and a boundary feature extraction branch to obtain a shallow feature map of the cell target and a backbone feature map of the cell target.

4. The intelligent blood disease diagnosis and treatment data processing system according to claim 3 is characterized in that: The feature sparse interaction unit comprises: A core feature extraction subunit, used for inputting the cell target shallow feature map and the cell target trunk feature map into a core feature distillation network to obtain a cell target shallow core feature vector and a cell target trunk core feature vector; A core feature global association coding subunit is used to perform global association coding on the shallow core feature vector of the cell target and the trunk core feature vector of the cell target to obtain a global coding matrix between the shallow and trunk core features of the cell target; The feature interaction sparse modulation subunit is used to perform associated interactive feature sparse modulation on the cell target shallow feature map and the cell target trunk feature map based on the global coding matrix between the cell target shallow layer and the trunk core features to obtain the cell target shallow layer and the trunk significant sparse interactive fusion feature map.

5. The intelligent blood disease diagnosis and treatment data processing system according to claim 4 is characterized in that: The core feature extraction subunit is used to: Performing feature decomposition on the shallow feature map of the cell target to obtain a set of shallow local feature vectors of the cell target; Calculating the semantic difference coefficient of each shallow local feature vector of each cell target relative to other shallow local feature vectors of other cell targets in the set of shallow local feature vectors of the cell targets to obtain a set of semantic difference coefficients of shallow local features of cell targets; Taking the inverse of each cell target shallow local feature semantic difference coefficient in the set of cell target shallow local feature semantic difference coefficients, and inputting the inverse of ... The set of saliency description factors of the shallow local features of the cell target is used as weight distribution, and the set of shallow local feature vectors of the cell target is weightedly aggregated to obtain the shallow core feature vector of the cell target.

6. The intelligent blood disease diagnosis and treatment data processing system according to claim 5 is characterized in that: The core feature global association encoding subunit is used to: The matrix product of the shallow core feature vector of the cell target and the transposed vector of the trunk core feature vector of the cell target is calculated to obtain the global encoding matrix between the shallow and trunk core features of the cell target.

7. The intelligent blood disease diagnosis and treatment data processing system according to claim 6 is characterized in that: The feature interaction sparse modulation subunit comprises: The feature sparse secondary subunit is used to input the global encoding matrix between the shallow layer and the backbone core features of the cell target into the sparse mask unit to obtain the correlation sparse weight matrix between the shallow layer and the backbone core features of the cell target; A position-by-position association interaction secondary subunit, used for inputting the cell target shallow feature map and the cell target trunk feature map into a position-by-position association interaction network to obtain a cell target shallow layer-trunk position-by-position interaction fusion feature map; The sparse modulation secondary subunit is used to perform feature-significant sparse modulation on the shallow-trunk position-by-position interactive fusion feature map of the cell target based on the associated sparse weight matrix between the shallow-trunk core features of the cell target to obtain the shallow-trunk significant sparse interactive fusion feature map of the cell target.

8. The blood disease intelligent diagnosis and treatment data processing system according to claim 7 is characterized in that: The sparse modulation secondary subunit is used for: The product of the correlation sparse weight matrix between the shallow-trunk core features of the cell target and each feature matrix along the channel dimension of the shallow-trunk position-by-position interactive fusion feature map of the cell target is calculated to obtain the shallow-trunk significant sparse interactive fusion feature map of the cell target.

9. The intelligent blood disease diagnosis and treatment data processing system according to claim 8, characterized in that: The cell recognition unit is used for: The shallow layer-trunk significant sparse interactive fusion feature map of the cell target is input into a classifier-based cell identifier to obtain the identification result, and the identification result is used to indicate whether it is an abnormal cell.

Images