Medical detection and imaging diagnosis system based on cloud intelligent analysis

Abstract

The invention discloses a medical detection and image diagnosis system based on cloud intelligent analysis, and relates to the technical field of medical information. Therefore, the task allocation scheme can comprehensively consider the real-time load, the computing power and the dynamic change of the task queue of each server, and finally finds an allocation strategy close to the global optimum, so that the load balance degree of the servers is greatly improved, the resource waste is avoided, and the task allocation efficiency is improved through closed-loop feedback and continuous learning. The method can keep up with updating of medical knowledge and changes of clinical practice in a self-adaptive mode, and the optimal auxiliary diagnosis level is kept all the time.

Description

Technical Field

[0001] This application relates to the field of medical information technology, and more specifically, to a medical testing and imaging diagnostic system based on cloud-based intelligent analysis. Background Technology

[0002] With the rapid development of medical imaging technology, imaging data such as CT (Computed Tomography), MRI (Magnetic Resonance Imaging), and X-rays are playing an increasingly important role in clinical diagnosis. However, the traditional imaging diagnostic process relies heavily on the personal experience and subjective judgment of radiologists, facing numerous challenges such as low diagnostic efficiency, heavy workload for doctors, poor diagnostic consistency among different doctors, and uneven distribution of medical resources.

[0003] In recent years, deep learning-based image segmentation technology has achieved remarkable results in medical image analysis, automatically identifying and delineating regions of interest such as lesions and organs, providing doctors with quantitative analysis data. However, existing technologies still face the following bottlenecks in practical application: Computing resource limitations: High-end medical image analysis models typically require powerful computing resources (such as high-performance GPUs (Graphics Processing Units)), and individual medical institutions cannot afford the deployment and maintenance costs.

[0004] Inefficient task scheduling: In a multi-server cloud environment, how to efficiently and evenly distribute massive image diagnostic tasks across servers to maximize overall processing efficiency and reduce task waiting time is a complex optimization problem. Traditional polling or random allocation methods cannot cope with dynamically changing load and task characteristics, easily leading to some servers being overloaded while others are idle, resulting in resource waste and response delays.

[0005] Model performance stagnation and insufficient generalization ability: Pre-trained AI (Artificial Intelligence) models experience a decline in generalization ability and accuracy when faced with data from different hospitals, different equipment, and different diseases. Once deployed, the model struggles to continuously optimize and iterate based on new data encountered in real-world applications and feedback from doctors, causing its performance to gradually lag behind clinical needs.

[0006] Lack of human-machine collaborative closed loop: Most existing AI-assisted diagnostic systems provide one-time analysis results and lack an effective mechanism to systematically collect doctors' corrections and confirmations and use them for model retraining, thus failing to form a virtuous cycle of AI analysis - doctor review - model optimization. Summary of the Invention

[0007] This application aims to provide a medical testing and imaging diagnostic system based on cloud-based intelligent analysis, which is designed to solve the problem of low scheduling efficiency of massive medical imaging diagnostic tasks in the cloud environment, thereby improving diagnostic efficiency and intelligence.

[0008] This application provides a cloud-based intelligent analysis-based medical testing and imaging diagnostic system, including: The data acquisition and uploading module is configured to acquire detection and diagnosis tasks from at least one medical terminal device and upload the detection and diagnosis tasks to the cloud data platform module; wherein, the detection and diagnosis tasks include medical imaging data and patient metadata associated with the medical imaging data; The cloud data platform module is configured to queue the detection and diagnosis tasks and use a window acquisition and numerical allocation mapping method to allocate the queued detection and diagnosis tasks to the server. The intelligent analysis engine module is deployed on the server and configured to call a pre-deployed image segmentation model to analyze incoming detection and diagnosis tasks and obtain task analysis results. The interactive diagnostic module is configured to transmit the task analysis results and the detection and diagnostic task to a device designated by the doctor for display, and to obtain the processing information fed back by the doctor through the designated device; wherein, the processing information is confirmation information or correction segmentation information; The closed-loop feedback and continuous learning module is configured to collect the task analysis results, detection and diagnosis tasks, and processing information, and to iteratively optimize the image segmentation model pre-deployed on the intelligent analysis engine module using the task analysis results, detection and diagnosis tasks, and processing information.

[0009] In one possible implementation, the cloud data platform module includes a timing adjustment submodule, a window acquisition submodule, and a value allocation and mapping submodule. The timing adjustment submodule is configured to configure timing for the detection and diagnosis tasks and send the detection and diagnosis tasks into the window acquisition submodule in ascending order of timing. The window acquisition submodule is configured to acquire a fixed number of detection and diagnosis tasks in sequence after each task allocation cycle or after the previous batch of target detection and diagnosis tasks has been allocated, thus obtaining the target detection and diagnosis tasks. The numerical allocation mapping submodule is configured to use a numerical allocation mapping algorithm to allocate the target detection and diagnosis tasks collected by the window acquisition submodule, determine the server corresponding to each target detection and diagnosis task, and perform task allocation based on the server corresponding to the target detection and diagnosis task.

[0010] In one possible implementation, the timing adjustment submodule is specifically configured as follows: For different detection and diagnostic tasks, different initial timing sequences are assigned to obtain the timing sequence corresponding to the detection and diagnostic task. After each clock cycle, the timing of the detection and diagnosis task is reduced by one unit, and all detection and diagnosis tasks are sent to the window acquisition submodule in ascending order of timing.

[0011] In one possible implementation, the numerical allocation mapping submodule is specifically configured as follows: Assign consecutive numerical ranges to the servers, and the numerical ranges corresponding to all servers form a total mapping range; Within this mapping range, a corresponding mapping value is randomly generated for each target detection and diagnosis task collected by the window acquisition submodule, and all target detection and diagnosis tasks are encoded into a vector to obtain the task allocation code; Multiple different task allocation codes are repeatedly obtained, and the task allocation codes are optimized multiple times using spiral exploration, order fraction exploration, information entropy collection exploration and fuzzy fluctuation exploration methods in sequence to obtain the target task allocation code; The target task allocation code is decoded to determine the server corresponding to each target detection and diagnosis task, and the task is allocated according to the server corresponding to the target detection and diagnosis task.

[0012] In one possible implementation, the task allocation code is optimized multiple times using a spiral exploration, order fractional exploration, information entropy collection exploration, and fuzzy fluctuation exploration method in sequence to obtain the target task allocation code, including: The task allocation code is optimized to the first order using a spiral exploration method, resulting in a first-order optimized task allocation code. The first-order optimized task allocation code is optimized by second-order optimization using the order fractional exploration method to obtain the second-order optimized task allocation code. The information entropy acquisition and exploration method is used to perform third-order optimization on the second-order optimized task allocation code to obtain the third-order optimized task allocation code; The fuzzy fluctuation exploration method is used to perform fourth-order optimization on the third-order optimized task allocation code to obtain the fourth-order optimized task allocation code; Repeat the first-order to fourth-order optimization until the optimization termination condition is met. Use the task allocation code after the fourth-order optimization in the last optimization process to determine the target task allocation code.

[0013] In one possible implementation, the task allocation code is optimized using a spiral exploration method to obtain the first-order optimized task allocation code as follows: In the formula, For the first i Task assignment code, For the first i A first-order optimized task allocation code Assign codes to the optimal task. It is a natural constant. t The number of optimization attempts is given, and b is the spiral encirclement factor, which is set to 1 or 1.5. The first random number between (0,1) Pi i =1,2,…,N, where N represents the total number of task assignment codes.

[0014] In one possible implementation, the first-order optimized task allocation code is optimized using a fractional-order exploration method to obtain a second-order optimized task allocation code: In the formula, For the first k A first-order optimized task allocation code For the first k A second-order optimized task allocation code for In the t +1 optimization speed during the optimization process for In the t The optimization speed in the next optimization process. for In the t -Optimization speed in the first optimization process for In the t -Optimization speed during the second optimization process for In the t -Optimization speed during the second optimization process k =1,2,…,N, This is the inertia weight, and it is set to 0.85 or 0.95; For fractional order factors, The second random number between (0,1); To remove In addition, other first-order optimized task assignment codes are randomly matched; for and The Euclidean distance between them This is the interaction control coefficient, and it is set to 2 or 3; The average Euclidean distance between the optimal task assignment code and all other first-order optimized task assignment codes. Assign the minimum of the average Euclidean distances corresponding to the encodings of all first-order optimized tasks. Assign the maximum value of the average Euclidean distance corresponding to the encoding of all first-order optimized tasks.

[0015] In one possible implementation, the second-order optimized task allocation code is optimized to a third-order optimization using an information entropy acquisition and exploration method, resulting in the third-order optimized task allocation code as follows: ; ; In the formula, To arrange the second-order optimized task assignment codes in descending order of fitness, the next... j Task assignment coding, For the first j The task allocation code after third-order optimization Assign codes to the optimal task. It is the first learning factor, and is set to 1 or 2; This is the second learning factor, and it is set to 0.65 or 1; The third random number between (0,1) The fourth random number between (0,1) To arrange the second-order optimized task assignment codes in descending order of fitness, the next... j -1 task allocation code, j =1,2,…,N, and j When it is 1, Set to random, second-order optimized task assignment encoding. Step size factor This is the minimum value of the step size factor, and it is set to 0.01; This is the preset maximum number of optimization attempts. This is a balancing factor, set to 0.4 or 0.5; Assign the minimum fitness value corresponding to the encoding of the task after second-order optimization. Assign the maximum fitness value corresponding to the encoding of the task after second-order optimization. For information collection factors, For information entropy, for The corresponding intermediate parameters, for The corresponding fitness level.

[0016] In one possible implementation, the fuzzy fluctuation exploration method is used to perform fourth-order optimization on the third-order optimized task allocation code, resulting in the fourth-order optimized task allocation code as follows: in, For the first h The task allocation code after third-order optimization For the first h A fourth-order optimized task allocation code for Inverse learning encoding within the solution space For fuzzy fluctuation factors, This is the contraction / expansion coefficient, which decreases linearly from 1 to 0 with increasing training iterations; Assign the average historical optimum value corresponding to the encoding to all tasks after third-order optimization. As a control factor for random fluctuation behavior between (0,1), This is the maximum value of the fuzzy fluctuation factor, and it is set to 0.85. This is the minimum value of the fuzzy fluctuation factor, and it is set to 0.01. Pi T This indicates the maximum number of optimization attempts preset.

[0017] In one possible implementation, the pre-deployed image segmentation model is set to U-Net or U-Net++.

[0018] Beneficial effects: This application provides a cloud-based intelligent analysis-based medical testing and imaging diagnostic system. It employs a window acquisition and numerical allocation mapping method to distribute queued testing and diagnostic tasks to servers. This allows the task allocation scheme to comprehensively consider the real-time load, computing power, and dynamic changes in the task queue of each server, ultimately finding a near-globally optimal allocation strategy. This greatly improves the load balancing of the servers and avoids resource waste. Furthermore, through closed-loop feedback and continuous learning, it can adaptively keep up with the updates in medical knowledge and changes in clinical practice, always maintaining the best level of assisted diagnosis. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the structure of a cloud-based intelligent analysis-based medical testing and imaging diagnostic system proposed in one embodiment of this application.

[0021] Explanation of reference numerals in the attached figures: 101-Data Acquisition and Upload Module, 102-Cloud Data Platform Module, 103-Intelligent Analysis Engine Module, 104-Interactive Diagnosis Module, 105-Closed-Loop Feedback and Continuous Learning Module. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] like Figure 1 As shown, this application provides a cloud-based intelligent analysis-based medical testing and imaging diagnostic system, including: The data acquisition and uploading module 101 is configured to acquire detection and diagnosis tasks from at least one medical terminal device and upload the detection and diagnosis tasks to the cloud data platform module; wherein, the detection and diagnosis tasks include medical image data and patient metadata associated with the medical image data; For example, the data acquisition and upload module 101 is deployed on the hospital's PACS (Picture Archiving and Communication System) or RIS (Radiology Information System), or as a standalone software client. After a doctor or technician completes a CT or MRI scan, the module automatically acquires medical image data and patient metadata (such as age, gender, medical history, etc.) synchronized from the HIS (Hospital Information System), packages this data into a diagnostic task, and uploads it to the cloud data platform module 102 via encrypted or dedicated network lines.

[0024] The cloud data platform module 102 is configured to queue the detection and diagnosis tasks and use a window acquisition and numerical allocation mapping method to allocate the queued detection and diagnosis tasks to the server. For example, detection and diagnostic tasks are received from the data acquisition and upload module 101 and placed into a global task queue. An initial timing value can be assigned to each new task (e.g., based on task priority or arrival time). Subsequently, the system decrements the timing values ​​of all tasks in the queue by 1 at a fixed clock cycle (e.g., per second). Thus, tasks with longer waiting times have smaller timing values ​​and higher priority.

[0025] When a task allocation cycle (e.g., 5 seconds) is reached, or when all tasks in the previous batch have been allocated, a fixed number (e.g., 50) of tasks can be sequentially retrieved from the head of the queue to form a target detection and diagnosis task batch. Then, a numerical allocation mapping method is used to even out the task load and improve the processing efficiency of the detection and diagnosis tasks.

[0026] The intelligent analysis engine module 103 is deployed on the server and configured to call a pre-deployed image segmentation model to analyze incoming detection and diagnosis tasks and obtain task analysis results. For example, the intelligent analysis engine module 103 can be deployed on a cloud server. When the server receives an assigned detection and diagnosis task, the intelligent analysis engine module 103 is activated, calling a pre-deployed and trained U-Net++ image segmentation model to process the medical image data, automatically segmenting the lesion region and generating task analysis results. It can also calculate quantitative features such as area, volume, and density, generating task analysis results that include segmentation masks and quantification reports.

[0027] The interactive diagnostic module 104 is configured to transmit the task analysis results and the detection and diagnostic task to a device designated by the doctor for display, and to obtain the processing information fed back by the doctor through the designated device; wherein, the processing information is confirmation information or correction segmentation information; The interactive diagnostic module 104 can transmit the task analysis results and diagnostic tasks to a device designated by the doctor for display. After logging into their account, the doctor can see a list of tasks awaiting review. Clicking on a task will display the original medical image and the AI-generated task analysis results (such as a colored segmented area overlaid on the original image) side by side on the terminal. The doctor can perform operations such as zooming and panning. If they approve of the AI ​​results, they can click the confirmation button, and the system will record this as confirmation information. If they believe that the AI ​​segmentation is incorrect, the doctor can use the drawing tools provided on the terminal to manually correct it, and the system will record this as corrected segmentation information.

[0028] The closed-loop feedback and continuous learning module 105 is configured to collect the task analysis results, detection and diagnosis tasks, and processing information, and to iteratively optimize the image segmentation model pre-deployed on the intelligent analysis engine module using the task analysis results, detection and diagnosis tasks, and processing information.

[0029] The closed-loop feedback and continuous learning module 105 runs continuously in the background, collecting all completed task data, including: original detection and diagnosis tasks (images + metadata), AI-generated task analysis results, and doctor feedback (confirmation or correction information). This data is cleaned, organized, and stored in a dedicated continuous learning database. When the database accumulates a certain amount of new data (e.g., 1000 valid feedbacks per week), or reaches a preset time period, the module automatically triggers the model retraining process. This new data is used to incrementally train or fully fine-tune the current U-Net++ model, generating a new model version with superior performance. After performance metrics verification, the new model is seamlessly deployed to the intelligent analysis engine module 103, replacing the old version, thus completing a system self-evolution.

[0030] Doctors' professional feedback (whether confirming or correcting) is the highest quality and most valuable labeled data. The system can systematically collect this data and correlate it with the original images and preliminary AI results, forming a continuously growing, timely, and diverse high-quality training dataset. When the model frequently makes mistakes on a specific type of lesion and is corrected by doctors, these corrected samples will be used primarily for model retraining, thereby accurately compensating for the model's shortcomings. This mechanism breaks the deadlock of traditional AI model performance stagnation. Key performance indicators of the image segmentation model, such as accuracy, recall, and Dice coefficient, will steadily improve with the continuous operation of the system; the model's generalization ability to different hospitals, different equipment, and different rare cases is significantly enhanced; the system can adaptively keep up with the updates in medical knowledge and changes in clinical practice, always maintaining the best level of assisted diagnosis.

[0031] In one possible implementation, the cloud data platform module 102 includes a timing adjustment submodule, a window acquisition submodule, and a value allocation mapping submodule. The timing adjustment submodule is configured to configure timing for the detection and diagnosis tasks and send the detection and diagnosis tasks into the window acquisition submodule in ascending order of timing. The window acquisition submodule is configured to acquire a fixed number of detection and diagnosis tasks in sequence after each task allocation cycle or after the previous batch of target detection and diagnosis tasks has been allocated, thus obtaining the target detection and diagnosis tasks. The numerical allocation mapping submodule is configured to use a numerical allocation mapping algorithm to allocate the target detection and diagnosis tasks collected by the window acquisition submodule, determine the server corresponding to each target detection and diagnosis task, and perform task allocation based on the server corresponding to the target detection and diagnosis task.

[0032] In one possible implementation, the timing adjustment submodule is specifically configured as follows: For different detection and diagnostic tasks, different initial timing sequences are assigned to obtain the timing sequence corresponding to the detection and diagnostic task. For example, when a detection and diagnosis task arrives, different initial timing sequences can be assigned to different types of segmentation tasks. The more urgent the segmentation task corresponding to the detection and diagnosis task, the lower the timing sequence, indicating that faster processing efficiency can be obtained. The initial timing sequence can contain multiple clock cycles, thereby realizing timing calculation.

[0033] After each clock cycle, the timing of the detection and diagnosis task is reduced by one unit, and all detection and diagnosis tasks are sent to the window acquisition submodule in ascending order of timing.

[0034] Timing adjustment and window acquisition ensure the fairness and continuity of task processing, while batch processing improves allocation efficiency.

[0035] In one possible implementation, the numerical allocation mapping submodule is specifically configured as follows: Assign consecutive numerical ranges to the servers, and the numerical ranges corresponding to all servers form a total mapping range; Within this mapping range, a corresponding mapping value is randomly generated for each target detection and diagnosis task collected by the window acquisition submodule, and all target detection and diagnosis tasks are encoded into a vector to obtain the task allocation code; For example, if there are 5 servers, the interval (0, 1] can be evenly divided into 5 intervals, such as (0, 0.2], (0.2, 0.4], (0.4, 0.6], (0.6, 0.8], and (0.8, 1]. When the mapped value is in different intervals, it is assigned to different servers, thus enabling encoding and decoding.

[0036] Multiple different task allocation codes are repeatedly obtained, and the task allocation codes are optimized multiple times using spiral exploration, order fraction exploration, information entropy collection exploration and fuzzy fluctuation exploration methods in sequence to obtain the target task allocation code; The target task allocation code is decoded to determine the server corresponding to each target detection and diagnosis task, and the task is allocated according to the server corresponding to the target detection and diagnosis task.

[0037] This multi-stage, multi-mechanism collaborative optimization enables the task allocation scheme to comprehensively consider the real-time load, computing power, and dynamic changes in the task queue of each server, ultimately finding a near-globally optimal allocation strategy. Server load balancing is greatly improved, avoiding resource waste; the system's average task waiting time and maximum task waiting time are significantly reduced; and the overall task processing throughput (the number of diagnostic tasks completed per unit time) is significantly increased.

[0038] In one possible implementation, the task allocation code is optimized multiple times using a spiral exploration, order fractional exploration, information entropy collection exploration, and fuzzy fluctuation exploration method in sequence to obtain the target task allocation code, including: The task allocation code is optimized to the first order using a spiral exploration method, resulting in a first-order optimized task allocation code. The first-order optimized task allocation code is optimized by second-order optimization using the order fractional exploration method to obtain the second-order optimized task allocation code. The information entropy acquisition and exploration method is used to perform third-order optimization on the second-order optimized task allocation code to obtain the third-order optimized task allocation code; The fuzzy fluctuation exploration method is used to perform fourth-order optimization on the third-order optimized task allocation code to obtain the fourth-order optimized task allocation code; Repeat the first-order to fourth-order optimization until the optimization termination condition is met. Use the task allocation code after the fourth-order optimization in the last optimization process to determine the target task allocation code.

[0039] Optionally, after each change to the task assignment code, an over-limit handling process can be performed to ensure that the task assignment code is always valid.

[0040] In one possible implementation, the task allocation code is optimized using a spiral exploration method to obtain the first-order optimized task allocation code as follows: In the formula, For the first i Task assignment code, For the first i A first-order optimized task allocation code Assign codes to the optimal task. It is a natural constant. t The number of optimization attempts is given, and b is the spiral encirclement factor, which is set to 1 or 1.5. The first random number between (0,1) Pi i =1,2,…,N, where N represents the total number of task assignment codes.

[0041] The spiral exploration ensures that the search process always revolves around the current optimal solution, has a strong local exploitation capability, and can quickly converge to the vicinity of a high-quality solution.

[0042] In one possible implementation, the first-order optimized task allocation code is optimized using a fractional-order exploration method to obtain a second-order optimized task allocation code: In the formula, For the first k A first-order optimized task allocation code For the first k A second-order optimized task allocation code for In the t +1 optimization speed during the optimization process for In the t The optimization speed in the next optimization process. for In the t -Optimization speed in the first optimization process for In the t -Optimization speed during the second optimization process for In the t -Optimization speed during the second optimization process k =1,2,…,N, This is the inertia weight, and it is set to 0.85 or 0.95; For fractional order factors, The second random number between (0,1); To remove In addition, other first-order optimized task assignment codes are randomly matched; for and The Euclidean distance between them This is the interaction control coefficient, and it is set to 2 or 3; The average Euclidean distance between the optimal task assignment code and all other first-order optimized task assignment codes. Assign the minimum of the average Euclidean distances corresponding to the encodings of all first-order optimized tasks. Assign the maximum value of the average Euclidean distance corresponding to the encoding of all first-order optimized tasks.

[0043] Fractional exploration introduces the concept of fractional calculus, which makes the influence of historical velocity and position information on the current update memorized and nonlocal, enhancing the algorithm's ability to escape local optima and explore the global world.

[0044] In one possible implementation, the second-order optimized task allocation code is optimized to a third-order optimization using an information entropy acquisition and exploration method, resulting in the third-order optimized task allocation code as follows: ; ; In the formula, To arrange the second-order optimized task assignment codes in descending order of fitness, the next... j Task assignment coding, For the first j The task allocation code after third-order optimization Assign codes to the optimal task. It is the first learning factor, and is set to 1 or 2; This is the second learning factor, and it is set to 0.65 or 1; The third random number between (0,1) The fourth random number between (0,1) To arrange the second-order optimized task assignment codes in descending order of fitness, the next... j -1 task allocation code, j =1,2,…,N, and j When it is 1, Set to random, second-order optimized task assignment encoding. Step size factor This is the minimum value of the step size factor, and it is set to 0.01; This is the preset maximum number of optimization attempts. This is a balancing factor, set to 0.4 or 0.5; Assign the minimum fitness value corresponding to the encoding of the task after second-order optimization. Assign the maximum fitness value corresponding to the encoding of the task after second-order optimization. For information collection factors, For information entropy, for The corresponding intermediate parameters, for The corresponding fitness level.

[0045] Information entropy collection and exploration utilizes information entropy to measure population diversity and dynamically adjusts the learning strategy. When population diversity is low (easily trapped in local optima), exploration is enhanced; when diversity is high, development is enhanced, thus achieving an intelligent balance between exploration and development.

[0046] In one possible implementation, the fuzzy fluctuation exploration method is used to perform fourth-order optimization on the third-order optimized task allocation code, resulting in the fourth-order optimized task allocation code as follows: in, For the first h The task allocation code after third-order optimization For the first h A fourth-order optimized task allocation code for Inverse learning encoding within the solution space For fuzzy fluctuation factors, This is the contraction / expansion coefficient, which decreases linearly from 1 to 0 with increasing training iterations; Assign the average historical optimum value corresponding to the encoding to all tasks after third-order optimization. As a control factor for random fluctuation behavior between (0,1), This is the maximum value of the fuzzy fluctuation factor, and it is set to 0.85. This is the minimum value of the fuzzy fluctuation factor, and it is set to 0.01. Pi T This indicates the maximum number of optimization attempts preset.

[0047] Fuzzy fluctuation exploration combines a reverse learning strategy with fuzzy logic. By introducing random perturbations and reverse exploration in the later stages of optimization, it effectively prevents the algorithm from converging too early and provides a last chance to escape the local optimum trap.

[0048] Optionally, a greedy strategy can be used to control various exploration methods to further improve the convergence speed of the algorithm.

[0049] In one possible implementation, the pre-deployed image segmentation model is set to U-Net or U-Net++.

[0050] It is worth noting that the image segmentation model can also be set up as other deep learning models to help doctors improve diagnostic efficiency; for example, it can also be used as an image classification model.

[0051] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0052] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0056] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0057] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A cloud-based intelligent analysis of medical detection and image diagnosis system, characterized in that, The application relates to a cloud-side data platform for medical diagnosis, which comprises the following modules: a data acquisition and uploading module configured to acquire detection and diagnosis tasks from at least one medical terminal device and upload the detection and diagnosis tasks to a cloud-side data platform module; wherein the detection and diagnosis tasks comprise medical image data and patient metadata associated with the medical image data; the cloud-side data platform module is configured to queue the detection and diagnosis tasks and distribute the queued detection and diagnosis tasks to servers by using a window acquisition and numerical value distribution mapping method; an intelligent analysis engine module deployed on the servers and configured to analyze the incoming detection and diagnosis tasks by calling a pre-deployed image segmentation model to obtain task analysis results; an interactive diagnosis module configured to transmit the task analysis results and detection and diagnosis tasks to a device designated by a doctor for display and obtain processing information fed back by the doctor through the designated device; wherein the processing information is confirmation information or correction segmentation information; a closed-loop feedback and continuous learning module configured to collect the task analysis results, detection and diagnosis tasks and processing information and use the task analysis results, detection and diagnosis tasks and processing information to iteratively optimize the image segmentation model pre-deployed on the intelligent analysis engine module.

2. The cloud-based intelligent analysis of medical detection and image diagnosis system according to claim 1, wherein, The cloud-side data platform module comprises a time sequence adjustment submodule, a window acquisition submodule and a numerical value distribution mapping submodule; the time sequence adjustment submodule is configured to configure time sequences for the detection and diagnosis tasks and send the detection and diagnosis tasks into the window acquisition submodule in order of increasing time sequences; the window acquisition submodule is configured to acquire a fixed number of detection and diagnosis tasks in order every time a task distribution period elapses or a target detection and diagnosis task acquired in the last batch is distributed, and obtain the target detection and diagnosis task; the numerical value distribution mapping submodule is configured to distribute the target detection and diagnosis task acquired by the window acquisition submodule by using a numerical value distribution mapping algorithm, determine the server corresponding to each target detection and diagnosis task and perform task distribution according to the server corresponding to the target detection and diagnosis task.

3. The cloud-based intelligent analysis of medical detection and image diagnosis system according to claim 2, wherein, The time sequence adjustment submodule is specifically configured as follows: different initial time sequences are allocated to different detection and diagnosis tasks to obtain the time sequences corresponding to the detection and diagnosis tasks; the time sequences corresponding to the detection and diagnosis tasks are reduced by a unit value every clock cycle, and all the detection and diagnosis tasks are sent into the window acquisition submodule in order of increasing time sequences.

4. The cloud-based intelligent analysis enabled medical detection and image diagnosis system as claimed in claim 2, wherein, The numerical value distribution mapping submodule is specifically configured as follows: continuous numerical value intervals are allocated to servers, and the numerical value intervals corresponding to all the servers form a total mapping interval; a mapping value corresponding to each target detection and diagnosis task acquired by the window acquisition submodule is randomly generated in the mapping interval, and all the target detection and diagnosis tasks are coded into vectors to obtain task distribution codes; a plurality of different task distribution codes are repeatedly obtained, and the task distribution codes are optimized by using a spiral surrounding exploration method, a step fraction exploration method, an information entropy acquisition exploration method and a fuzzy fluctuation exploration method in sequence for multiple times to obtain a target task distribution code. The target task allocation code is decoded to determine the server corresponding to each target detection and diagnosis task, and the task allocation is performed according to the server corresponding to the target detection and diagnosis task.

5. The cloud-based intelligent analysis enabled medical detection and image diagnosis system of claim 4, wherein, The task allocation code is sequentially optimized by using spiral surrounding exploration, fractional order exploration, information entropy collection exploration and fuzzy fluctuation exploration methods to obtain a target task allocation code, including: The task allocation code is first-order optimized by using the spiral surrounding exploration method to obtain the task allocation code after first-order optimization; The task allocation code after first-order optimization is second-order optimized by using the fractional order exploration method to obtain the task allocation code after second-order optimization; The task allocation code after second-order optimization is third-order optimized by using the information entropy collection exploration method to obtain the task allocation code after third-order optimization; The task allocation code after third-order optimization is fourth-order optimized by using the fuzzy fluctuation exploration method to obtain the task allocation code after fourth-order optimization; The first-order optimization to the fourth-order optimization is repeatedly executed until the optimization end condition is met, and the task allocation code after the fourth-order optimization in the last optimization process is used to determine the target task allocation code. 6.The cloud-based intelligent analysis of medical detection and image diagnosis system according to claim 5, wherein, The task allocation code is first-order optimized by using the spiral surrounding exploration method to obtain the task allocation code after first-order optimization as follows: wherein is the task assignment code for the i th task, is the task assignment code for the i th first-order optimization, is the optimal task assignment code, is a natural constant, t is the number of optimizations, b is a spiral wrap factor, and is set to 1 or 1.5; is a first random number between (0, 1), is a constant, i = 1, 2, …, N, N represents the total number of task assignment codes.

7. The cloud-based intelligent analysis enabled medical detection and image diagnosis system of claim 6, wherein, The task allocation code after first-order optimization is second-order optimized by using the fractional order exploration method to obtain the task allocation code after second-order optimization as follows: wherein, is the task assignment encoding after the first k -order optimization, is the task assignment encoding after the second k -order optimization, is the optimization speed in the first t +1-order optimization process, is the optimization speed in the first t -order optimization process, is the optimization speed in the first t -1-order optimization process, is the optimization speed in the first t -2-order optimization process, is the optimization speed in the first t -2-order optimization process, k = 1, 2, …, N, is the inertia weight, and is set to 0.85 or 0.95; is the fractional order factor, is a second random number between (0, 1); is the task assignment encoding after the first -order optimization except for is the Euclidean distance between and is the interaction control coefficient, and is set to 2 or 3; is the average Euclidean distance between the optimal task assignment encoding and all the task assignment encodings after the first-order optimization, is the minimum value in the average Euclidean distances corresponding to all the task assignment encodings after the first-order optimization, is the maximum value in the average Euclidean distances corresponding to all the task assignment encodings after the first-order optimization. 8.The cloud-based intelligent analysis of medical detection and image diagnosis system according to claim 6, wherein, The task allocation code after second-order optimization is third-order optimized by using the information entropy collection exploration method to obtain the task allocation code after third-order optimization as follows: ； ； wherein, is the first task allocation code after the second-order optimization, j is the task allocation code, is the second task allocation code after the second-order optimization, j is the third task allocation code after the second-order optimization, is the optimal task allocation code, is the first learning factor, and is set to 1 or 2; is the second learning factor, and is set to 0.65 or 1; is a third random number between (0, 1), is a fourth random number between (0, 1), is the first task allocation code after the second-order optimization, j is the -1th task allocation code, j = 1, 2, …, N, and j is 1, is set to a random second-order optimized task allocation code, is a step factor, is a minimum value of the step factor, and is set to 0.01; is a preset maximum number of optimizations, is a balance factor, and is set to 0.4 or 0.5; is a minimum value of the fitness corresponding to the second-order optimized task allocation code, is a maximum value of the fitness corresponding to the second-order optimized task allocation code, is an information collection factor, is an information entropy, is a corresponding intermediate parameter, is a corresponding fitness. 9.The cloud-based intelligent analysis of medical detection and image diagnosis system according to claim 6, wherein, The task allocation code after third-order optimization is fourth-order optimized by using the fuzzy fluctuation exploration method to obtain the task allocation code after fourth-order optimization as follows: wherein, is the task assignment encoding after the first h third-order optimization, is the task assignment encoding after the first h fourth-order optimization, is the reverse learning encoding within the solution space, is the fuzzy volatility factor, is the shrinkage expansion coefficient, and linearly decreases from 1 to 0 as the number of training increases; is the average historical optimal value corresponding to all the task assignment encodings after the third-order optimization, is the random volatility behavior control factor between (0, 1), is the maximum value of the fuzzy volatility factor, and is set to 0.85; is the minimum value of the fuzzy volatility factor, and is set to 0.01; is the constant pi, T represents the preset maximum number of optimizations.

10. The cloud-based intelligent analysis enabled medical detection and image diagnosis system as claimed in claim 2, wherein, The pre-deployed image segmentation model is set as U-Net or U-Net++.

Images