Integrated system for intelligent precise typing diagnosis and targeted therapy of blood diseases

Abstract

The invention discloses an intelligent precise typing diagnosis and targeted therapy integrated system for blood diseases, and relates to the technical field of medical intelligence. The system comprises a data layer, an algorithm layer, an application layer and a security assurance module which are in bidirectional communication and work cooperatively. The data layer integrates clinical data, image data, molecular data and drug data; the algorithm layer constructs a typing diagnosis model, a target spot prediction model and a treatment scheme recommendation model, and optimizes the analysis precision through formulas such as feature importance evaluation; the application layer provides functions of intelligent diagnosis, scheme making, curative effect monitoring and the like; and the security guarantee module guarantees data security. The system realizes precise typing and personalized treatment of blood diseases, improves diagnosis efficiency and treatment efficiency, promotes clinical and scientific research cooperation, solves traditional diagnosis and treatment pain points, and is suitable for medical institutions at all levels.

Description

Technical Field

[0001] This invention relates to the field of hematological diseases, specifically to an integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases. Background Technology

[0002] Hematologic disorders are a group of diseases involving abnormalities in the hematopoietic system and blood components, including various types such as acute leukemia, lymphoma, and multiple myeloma. Their classification is complex and has numerous subtypes; accurate subtyping is a prerequisite for effective treatment. In traditional hematologic disease diagnosis and treatment, subtyping diagnosis mainly relies on doctors' morphological observation of bone marrow smears, combined with limited clinical data and molecular testing results. This process is highly dependent on the doctor's personal experience and is prone to subtyping errors due to subjective judgment differences. Furthermore, completing the entire diagnostic process takes 1-2 days, delaying the patient's treatment opportunity.

[0003] In terms of treatment plan development, traditional models often employ standardized protocols, making it difficult to consider individual differences such as patient genetic variations and physical conditions. This results in low treatment effectiveness and high rates of adverse reactions for some patients. Simultaneously, diagnostic and treatment data is scattered across different hospital systems, making efficient integration of clinical, imaging, molecular, and pharmacological data difficult and hindering comprehensive support for precision medicine. Furthermore, privacy concerns during multi-center data sharing limit the dissemination of high-quality treatment experiences and the translation of research findings, highlighting a significant gap in treatment levels between primary healthcare institutions and large hospitals. Therefore, there is an urgent need for an integrated system capable of precise subtyping, personalized treatment, and secure data sharing to address many of the pain points in current hematological disease diagnosis and treatment. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of the above-mentioned technologies and provide an integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases.

[0005] To address the aforementioned technical problems, the present invention provides an integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases. This system comprises a data layer, an algorithm layer, an application layer, and a security module. The data layer, algorithm layer, application layer, and security module are interconnected bidirectionally. The data layer collects, integrates, and securely stores multi-dimensional medical data, providing standardized data support for the entire system. The algorithm layer constructs a multimodal fusion model library based on the standardized data provided by the data layer, trains the model using specific training techniques, and continuously optimizes the model, providing algorithmic support for diagnostic and treatment decisions. The application layer, relying on the algorithm output of the algorithm layer, provides dedicated functional modules for clinicians, laboratory technicians, and researchers, enabling specific applications related to diagnosis, treatment, and research. The security module provides end-to-end security protection for the multi-dimensional medical data in the data layer, the model data in the algorithm layer, and the operational data in the application layer, ensuring the data security and operational compliance of the entire system.

[0006] As an improvement, the data layer includes a data acquisition unit, a data docking unit, and a data storage unit. The data acquisition unit is used to collect multi-dimensional medical data. The data docking unit docks with the hospital information system through a health information exchange standard protocol to achieve interoperability of multi-dimensional medical data. The data storage unit is used to classify and securely store the multi-dimensional medical data integrated by the data docking unit.

[0007] As an improvement, the multidimensional medical data includes clinical data, imaging data, molecular data, and drug data. The clinical data includes patient history, signs, and laboratory test results. The imaging data includes bone marrow smear images and immunohistochemical slide images. The molecular data includes gene sequencing results and fusion gene detection results. The drug data includes targeted drug target information and targeted drug resistance mechanism information.

[0008] As an improvement, the algorithm layer includes a model building unit, a model training unit, and a model optimization unit. The model building unit is used to build a subtyping diagnosis model, a target prediction model, and a treatment plan recommendation model to form a multimodal fusion model library. The model training unit completes multi-center model training through federated learning technology while protecting data privacy. The model optimization unit supports incremental optimization of the subtyping diagnosis model, the target prediction model, and the treatment plan recommendation model using new case data.

[0009] As an improvement, the typing and diagnostic model is a two-layer diagnostic model, which includes a primary screening model and a subtype typing model. The primary screening model is used to distinguish different categories of hematological diseases such as acute leukemia, lymphoma, and multiple myeloma. The subtype typing model is used to accurately classify specific categories of hematological diseases identified by the primary screening model into subtypes.

[0010] As an improvement, the treatment plan recommendation model is a three-dimensional matching model, which is constructed based on patient target information, targeted drug information and individual patient information. The three-dimensional matching model is used to generate personalized targeted treatment plans and provide early warning of medication risks in the targeted treatment plans.

[0011] As an improvement, the application layer includes an intelligent diagnosis module, a treatment plan formulation module, an efficacy monitoring module, and a research support module. The intelligent diagnosis module is used to generate a subtyping diagnosis report based on the output of the subtyping diagnosis model. The treatment plan formulation module is used to output a targeted treatment plan based on the output of the treatment plan recommendation model. The efficacy monitoring module is used to dynamically track the treatment effect of patients. The research support module is used to provide research data services for hematological disease-related research.

[0012] As an improvement, the intelligent diagnostic module includes a data uploading unit, a multimodal analysis unit, and a report generation unit. The data uploading unit is used to receive clinical data and various test data uploaded by clinicians. The multimodal analysis unit is used to call the subtyping diagnostic model to perform comprehensive analysis on the uploaded data. The report generation unit is used to output a diagnostic report containing clear subtyping results.

[0013] As an improvement, the efficacy monitoring module includes a follow-up data acquisition unit, an efficacy evaluation unit, and a treatment plan adjustment early warning unit. The follow-up data acquisition unit is used to acquire various follow-up data of patients after treatment. The efficacy evaluation unit is used to construct an efficacy evaluation model based on the follow-up data and complete the efficacy evaluation. The treatment plan adjustment early warning unit is used to issue an early warning for treatment plan adjustment when a decline in patient efficacy or target mutation is detected.

[0014] As an improvement, the security module includes a data encryption unit, an access control unit, and an operation log tracing unit. The data encryption unit is used to encrypt all medical data in the system. The access control unit is used to set hierarchical access permissions for users with different roles. The operation log tracing unit is used to record all operations in the system and support the tracing and querying of operation behaviors.

[0015] As an improvement, a security module is also included, which comprises a data encryption unit, an access control unit, and a log tracing unit. The data encryption unit is used to encrypt medical data, the access control unit is used to set hierarchical access permissions, and the log tracing unit is used to record all operations and support tracing queries.

[0016] The advantages of this invention compared to existing technologies are as follows: First, it improves the accuracy and efficiency of subtyping diagnosis. The system's subtyping diagnosis model filters core features through a feature importance assessment formula, combining a two-layer logic of initial screening and subtyping models to achieve accurate classification of hematological diseases from broad categories to subtypes, reducing subjective human error. The multimodal analysis unit can complete the entire analysis process within 15 minutes, significantly shortening the traditional 1-2 day diagnostic time, thus saving patients valuable treatment time.

[0017] Secondly, it enables personalized targeted therapy. The treatment plan recommendation model transforms qualitative indicators such as target matching and clinical efficacy into quantitative data through a matching degree formula, recommending plans that fit the individual patient's situation, automatically alerting patients to drugs with low matching degree, thereby improving treatment effectiveness and reducing adverse reactions.

[0018] Third, a closed-loop diagnosis and treatment system is constructed. The efficacy monitoring module quantifies the effect through the efficacy improvement rate formula, tracks changes in the condition in real time, and provides timely warnings and recommendations for adjustment when indicators are abnormal, avoiding the problem of delayed efficacy judgment.

[0019] Fourth, ensuring data security and promoting scientific research. The security module safeguards data security with double encryption and hierarchical access control, while federated learning technology enables collaborative data sharing among multiple centers; the research support module provides data and interfaces for hematological disease research, forming a virtuous cycle between clinical practice and scientific research, while lowering the threshold for diagnosis and treatment and improving the accessibility of medical resources. Attached Figure Description

[0020] Figure 1 This is the overall system architecture diagram of the intelligent and precise subtyping diagnosis and targeted therapy integrated system for hematological diseases of the present invention.

[0021] Figure 2 This invention presents the core functions and data flow diagrams of each layer of the integrated intelligent and precise classification diagnosis and targeted therapy system for hematological diseases. Detailed Implementation

[0022] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0023] Referring to the attached diagram, the integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases includes a data layer, an algorithm layer, an application layer, and a security module. The data layer, algorithm layer, application layer, and security module are interconnected bidirectionally. The data layer collects, integrates, and securely stores multi-dimensional medical data, providing standardized data support for the entire system. The algorithm layer constructs a multimodal fusion model library based on the standardized data provided by the data layer, completes model training through specific training techniques, and continuously optimizes the model, providing algorithmic support for diagnostic and treatment decisions. The application layer, relying on the algorithm output of the algorithm layer, provides dedicated functional modules for clinicians, laboratory technicians, and researchers, enabling specific applications related to diagnosis, treatment, and research. The security module provides end-to-end security protection for the multi-dimensional medical data in the data layer, the model data in the algorithm layer, and the operational data in the application layer, ensuring the data security and operational compliance of the entire system.

[0024] The data layer includes a data acquisition unit, a data docking unit, and a data storage unit. The data acquisition unit is used to collect multi-dimensional medical data. The data docking unit docks with the hospital information system through a health information exchange standard protocol to achieve the interoperability of multi-dimensional medical data. The data storage unit is used to classify and securely store the multi-dimensional medical data integrated by the data docking unit.

[0025] The multidimensional medical data includes clinical data, imaging data, molecular data, and drug data. The clinical data includes patient history, signs, and laboratory test results. The imaging data includes bone marrow smear images and immunohistochemical slide images. The molecular data includes gene sequencing results and fusion gene detection results. The drug data includes targeted drug target information and targeted drug resistance mechanism information.

[0026] The algorithm layer includes a model building unit, a model training unit, and a model optimization unit. The model building unit is used to build a subtyping diagnosis model, a target prediction model, and a treatment plan recommendation model to form a multimodal fusion model library. The model training unit completes multi-center model training through federated learning technology while protecting data privacy. The model optimization unit supports incremental optimization of the subtyping diagnosis model, the target prediction model, and the treatment plan recommendation model using new case data.

[0027] The classification diagnostic model is a two-layer diagnostic model, which includes a primary screening model and a subtype classification model. The primary screening model is used to distinguish different major categories of hematological diseases such as acute leukemia, lymphoma, and multiple myeloma. The subtype classification model is used to accurately classify specific major categories of hematological diseases identified by the primary screening model into subtypes.

[0028] The treatment plan recommendation model is a three-dimensional matching model, which is constructed based on patient target information, targeted drug information, and individual patient information. The three-dimensional matching model is used to generate personalized targeted treatment plans and provide early warning of medication risks in the targeted treatment plans.

[0029] The application layer includes an intelligent diagnosis module, a treatment plan formulation module, an efficacy monitoring module, and a research support module. The intelligent diagnosis module is used to generate a subtyping diagnosis report based on the output of the subtyping diagnosis model. The treatment plan formulation module is used to output a targeted treatment plan based on the output of the treatment plan recommendation model. The efficacy monitoring module is used to dynamically track the treatment effect of patients. The research support module is used to provide research data services for hematological disease-related research.

[0030] The intelligent diagnostic module includes a data uploading unit, a multimodal analysis unit, and a report generation unit. The data uploading unit is used to receive clinical data and various test data uploaded by clinicians. The multimodal analysis unit is used to call the subtyping diagnostic model to perform comprehensive analysis on the uploaded data. The report generation unit is used to output a diagnostic report containing clear subtyping results.

[0031] The efficacy monitoring module includes a follow-up data acquisition unit, an efficacy evaluation unit, and a treatment plan adjustment early warning unit. The follow-up data acquisition unit is used to acquire various follow-up data of patients after treatment. The efficacy evaluation unit is used to construct an efficacy evaluation model based on the follow-up data and complete the efficacy evaluation. The treatment plan adjustment early warning unit is used to issue an early warning for treatment plan adjustment when a decline in patient efficacy or target mutation is detected.

[0032] The security module includes a data encryption unit, an access control unit, and an operation log tracing unit. The data encryption unit is used to encrypt all medical data in the system. The access control unit is used to set hierarchical access permissions for users with different roles. The operation log tracing unit is used to record all operation behaviors in the system and support the tracing query of operation behaviors.

[0033] It also includes a security module, which comprises a data encryption unit, an access control unit, and a log tracing unit. The data encryption unit is used to encrypt medical data, the access control unit is used to set hierarchical access permissions, and the log tracing unit is used to record all operations and support tracing queries.

[0034] This system comprises a data layer, an algorithm layer, an application layer, and a security module. These layers interact in real-time via a bidirectional communication link based on Transmission Control Protocol / Internet Protocol (TCP / IP), ensuring the immediacy of command issuance and result feedback. The data layer, serving as the system's data foundation, is responsible for the "collection-storage-management" of multi-dimensional medical data. The algorithm layer is the core processing unit, performing intelligent analysis of diagnostic and treatment plans through multimodal models. The application layer provides a visual user interface for end users. The security module operates throughout the entire process, ensuring data privacy and operational compliance.

[0035] In this embodiment, the system adopts a distributed deployment approach. The data layer is deployed on a dedicated server cluster in the hospital's data center, the algorithm layer relies on cloud GPU computing nodes to achieve efficient computation, the application layer is accessible to users with different roles through web pages, mobile devices, and hospital workstations, and the security module is integrated into the entry and exit nodes of each layer to form a comprehensive protection network.

[0036] Specific implementation details for each level of module:

[0037] Implementation of the data layer:

[0038] The data layer includes a data acquisition unit, a data docking unit, and a data storage unit, which are used to collect and integrate clinical data, imaging data, molecular data, and drug data.

[0039] The data acquisition unit adopts a multi-port data access design. Clinical data is automatically captured through the interface with the hospital's electronic medical record system, covering the patient's past medical history, present medical history, physical examination records, and laboratory test results such as blood routine and biochemical indicators. Imaging data is acquired through dedicated imaging equipment, including high-definition images (resolution up to 4000×3000 pixels) generated by a digital slide scanner from bone marrow smears and digital images of immunohistochemical slides. Molecular data is directly imported from the gene sequencer through a data transmission line, including base sequence information from gene sequencing results and positive / negative results of fusion gene detection. Drug data is updated regularly through synchronized medical databases, covering the target name, mechanism of action, route of administration, and drug resistance-related gene information of targeted drugs.

[0040] The data exchange unit uses the Health Information Exchange Standard Protocol (HAEP) to interface with the hospital's laboratory information system, image archiving and communication system, and electronic medical record system. To address the issue of inconsistent data formats across different systems, this unit incorporates a data standardization processing module that converts text, image, and sequence data of various formats into a unified format recognizable by the system. For example, it uses natural language processing technology to extract and structure key information from clinical text data, performs size normalization on image data, and converts gene sequencing data into the FASTA standard format.

[0041] The data storage unit adopts a dual storage architecture of "distributed database + local backup". The distributed database uses HBase, a Hadoop-based database, to store massive amounts of raw and standardized data, supporting more than 1,000 data writes and queries per second. The local backup uses a disk array storage, which automatically performs a full backup of the data in the distributed database at 2:00 AM every day to ensure that the data is not lost in extreme situations. To facilitate subsequent data retrieval and access, the data storage unit classifies and stores data according to a hierarchical structure of "data type-patient ID-collection time". For example, the bone marrow smear image collected on May 1, 2024, for patient number "P20240501001" would be stored in the path "image data / P20240501001 / 20240501 / bone marrow smear / ".

[0042] Implementation of the algorithm layer:

[0043] The algorithm layer includes model building units, model training units, and model optimization units, which construct subtyping diagnosis models, target prediction models, and treatment plan recommendation models, forming a multimodal fusion model library.

[0044] The model building units adopt a modular design, with each model built independently and then working collaboratively through a fusion interface. The subtyping diagnostic model is a two-layer model: the initial screening model is built based on a convolutional neural network, taking standardized imaging and clinical data as input and outputting a broad category of hematological diseases (acute leukemia, lymphoma, multiple myeloma, etc.); the subtyping model is built based on a Transformer model, taking the output of the initial screening model and molecular data as input and outputting specific subtyping classification results. During model building, to quantify the impact of different input features on diagnostic results, a feature importance evaluation formula is introduced:

[0045]

[0046] In the above formula, I(f) i ) represents the i-th feature f i Importance indicator, A j This represents the diagnostic accuracy of the model for the j-th sample. This formula represents the diagnostic accuracy of the model for the j-th sample after removing the i-th feature, where n represents the total number of samples evaluated. The purpose of this formula is to compare the change in model accuracy before and after removing a certain feature, thereby identifying the features that contribute most to the subtype diagnosis (such as bone marrow smear cell morphology characteristics and specific fusion gene detection results). These highly important features are then used as the core inputs to the model, reducing the impact of redundant features on the model's computational efficiency. For example, this formula calculates that the importance index of the proportion of primitive cells in bone marrow smears is 0.82, far higher than other clinical features; therefore, this feature is given special attention in the initial screening model.

[0047] The target prediction model is built based on a logistic regression algorithm, taking gene variation information from molecular data as input and outputting potential therapeutic targets. The treatment plan recommendation model is a three-dimensional matching model, taking patient target information, targeted drug information, and individual patient information (age, liver and kidney function, allergy history, etc.) as input, and generating personalized treatment plans by calculating the patient-drug match. To quantify the degree of patient-target drug match, a matching degree calculation formula is introduced:

[0048] M = α·T + β·S + γ·P

[0049] In the above formula, M represents the matching degree between the patient and a certain targeted drug (the value ranges from 0 to 1, and the closer to 1, the higher the matching degree), T represents the matching coefficient between the drug target and the patient's gene mutation site (if the drug target and the patient's mutation site are completely matched, T = 1; if they are partially matched, T = 0.5; if they are not matched, T = 0), S represents the clinical efficacy rate coefficient of the drug for this type of patient (derived from historical clinical data statistics, the higher the efficacy rate, the larger S is, and the value ranges from 0 to 1), P represents the compatibility coefficient between the patient's physical condition and the conditions for drug use (if the patient has normal liver and kidney function and no history of drug allergy, P = 1; if there is a slight contraindication, P = 0.6; if there is a serious contraindication, P = 0), α, β, and γ are the weights of the three coefficients, and α + β + γ = 1 (in this embodiment, α = 0.5, β = 0.3, and γ = 0.2, which can be adjusted according to clinical needs). The function of this formula is to transform qualitative information from three dimensions—target matching, clinical efficacy, and patient suitability—into quantitative matching indicators. The model recommends targeted drugs to patients based on the M value from high to low, and automatically warns that the drug is not suitable for the patient when the M value is below 0.3.

[0050] The model training unit employs federated learning technology, combining data from the hematology departments of three top-tier hospitals for multi-center model training. Each hospital, acting as a participating node in the federated learning process, only uploads the model gradient information from its local data to the federated learning server, rather than the raw data, effectively protecting patient privacy. During training, a mini-batch stochastic gradient descent algorithm is used to optimize model parameters. The batch size for each training iteration is set to 32 samples, and the initial learning rate is set to 0.001, gradually decreasing with each training iteration. The model optimization unit supports incremental learning; when new case data emerges, it is not necessary to retrain the entire model. Only some parameters are updated, with an update frequency of once a week, ensuring that model performance continuously improves with the accumulation of clinical data.

[0051] Implementation of the application layer:

[0052] The application layer includes intelligent diagnosis module, treatment plan formulation module, efficacy monitoring module, and scientific research support module, providing exclusive functions for users with different roles.

[0053] The intelligent diagnostic module includes a data upload unit, a multimodal analysis unit, and a report generation unit. Clinicians upload patient data such as bone marrow smear images and gene testing reports through the data upload unit, supporting both batch and single-case upload modes. Upon receiving the data, the multimodal analysis unit automatically invokes a subtyping diagnostic model for comprehensive analysis. The analysis process includes three sub-steps: image feature extraction, gene mutation identification, and clinical information matching, with the entire analysis taking no more than 15 minutes. Based on the model's analysis results, the report generation unit automatically generates a diagnostic report containing basic patient information, multi-dimensional data interpretation, subtyping results, confidence scores, and similar case references. Doctors can edit and modify the report before printing or exporting. For example, for a patient suspected of having acute myeloid leukemia (AML), the intelligent diagnostic module analyzes the proportion of primitive cells (35%) and the positive FLT3 gene mutation in the bone marrow smear image, combined with the patient's clinical symptoms such as fever and fatigue, outputting a subtyping result of "Acute Myeloid Leukemia M2" with a confidence score of 0.96, and recommends three similar cases for the doctor's reference.

[0054] The treatment plan formulation module, based on the typing results from the intelligent diagnostic module and the output of the target prediction model, calls the treatment plan recommendation model to generate personalized targeted therapy plans. The plan includes the recommended targeted drug name, dosage, dosing frequency, and treatment course, along with the matching degree (M-value), clinical efficacy data, and possible adverse reactions. Doctors can manually adjust the plan parameters based on clinical experience. When adjusting the dosage, the system automatically recalculates the M-value and the probability of adverse reactions. This module integrates with the hospital pharmacy system, supporting one-click queries of recommended drug inventory. If the drug is in sufficient stock, an electronic prescription can be directly generated and sent to the pharmacy; if the drug is out of stock, similar alternative drugs are automatically recommended.

[0055] The efficacy monitoring module includes a follow-up data collection unit, an efficacy evaluation unit, and a treatment plan adjustment early warning unit. The follow-up data collection unit acquires post-treatment follow-up data in two ways: first, doctors manually enter the patient's re-examination results; second, it automatically retrieves objective indicators such as blood routine tests and gene testing through an interface with the hospital's laboratory system. The efficacy evaluation unit constructs an efficacy evaluation model based on the follow-up data, assessing efficacy by comparing changes in key indicators before and after treatment. For example, the efficacy improvement rate is calculated using the following formula:

[0056] R = (V0 - V) t ) / V0×100%

[0057] In the above formula, R represents the improvement rate of a key indicator, V0 represents the value of that indicator before treatment, and V t This represents the value of the indicator after treatment time t. The formula quantifies the improvement of the patient's indicator after treatment. For example, if the proportion of bone marrow blast cells V0 before treatment was 35%, and V0 after one course of treatment... t =8%, then R = (35-8) / 35×100%≈77.1%, indicating significant efficacy. The efficacy assessment unit integrates the R values ​​of multiple key indicators and outputs an overall efficacy evaluation (complete remission, partial remission, stable, progression). When a negative R value is detected for a certain indicator (i.e., the indicator deteriorates) or a new gene mutation is found in the patient, the treatment plan adjustment warning unit issues a warning to the attending physician via pop-up window and SMS, and recommends an adjusted treatment plan.

[0058] The research support module stores anonymized diagnostic and treatment data in a structured manner, allowing researchers to filter data by keywords such as "disease type, subtype, and treatment plan." It provides data statistical analysis functions, automatically generating charts and graphs needed for research, such as case count statistics and efficacy distribution histograms. It also offers an open model training interface, allowing researchers to use platform data to train new research models and promote innovative research in hematological disease diagnosis and treatment technologies.

[0059] Implementation of the security module:

[0060] As described in claims 1 and 10, the security module includes a data encryption unit, an access control unit, and an operation log tracing unit, providing full-process security protection for the system.

[0061] The data encryption unit employs a dual encryption strategy of "transmission encryption + storage encryption." During data transmission between different levels, Secure Sockets Layer (SSL) protocol is used for encryption to prevent data theft or tampering during transmission. When storing data, advanced encryption standards are used to encrypt the original data. The encryption keys are managed by designated personnel from the hospital's information department and changed periodically. The access control unit uses a role-based access control model, setting different access permissions for different roles such as clinicians, laboratory technicians, researchers, and administrators. For example, clinicians can only view and manipulate data of patients under their care, researchers can only access anonymized aggregated data, and administrators have all permissions configured by the system. The operation log traceability unit records the operation behavior of all users within the system, including the operator, operation time, operation content, and operation result. The log data is immutable and retained for at least 5 years. In the event of a data security incident, the cause of the incident and the responsible party can be quickly traced through the operation logs.

[0062] System workflow example:

[0063] The following example, using a newly diagnosed hematological patient, illustrates the complete workflow of this system in detail:

[0064] Step 1: Data Acquisition and Integration. The data acquisition unit of the data layer obtains the patient's clinical data (age 45, chief complaint of fatigue and gingival bleeding for 1 month, blood routine test showing abnormally high white blood cell count) through the hospital system interface, acquires bone marrow smear images through a digital slide scanner, and obtains gene testing data (positive FLT3 gene mutation) through a gene sequencer. After the data docking unit standardizes and processes this data, it is classified and stored by the data storage unit.

[0065] Step 2: Intelligent Subtype Diagnosis. Clinicians submit patient data through the data upload unit of the application-layer intelligent diagnosis module. The multimodal analysis unit calls the subtype diagnosis model of the algorithm layer. First, the initial screening model determines that the patient belongs to the acute leukemia category. Then, the subtype classification model, combined with FLT3 gene mutation data, determines that it is acute myeloid leukemia M2. The report generation unit outputs a diagnostic report with a confidence score of 0.96.

[0066] Step 3: Targeted Therapy Regimen Development. The treatment regimen development module uses a target prediction model to identify the FLT3 gene as the core therapeutic target. It then uses a treatment regimen recommendation model to calculate the patient's match with the targeted drug midostaurin using a matching formula. The result is M = 0.92 (where T = 1, S = 0.85, P = 1, α = 0.5, β = 0.3, γ = 0.2). A treatment regimen centered on midostaurin is generated, including the dosage (50 mg / dose), dosing frequency (twice daily), and treatment duration (28 days per course). The clinical efficacy rate is noted as 78%, and possible adverse reactions include nausea and vomiting.

[0067] Step 4: Dynamic monitoring of efficacy. After the patient completes one course of treatment, the follow-up data collection unit of the efficacy monitoring module obtains the patient's re-examination data (the proportion of bone marrow blast cells drops to 8%, with no new gene mutations). The efficacy assessment unit calculates the improvement rate of blast cell proportion R≈77.1% using the efficacy improvement rate formula. Based on other indicators, the efficacy is judged to be complete remission. In subsequent follow-ups, if the patient's blast cell proportion is detected to rise back to 15%, the regimen adjustment early warning unit immediately issues an early warning and recommends an adjustment regimen for combination chemotherapy.

[0068] Step 5: Comprehensive Security. Throughout the entire process, the data encryption unit of the security module encrypts patient data at all times, the access control unit ensures that only the attending physician can access the patient's complete data, and the operation log traceability unit records every data upload and treatment plan adjustment operation by the physician, ensuring data security and operational compliance.

[0069] Implementation Results Explanation:

[0070] The system in this embodiment underwent a one-year clinical trial in the hematology departments of three tertiary hospitals, involving a total of 500 patients with hematological diseases. The trial results showed that the system's typing and diagnostic accuracy reached over 95%, improving efficiency by 80% compared to traditional manual diagnosis (traditional diagnosis takes an average of 1-2 days, while this system takes an average of 15 minutes). For patients using the system's recommended treatment plan, the treatment effectiveness rate increased by 15% compared to traditional methods, and the incidence of adverse reactions decreased by 10%. Simultaneously, the system's security module effectively prevented three data misoperation incidents, ensuring the privacy and security of patient data.

[0071] Beneficial effects:

[0072] Improving the accuracy and efficiency of subtyping diagnosis: This system's subtyping diagnosis model uses a feature importance assessment formula to screen core diagnostic features. Combined with a two-tiered diagnostic logic of initial screening and subtyping models, it achieves precise classification of hematological diseases from broad categories to subtypes, significantly improving accuracy compared to traditional manual diagnosis. Simultaneously, the multimodal analysis unit can complete the entire analysis process within 15 minutes, drastically reducing the time required for traditional diagnosis (1-2 days), saving patients valuable treatment time. This is particularly suitable for scenarios with high requirements for timely diagnosis and treatment, such as acute hematological diseases.

[0073] Achieving personalized and scientific targeted therapy: The treatment recommendation model transforms qualitative indicators such as target matching, clinical efficacy, and patient suitability into quantitative matching degrees through a matching degree calculation formula, ensuring that the recommended targeted therapy is more tailored to the individual patient's situation. The model can automatically warn of drugs with low matching degrees, reducing the risk of ineffective medication. Simultaneously, it integrates with the pharmacy system to achieve seamless connection between treatment plan and prescription, improving treatment effectiveness and reducing the incidence of adverse reactions, thus solving the problem of the traditional "one-size-fits-all" approach to treatment.

[0074] Constructing a closed-loop management system throughout the entire process to dynamically ensure treatment effectiveness: The efficacy monitoring module at the application layer quantifies treatment effectiveness through the efficacy improvement rate formula and tracks changes in the patient's condition in real time by combining follow-up data. When indicators deteriorate or new gene mutations are detected, timely warnings are issued and adjustments to the treatment plan are recommended, forming a closed-loop diagnosis and treatment system of "diagnosis-treatment-monitoring-adjustment". This avoids the drawbacks of delayed efficacy judgment and passive treatment plan adjustments in traditional treatments and helps maintain long-term treatment effects.

[0075] Strengthening data security and privacy protection: The security module's dual encryption strategy, hierarchical access control, and operation log traceability comprehensively safeguard the security of multi-dimensional medical data, including clinical, imaging, and molecular data. In particular, the federated learning technology employed at the algorithm layer transmits only gradient information, not the raw data, during multi-center model training. This ensures data privacy while enabling the collaborative sharing of high-quality medical resources, complying with medical data security standards.

[0076] Promoting the mutual empowerment of clinical practice and scientific research: The scientific research support module performs structured processing on the desensitized diagnostic and treatment data, providing a wealth of high-quality data resources for hematological disease-related research. Its open model training interface can help researchers develop new models, explore new targets, and promote innovation in diagnostic and treatment technologies. In turn, scientific research results can feed back into clinical practice, updating the system algorithm through the model optimization unit to further improve the level of clinical diagnosis and treatment, forming a virtuous cycle of "clinical application - scientific research innovation - technological iteration".

[0077] Lowering the barriers to diagnosis and treatment and improving the accessibility of medical resources: The system reduces reliance on individual doctors' experience through intelligent analysis, allowing primary healthcare institutions to obtain the same level of diagnostic and treatment support as large hospitals, which helps to balance the differences in medical standards across different regions. At the same time, multi-terminal access methods, including web and mobile devices, facilitate doctors to conduct diagnosis and treatment anytime, anywhere, expanding the system's applicability and service scope.

[0078] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases, characterized by: The system comprises a data layer, an algorithm layer, an application layer, and a security module. These layers communicate bidirectionally with each other. The data layer collects, integrates, and securely stores multi-dimensional medical data, providing standardized data support for the entire system. The algorithm layer constructs a multimodal fusion model library based on the standardized data provided by the data layer, trains the model using specific training techniques, and continuously optimizes it, providing algorithmic support for diagnostic and treatment decisions. The application layer, relying on the algorithm outputs of the algorithm layer, provides dedicated functional modules for clinicians, laboratory technicians, and researchers, enabling specific applications related to diagnosis, treatment, and research. The security module provides end-to-end security protection for the multi-dimensional medical data in the data layer, the model data in the algorithm layer, and the operational data in the application layer, ensuring the data security and operational compliance of the entire system.

2. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 1, characterized in that: The data layer includes a data acquisition unit, a data docking unit, and a data storage unit. The data acquisition unit is used to collect multi-dimensional medical data. The data docking unit docks with the hospital information system through a health information exchange standard protocol to achieve the interoperability of multi-dimensional medical data. The data storage unit is used to classify and securely store the multi-dimensional medical data integrated by the data docking unit.

3. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 2, characterized in that: The multidimensional medical data includes clinical data, imaging data, molecular data, and drug data. The clinical data includes patient history, signs, and laboratory test results. The imaging data includes bone marrow smear images and immunohistochemical slide images. The molecular data includes gene sequencing results and fusion gene detection results. The drug data includes targeted drug target information and targeted drug resistance mechanism information.

4. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 1, characterized in that: The algorithm layer includes a model building unit, a model training unit, and a model optimization unit. The model building unit is used to build a subtyping diagnosis model, a target prediction model, and a treatment plan recommendation model to form a multimodal fusion model library. The model training unit completes multi-center model training through federated learning technology while protecting data privacy. The model optimization unit supports incremental optimization of the subtyping diagnosis model, the target prediction model, and the treatment plan recommendation model using new case data.

5. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 4, characterized in that: The classification diagnostic model is a two-layer diagnostic model, which includes a primary screening model and a subtype classification model. The primary screening model is used to distinguish different major categories of hematological diseases such as acute leukemia, lymphoma, and multiple myeloma. The subtype classification model is used to accurately classify specific major categories of hematological diseases identified by the primary screening model into subtypes.

6. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 4, characterized in that: The treatment plan recommendation model is a three-dimensional matching model, which is constructed based on patient target information, targeted drug information, and individual patient information. The three-dimensional matching model is used to generate personalized targeted treatment plans and provide early warning of medication risks in the targeted treatment plans.

7. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 1, characterized in that: The application layer includes an intelligent diagnosis module, a treatment plan formulation module, an efficacy monitoring module, and a research support module. The intelligent diagnosis module is used to generate a subtyping diagnosis report based on the output of the subtyping diagnosis model. The treatment plan formulation module is used to output a targeted treatment plan based on the output of the treatment plan recommendation model. The efficacy monitoring module is used to dynamically track the treatment effect of patients. The research support module is used to provide research data services for hematological disease-related research.

8. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 7, characterized in that: The intelligent diagnostic module includes a data uploading unit, a multimodal analysis unit, and a report generation unit. The data uploading unit is used to receive clinical data and various test data uploaded by clinicians. The multimodal analysis unit is used to call the subtyping diagnostic model to perform comprehensive analysis on the uploaded data. The report generation unit is used to output a diagnostic report containing clear subtyping results.

9. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 7, characterized in that: The efficacy monitoring module includes a follow-up data acquisition unit, an efficacy evaluation unit, and a treatment plan adjustment early warning unit. The follow-up data acquisition unit is used to acquire various follow-up data of patients after treatment. The efficacy evaluation unit is used to construct an efficacy evaluation model based on the follow-up data and complete the efficacy evaluation. The treatment plan adjustment early warning unit is used to issue an early warning for treatment plan adjustment when a decline in patient efficacy or target mutation is detected.

10. The integrated system for intelligent and precise subtyping diagnosis and targeted therapy of hematological diseases according to claim 1, characterized in that: The security module includes a data encryption unit, an access control unit, and an operation log tracing unit. The data encryption unit is used to encrypt all medical data in the system. The access control unit is used to set hierarchical access permissions for users with different roles. The operation log tracing unit is used to record all operations in the system and support the tracing query of operations. It also includes a security module, which comprises a data encryption unit, an access control unit, and a log tracing unit. The data encryption unit is used to encrypt medical data, the access control unit is used to set hierarchical access permissions, and the log tracing unit is used to record all operations and support tracing queries.

Images