Superselective ocular artery thrombolysis treatment comprehensive data processing and analysis system

Abstract

The invention relates to the technical field of medical data processing, in particular to a super-selective ocular artery thrombolysis treatment comprehensive data processing and analyzing system. The system comprises a thrombus positioning module for constructing a thrombus segmentation model to identify a thrombus position; the medication position optimization module judges whether the patient is suitable for adopting super-selective ocular artery thrombolysis treatment or not according to the thrombus position, the eyeball structure data, the orbit structure data and the artery structure data, and obtains recommendation data of the medication position through a medication position optimization algorithm; a first recommendation unit in the medication dosage recommendation module processes historical thrombus data of all patients through a machine learning algorithm to obtain a first recommended medication dosage, and a second recommendation unit obtains a second recommended medication dosage according to personal data of the patients; the comprehensive recommendation unit obtains recommended medication doses according to the data and the medication doses recommended by the experts; the curative effect analysis module is used for collecting treatment data and constructing a curative effect score according to the treatment data to assist doctors in making decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and particularly to a comprehensive data processing and analysis system for ultra-selective ophthalmic artery thrombolysis treatment. Background Art

[0002] With the in-depth integration of medical technology and data science, the processing and analysis of medical data have become one of the important technical directions in the modern medical field. Especially in the field of ophthalmology, the diagnosis and treatment of diseases such as ophthalmic artery occlusion have put forward higher requirements for accurate data analysis and real-time processing. However, the existing technologies have the following deficiencies: On the one hand, existing medical data analysis systems usually only analyze a single data source such as imaging data or biochemical indicators, and it is difficult to effectively integrate multi-modal data, such as the medical history records of patients, real-time images, and treatment equipment parameters.

[0003] On the other hand, in the treatment of acute diseases, such as ophthalmic artery occlusion thrombolysis treatment, the real-time and high accuracy of data are crucial. However, the current systems mostly rely on manual analysis, with low processing efficiency and prone to errors, unable to meet the clinical emergency needs. Although some medical analysis tools based on machine learning have been applied to specific scenarios to provide auxiliary decision-making, these systems lack sufficient algorithm training data and are difficult to achieve comprehensive diagnosis and prediction support for complex cases.

[0004] In view of the above problems, the present invention combines medical data processing technology and artificial intelligence analysis technology to propose a comprehensive data processing and analysis system for ultra-selective ophthalmic artery thrombolysis treatment, aiming to achieve data collection, comprehensive analysis, and intelligent auxiliary decision-making, provide efficient technical support for doctors, and improve the work efficiency of doctors. Summary of the Invention

[0005] The purpose of the present invention is to provide a comprehensive data processing and analysis system for ultra-selective ophthalmic artery thrombolysis treatment to provide auxiliary decision-making for ophthalmic artery thrombolysis treatment and improve the work efficiency of doctors. First, the thrombus localization module constructs a thrombus segmentation model to identify the thrombus location; the drug administration position optimization module determines whether the patient is suitable for ultra-selective ophthalmic artery thrombolysis treatment according to the thrombus location, eyeball structure data, orbital structure data, and arterial structure data, and obtains recommended data for the drug administration position through a drug administration position optimization algorithm; the first recommendation unit in the drug dosage recommendation module processes the historical thrombus data of all patients through a machine learning algorithm to obtain the first recommended drug dosage, the second recommendation unit obtains the second recommended drug dosage according to the patient's personal data, and the comprehensive recommendation unit obtains the recommended drug dosage according to the above data; the efficacy analysis module is used to collect treatment data and construct an efficacy score based on the treatment data to assist doctors in making decisions.

[0006] To achieve the above object, the present invention provides the following technical solutions: A comprehensive data processing and analysis system for super-selective ophthalmic artery thrombolysis treatment, comprising: The thrombus localization module includes an image preprocessing unit and a thrombus recognition unit, wherein the thrombus recognition unit includes a thrombus segmentation model for identifying the thrombus position; the thrombus segmentation model includes a thrombus feature extraction layer, a thrombus multi-modal fusion layer, a thrombus segmentation layer and a thrombus output layer; Further, the image preprocessing unit includes: Obtain thrombus multi-modal images with the same resolution, where the thrombus multi-modal images include DSA images, OCTA images and CDU images; Perform data alignment on the thrombus multi-modal images using a timestamp-based synchronization technique to obtain aligned thrombus multi-modal images; perform spatial alignment on the aligned thrombus multi-modal images using an image registration algorithm to obtain registered thrombus multi-modal images; perform data enhancement on the registered thrombus multi-modal images to obtain enhanced thrombus multi-modal images.

[0007] Further, the thrombus segmentation model further includes: The thrombus feature extraction layer includes a first branch layer, a second branch layer and a third branch layer, where the first branch layer includes A convolutional layer, a BatchNorm layer and a ReLU layer for processing DSA images to obtain a large blood vessel structure feature map; the second branch layer includes a depth convolutional layer and a dilated convolutional layer for processing OCTA images to obtain a fine blood vessel structure feature map; the third branch layer includes a 1D convolutional layer for processing CDU images to obtain a blood flow feature map; The thrombus feature fusion layer includes a pixel-level fusion layer, a feature-level fusion layer and an output layer, where the pixel-level fusion layer combines the large blood vessel structure feature map, the fine blood vessel structure feature map and the blood flow feature map through linear weighted averaging to obtain a first comprehensive feature; the feature-level fusion layer uses a Transformer module to process the global association between different feature maps to obtain a second comprehensive feature; the output layer obtains and processes the first comprehensive feature and the second comprehensive feature to obtain a multi-modal feature map for thrombus area segmentation; The thrombus segmentation layer includes an encoder layer, a skip connection layer, a decoder layer and a dynamic segmentation head layer, where the encoder layer includes 4 convolutional blocks, and each convolutional block includes a convolutional layer, a BatchNorm layer, a ReLU layer and a max pooling operation layer; the skip connection layer uses an attention U-Net structure for screening the information of the skip connection; the decoder layer includes an upsampling layer and a multi-scale fusion layer; the dynamic segmentation head layer combines dynamic convolution to adjust the weights of different modal features and outputs a thrombus segmentation mask; The thrombus output layer outputs a binary image of the thrombus segmentation and marks the thrombus area.

[0008] The medication position optimization module is used to determine whether a patient is suitable for super-selective ophthalmic artery thrombolysis treatment based on the thrombus position, eye structure data, orbital structure data, and arterial structure data. If suitable, it analyzes through the medication position optimization algorithm to obtain the first medication position data and the second medication position data. The medication position data includes the insertion point coordinates, release point coordinates, insertion angle, and insertion depth. Furthermore, the medication position optimization module further includes: The formula for calculating the super-selective ophthalmic artery thrombolysis applicability score is: ; Wherein, represents the super-selective ophthalmic artery thrombolysis applicability score, represents the thrombus position depth weight coefficient of represents the complexity of the eye structure weight coefficient of represents the complexity of the orbital structure weight coefficient of represents the complexity of the arterial structure weight coefficient of; If the super-selective ophthalmic artery thrombolysis applicability score is greater than the preset applicability threshold, super-selective ophthalmic artery thrombolysis treatment is not suitable; If the super-selective ophthalmic artery thrombolysis applicability score is not greater than the preset applicability threshold, it enters the medication position optimization algorithm, including: obtaining an obstacle avoidance node set based on the eye structure data, orbital structure data, and arterial structure data, constructing an obstacle avoidance area according to the maximum diameter of the obstacle avoidance nodes to obtain an obstacle avoidance area set, and removing the obstacle avoidance area set from the three-dimensional space model through Boolean operations to obtain a feasible path space; Using the octree algorithm to divide the feasible path space and adopting the Dijkstra algorithm to obtain the first medication position data; Removing the insertion point coordinates of the first medication position data and recalculating to obtain the second medication position data.

[0009] The medication dose recommendation module includes a first recommendation unit, a second recommendation unit, and a comprehensive recommendation unit. Among them, the first recommendation unit processes the patient's historical thrombus data through a machine learning algorithm to obtain the first recommended medication dose; the second recommendation unit obtains the second recommended medication dose according to the patient's personal physical data and historical medication data; the comprehensive recommendation unit obtains the recommended medication dose based on the first recommended medication dose, the second recommended medication dose, and the expert recommended medication dose; Furthermore, the first recommendation unit further includes: Construct a thrombus data set based on historical patient data, where the thrombus data set includes blood vessel diameter, blood flow rate, thrombus area, thrombus location, and historical patient medication records; Denoise and normalize the thrombus data set to obtain a standard thrombus data set, and use a machine learning regression model for training, with the output being the first recommended medication dose ; Obtain the individual data of the patient and process it using the trained machine learning regression model to obtain the first recommended medication dose for the patient. Determine whether it is within the safe range based on the historical dose data. If it is greater than the maximum medication dose, the value of the first recommended medication dose is the maximum medication dose.

[0010] Further, the second recommendation unit further includes: Obtain the historical data of the patient and process it to obtain the individual data of the patient. Calculate the second recommended medication dose based on the individual data of the patient. The formula is: ; Wherein, represents the second recommended medication dose, represents the first recommended medication dose, represents the patient's weight, represents the standard weight, represents the age adjustment coefficient, represents the medical history adjustment coefficient.

[0011] Further, the recommended medication dose further includes: Obtain the individual data of the patient and obtain the first recommended medication dose for the patient through the first recommendation unit, and obtain the second recommended medication dose through the second recommendation unit. Obtain the recommended medication dose based on the first recommended medication dose, the second recommended medication dose, and the expert-recommended medication dose. The formula is: ; Wherein, represents the recommended medication dose, represents the weight coefficient of the first recommended medication dose , represents the weight coefficient of the second recommended medication dose, represents the patient's weight, represents the standard weight, represents the age adjustment coefficient, represents the medical history adjustment coefficient, represents the expert-recommended medication dose 's weight coefficient.

[0012] The efficacy analysis module is used to collect treatment data, including blood flow recovery speed, vascular patency, imaging comparison data, and patient symptoms, and generate reports; construct an efficacy score based on the treatment data for assisting in decision-making.

[0013] Furthermore, the calculation formula for the efficacy score in the efficacy analysis module is as follows: ; Wherein, represents the efficacy score, represents the weight coefficient of the imaging index score of, represents the blood flow recovery index of the weight coefficient, represents the symptom improvement score of the weight coefficient, represents the complication score of the weight coefficient.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By integrating the advantages of different modal images, it ensures the comprehensive and accurate identification of thrombus; the combination of pixel-level fusion and feature-level fusion greatly improves the comprehensiveness and accuracy of thrombus feature expression. The binary image and superimposed image of thrombus segmentation enable doctors to intuitively understand the segmentation results, enhancing the clinical usability of the model; this model can adapt to a variety of imaging modalities and diverse thrombus features, and at the same time combines a dynamic adjustment mechanism and multi-scale feature extraction ability, having good adaptability to different patients and complex thrombus conditions, providing scientific support for subsequent treatment decisions, and significantly improving the success rate and safety of ophthalmic artery thrombolysis treatment.

[0015] 2. Through the applicability score of super-selective ophthalmic artery thrombolysis, it converts complex anatomical structures and patient individual differences into quantifiable indicators, helping to objectively and accurately evaluate whether a patient is suitable for super-selective ophthalmic artery thrombolysis treatment; combining the structural data of the eyeball, orbit, and artery, using obstacle avoidance nodes and obstacle avoidance areas to construct an accurate three-dimensional space model to ensure that the catheter path does not damage key tissues; by calculating the first drug administration position data and the second drug administration position data, it increases the robustness and fault tolerance of the treatment plan, further ensuring the accuracy and safety of drug release; the visualized results of key parameters such as the insertion point and release point generated provide intuitive references for doctors, improving the clinical decision-making efficiency and reducing the dependence on surgical experience.

[0016] 3. The first recommendation unit uses a machine learning regression model to train the standardized thrombus dataset to ensure that the recommended dose is more in line with the patient's specific physiological conditions and disease conditions. The second recommendation unit dynamically adjusts the recommended dose according to the patient's individual characteristics by analyzing the patient's physical data and medication data to ensure the personalization of the treatment plan. By introducing the recommended medication dose by experts and combining manual experience and model prediction results, the recommended dose is further optimized to make it more in line with the actual clinical needs, and an accurate and safe thrombolytic drug dose is output. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 FIG. is a schematic structural diagram of a comprehensive data processing and analysis system for super-selective ophthalmic artery thrombolysis treatment provided by an embodiment of the present invention; Figure 2 FIG. 1 is an angiographic image provided by an embodiment of the present invention; Figure 3 FIG. 2 is an angiographic image provided by an embodiment of the present invention; Figure 4 FIG. is a schematic structural diagram of a thrombus segmentation model provided by an embodiment of the present invention; Figure 5 FIG. is a flowchart of a medication position optimization algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] The specific content of Embodiment 1 is as follows:

[0020] In order to make full use of the existing medical data, shorten the treatment time of ophthalmic artery thrombosis, and improve the work efficiency of doctors, a certain hospital introduced a comprehensive data processing and analysis system for super-selective ophthalmic artery thrombolysis treatment provided by the present invention to assist doctors in making decisions. The specific implementation method is as follows: The system structure is as Figure 1 shown, mainly including four modules. Among them, the thrombus positioning module includes an image preprocessing unit and a thrombus recognition unit. The thrombus recognition unit includes a thrombus segmentation model for identifying the thrombus position. The thrombus segmentation model includes a thrombus feature extraction layer, a thrombus multi-modal fusion layer, a thrombus segmentation layer, and a thrombus output layer; Further, the image preprocessing unit includes: Obtain thrombus multimodal images with the same resolution, where the thrombus multimodal images include DSA images, OCTA images, and CDU images, and the specific images are as Figure 2 and Figure 3 shown; Perform data alignment on the thrombus multimodal images using timestamp-based synchronization technology to obtain aligned thrombus multimodal images; perform spatial alignment on the aligned thrombus multimodal images using an image registration algorithm to obtain registered thrombus multimodal images; perform data enhancement on the registered thrombus multimodal images to obtain enhanced thrombus multimodal images.

[0021] By performing multimodal data alignment, spatial registration, and data enhancement on thrombus multimodal images, the quality and consistency of the images are comprehensively improved, providing higher-quality input data for subsequent fusion and analysis, and providing accurate and reliable data support for thrombus segmentation.

[0022] Furthermore, the thrombus segmentation model is as Figure 4 shown, and further includes: The thrombus feature extraction layer includes a first branch layer, a second branch layer, and a third branch layer. The first branch layer includes a convolutional layer, a BatchNorm layer, and a ReLU layer, which are used to process DSA images to obtain a large blood vessel structure feature map; the second branch layer includes a depth convolutional layer and an atrous convolutional layer, which are used to process OCTA images to obtain a fine blood vessel structure feature map; the third branch layer includes a 1D convolutional layer, which is used to process CDU images to obtain a blood flow feature map; The thrombus feature fusion layer includes a pixel-level fusion layer, a feature-level fusion layer, and an output layer. The pixel-level fusion layer combines the large blood vessel structure feature map, the fine blood vessel structure feature map, and the blood flow feature map through linear weighted averaging to obtain a first comprehensive feature; the feature-level fusion layer uses a Transformer module to capture the global correlation between different feature maps to obtain a second comprehensive feature; the output layer obtains and processes the first comprehensive feature and the second comprehensive feature to obtain a multimodal feature map for thrombus region segmentation; The thrombus segmentation layer includes an encoder layer, a skip connection layer, a decoder layer, and a dynamic segmentation head layer. The encoder layer includes 4 convolutional blocks, and each convolutional block contains a convolutional layer, a BatchNorm layer, a ReLU layer, and a max pooling operation layer; an SE or CBAM mechanism is added to the encoding layer to enhance the thrombus region features; the skip connection layer uses an attention U-Net structure to filter the information of the skip connection based on feature importance; the decoder layer includes an upsampling layer and a multi-scale fusion layer. In the upsampling layer, the upsampling module restores the resolution and uses bilinear interpolation combined with a convolutional operation. The multi-scale fusion layer extracts features of different scales; the dynamic segmentation head layer combines dynamic convolution to adjust the weights of different modal features and outputs a thrombus segmentation mask, including the position and range of the thrombus; The thrombus output layer outputs the binary image of thrombus segmentation, marks the thrombus area, further generates a visualization graph of the segmentation result, and superimposes it on the original image.

[0023] As shown in Table 1, the training results of the thrombus segmentation model show that it converges to a very good effect.

[0024]

[0025] Through innovative designs in multiple aspects such as multi-modal feature extraction and fusion, attention mechanism optimization, multi-scale information fusion, dynamic segmentation head weight adjustment, and visualization output, the thrombus segmentation model significantly improves the accuracy, robustness, and adaptability of thrombus location segmentation, providing efficient and reliable data support for ophthalmic artery thrombolysis treatment.

[0026] The medication position optimization module is used to determine whether the patient is suitable for super-selective ophthalmic artery thrombolysis treatment based on thrombus location, eyeball structure data, orbital structure data, and arterial structure data. If suitable, it analyzes through the medication position optimization algorithm to obtain the first medication position data and the second medication position data. The medication position data includes insertion point coordinates, release point coordinates, insertion angle, and insertion depth. Furthermore, the medication position optimization module further includes: The formula for calculating the super-selective ophthalmic artery thrombolysis applicability score is: ; Where, represents the super-selective ophthalmic artery thrombolysis applicability score, represents the thrombus position depth is the weight coefficient of represents the eyeball structure complexity is the weight coefficient of represents the orbital structure complexity is the weight coefficient of represents the arterial structure complexity is the weight coefficient of; If the super-selective ophthalmic artery thrombolysis applicability score is greater than the preset applicability threshold, super-selective ophthalmic artery thrombolysis treatment is not suitable; If the super-selective ophthalmic artery thrombolysis applicability score is not greater than the preset applicability threshold, enter the medication position optimization algorithm, and the algorithm structure is as Figure 5 shown; Obtain the obstacle avoidance node set according to the eyeball structure data, orbital structure data, and arterial structure data, construct the obstacle avoidance area based on the maximum diameter of the obstacle avoidance nodes to obtain the obstacle avoidance area set, and remove the obstacle avoidance area set from the three-dimensional space model through Boolean operations to obtain the feasible path space; The feasible path space is segmented using the octree algorithm, and the optimal path search logic of the Dijkstra algorithm is adopted to obtain the first drug administration position data; The insertion point coordinates of the first drug administration position data are removed and recalculated to obtain the second drug administration position data.

[0027] Furthermore, the thrombus position depth refers to the depth of the thrombus in the target artery; Furthermore, the eyeball structure complexity refers to establishing a surface function to fit the shape of the eyeball, then extracting the structural features of the surface function, such as radius, aspect ratio, and surface area, etc., and then calculating the eyeball structure complexity by weighted calculation; Furthermore, the orbital structure complexity refers to evaluating the degree of restriction on the treatment path and instrument operation based on the orbital bone and spatial structure. By extracting features such as orbital volume, transverse and longitudinal diameters of the orbital opening area, the number and distribution of bony protrusions in the orbit that may interfere with the treatment path, etc., and then calculating the orbital structure complexity by weighted calculation; Furthermore, the artery structure complexity is used to evaluate the spatial complexity and anatomical variation of the target artery. By extracting features such as the number of the arterial trunk and its branches, the total curvature and change rate of the arterial path, the standard deviation of the arterial diameter change along the path, etc., and then calculating the artery structure complexity by weighted calculation.

[0028] The drug administration position optimization module first screens suitable patients through the super-selective ophthalmic artery thrombolysis applicability score to avoid the risks and side effects that may be caused by inappropriate treatment plans; then constructs a three-dimensional obstacle avoidance area using the eyeball, orbital, and artery structure data, which can accurately avoid important tissues and risk areas, and minimize the probability of accidental injury; then combines multiple algorithms to improve the path planning efficiency and accuracy, and finally outputs the first position data and the second position data for doctors to choose, providing efficient and intuitive operation guidance for doctors.

[0029] The drug dosage recommendation module includes a first recommendation unit, a second recommendation unit, and a comprehensive recommendation unit. The first recommendation unit processes the historical thrombus data of the patient through a machine learning algorithm to obtain the first recommended drug dosage; the second recommendation unit obtains the second recommended drug dosage according to the patient's personal physical data and historical drug use data; the comprehensive recommendation unit obtains the recommended drug dosage according to the first recommended drug dosage, the second recommended drug dosage, and the expert recommended drug dosage.

[0030] Furthermore, the first recommendation unit further includes: Construct a thrombus data set according to the historical patient data, and the thrombus data set includes blood vessel diameter, blood flow, thrombus area, thrombus position, and historical patient drug use records; Denoise and normalize the thrombus dataset to obtain a standard thrombus dataset, and use a machine learning regression model for training, with the output being the first recommended drug dosage. ; Obtain the individual data of the patient and process it using the trained machine learning regression model to obtain the patient's first recommended drug dosage. Determine whether it is within the safe range based on the historical dosage data. If it is greater than the maximum drug dosage, the value of the first recommended drug dosage is the maximum drug dosage.

[0031] As shown in Table 2 are the first recommended drug dosages of some patients in the test set.

[0032]

[0033] By processing key factors such as blood vessel diameter, blood flow, thrombus area, location, and medication records, and comprehensively analyzing and considering the influence of different variables on the drug dosage using machine learning algorithms, potential medication patterns and dosage rules can be mined, thereby providing a reliable reference for the basic dosage for doctors.

[0034] Furthermore, the second recommendation unit further includes: Obtain the historical data of the patient and process it to obtain the patient's individual data, and calculate the second recommended drug dosage based on the patient's individual data. The formula is: ; Where, represents the second recommended drug dosage, represents the first recommended drug dosage, represents the patient's weight, represents the standard weight, represents the age adjustment coefficient, represents the medical history adjustment coefficient.

[0035] Furthermore, the age adjustment coefficient refers to that the calculation of the coefficient combines the baseline adjustment value based on the empirical value of clinical data and the change value according to the individual's metabolic ability , and the calculation formula of the age adjustment coefficient is: ; Where, represents the age adjustment coefficient, represents the exponential function, represents the actual age, represents the optimal metabolic age, represents the standard deviation of the metabolic ability.

[0036] Further, the medical history adjustment coefficient is used to reflect the impact of the patient's past medical history on the current medication dose. By collecting the patient's medical history and assigning risk weights to each type of medical history, the medical history adjustment coefficient is obtained through weighted calculation.

[0037] The second recommendation unit calculates the exclusive medication dose by combining the patient's individual data, provides dose recommendations tailored to the individual characteristics based on the patient's personal physical data, fully embodies the concept of personalized medicine, ensures that the recommended dose conforms to the patient's individual physiological and pathological characteristics, enhances medication safety, and avoids dosing errors.

[0038] Further, the recommended medication dose also includes: Obtain the patient's individual data and get the patient's first recommended medication dose through the first recommendation unit, get the second recommended medication dose through the second recommendation unit, and obtain the recommended medication dose based on the first recommended medication dose, the second recommended medication dose, and the expert-recommended medication dose. The formula is: ; Wherein, represents the recommended medication dose, represents the weight coefficient of the first recommended medication dose of, represents the weight coefficient of the second recommended medication dose, represents the patient's weight, represents the standard weight, represents the age adjustment coefficient, represents the medical history adjustment coefficient, represents the expert-recommended medication dose of the weight coefficient.

[0039] The comprehensive recommendation unit obtains the final recommended medication dose by integrating the first recommended medication dose, the second recommended medication dose, and the expert-recommended medication dose, ensuring the comprehensiveness, scientificity, and flexibility of the recommended dose, taking into account the balance between data-driven and clinical experience, and improving the accuracy and individual adaptability of the recommended dose.

[0040] The efficacy analysis module is used to collect treatment data, including blood flow recovery speed, blood vessel patency, imaging contrast data, and patient symptoms, and generate a report; construct an efficacy score based on the treatment data for assisting in decision-making.

[0041] Further, the calculation formula for the efficacy score in the efficacy analysis module is: ; Wherein, represents the efficacy score, represents the score of the imaging index of the weight coefficient, Indicates the blood flow recovery index The weight coefficient of Indicates the symptom improvement score The weight coefficient of Indicates the complication score The weight coefficient of

[0042] Furthermore, the imaging index score is a comprehensive evaluation based on the changes in medical images after treatment. It is obtained by collecting imaging data before and after treatment, extracting features such as the size of the lesion area, vascular patency, and improvement of vascular morphology, and then performing weighted calculation after normalization.

[0043] Furthermore, the blood flow recovery index is an index to measure the improvement degree of hemodynamics after treatment, reflecting the speed, pressure, and stability of blood flow, and is obtained by weighted calculation of the above features.

[0044] Furthermore, the symptom improvement score is a comprehensive assessment of the patient's subjective symptoms and signs, used to quantify the relief effect of the treatment on the patient's actual pain, and is obtained by collecting and processing the patient's symptom data before and after treatment, and then performing weighted calculation; Furthermore, the complication score reflects the impact of adverse events during or after treatment, and is obtained by counting the complications during and after surgery, classifying the risk levels, and performing weighted calculation.

[0045] The efficacy analysis module collects the key data of the patient's treatment, automatically generates a treatment report, provides intuitive efficacy feedback for the patient, saves the time of medical staff, and improves work efficiency; constructs an efficacy score through a multi-factor scoring system to comprehensively and objectively evaluate the treatment effect, which helps doctors make quick decisions and optimize treatment plans.

[0046] The thrombus positioning module realizes the precise positioning of the thrombus position, provides data support for subsequent treatment decisions, the medication position optimization module provides an intuitive reference for doctors by generating the first medication position data and the second medication position data, and improves the decision-making efficiency; the medication dose recommendation module combines artificial experience and prediction results to optimize the recommended dose, and the efficacy analysis module helps doctors make quick decisions and optimize treatment plans through data collection and analysis. Through a comprehensive data processing and analysis system for super-selective ophthalmic artery thrombolysis therapy provided by the present invention, not only the accuracy and safety of super-selective ophthalmic artery thrombolysis therapy are improved, but also comprehensive, accurate, and real-time auxiliary guidance is provided for doctors, improving the work efficiency of doctors.

[0047] The specific content of Example 2 is as follows: A certain hospital introduced a comprehensive data processing and analysis system for super-selective ophthalmic artery thrombolysis therapy provided by the present invention to provide auxiliary decision-making for doctors. The specific implementation method is as follows: The thrombus localization module includes an image preprocessing unit and a thrombus recognition unit, wherein the thrombus recognition unit includes a thrombus segmentation model for identifying the thrombus location; the thrombus segmentation model includes a thrombus feature extraction layer, a thrombus multi-modal fusion layer, a thrombus segmentation layer, and a thrombus output layer; Further, the image preprocessing unit includes: Obtain thrombus multi-modal images with the same resolution, where the thrombus multi-modal images include DSA images, OCTA images, and CDU images; Perform data alignment on the thrombus multi-modal images using a timestamp-based synchronization technique to obtain aligned thrombus multi-modal images; perform spatial alignment on the aligned thrombus multi-modal images using an image registration algorithm to obtain registered thrombus multi-modal images; perform data augmentation on the registered thrombus multi-modal images to obtain enhanced thrombus multi-modal images.

[0048] Further, the thrombus segmentation model further includes: The thrombus feature extraction layer includes a first branch layer, a second branch layer, and a third branch layer, where the first branch layer includes a convolutional layer, a BatchNorm layer, and a ReLU layer for processing DSA images to obtain a large blood vessel structure feature map; the second branch layer includes a depth convolutional layer and a dilated convolutional layer for processing OCTA images to obtain a fine blood vessel structure feature map; the third branch layer includes a 1D convolutional layer for processing CDU images to obtain a blood flow feature map; The thrombus feature fusion layer includes a pixel-level fusion layer, a feature-level fusion layer, and an output layer. The pixel-level fusion layer combines the large blood vessel structure feature map, the fine blood vessel structure feature map, and the blood flow feature map through linear weighted averaging to obtain a first comprehensive feature; the feature-level fusion layer uses a Transformer module to process the global correlation between different feature maps to obtain a second comprehensive feature; the output layer obtains and processes the first comprehensive feature and the second comprehensive feature to obtain a multi-modal feature map for thrombus region segmentation; The thrombus segmentation layer includes an encoder layer, a skip connection layer, a decoder layer, and a dynamic segmentation head layer. The encoder layer includes 4 convolutional blocks, and each convolutional block contains a convolutional layer, a BatchNorm layer, a ReLU layer, and a max pooling operation layer; the skip connection layer uses an attention U-Net structure to filter the information of the skip connection; the decoder layer includes an upsampling layer and a multi-scale fusion layer; the dynamic segmentation head layer combines dynamic convolution to adjust the weights of different modal features and outputs a thrombus segmentation mask; The thrombus output layer outputs a binary image of the thrombus segmentation and marks the thrombus region.

[0049] The medication location optimization module is used to determine whether a patient is suitable for ultra - selective ophthalmic artery thrombolysis based on the thrombus location, eye structure data, orbital structure data, and arterial structure data. If suitable, it analyzes through the medication location optimization algorithm to obtain the first medication location data and the second medication location data. The medication location data includes the insertion point coordinates, release point coordinates, insertion angle, and insertion depth. Furthermore, the medication location optimization module further includes: The formula for calculating the applicability score of ultra - selective ophthalmic artery thrombolysis is: ; Wherein, represents the applicability score of ultra - selective ophthalmic artery thrombolysis, represents the thrombus location depth the weight coefficient of represents the complexity of the eye structure the weight coefficient of represents the complexity of the orbital structure the weight coefficient of represents the complexity of the arterial structure the weight coefficient of; If the applicability score of the ultra - selective ophthalmic artery thrombolysis is greater than the preset applicability threshold, ultra - selective ophthalmic artery thrombolysis treatment is not suitable; If the applicability score of the ultra - selective ophthalmic artery thrombolysis is not greater than the preset applicability threshold, it enters the medication location optimization algorithm, including: obtaining an obstacle - avoidance node set based on the eye structure data, orbital structure data, and arterial structure data, constructing an obstacle - avoidance area according to the maximum diameter of the obstacle - avoidance nodes to obtain an obstacle - avoidance area set, and removing the obstacle - avoidance area set from the three - dimensional space model through Boolean operations to obtain a feasible path space; Using the octree algorithm to segment the feasible path space and adopting the Dijkstra algorithm to obtain the first medication location data; Removing the insertion point coordinates of the first medication location data and recalculating to obtain the second medication location data.

[0050] The medication dose recommendation module includes a first recommendation unit, a second recommendation unit, and a comprehensive recommendation unit. The first recommendation unit processes the historical thrombus data of all patients through a machine learning algorithm to obtain the first recommended medication dose; the second recommendation unit obtains the second recommended medication dose according to the patient's personal physical data, thrombus data, and historical medication data; the comprehensive recommendation unit obtains the recommended medication dose based on the first recommended medication dose, the second recommended medication dose, and the expert - recommended medication dose.

[0051] Furthermore, the first recommendation unit further includes: Construct a thrombus dataset based on historical patient data, where the thrombus dataset includes blood vessel diameter, blood flow rate, thrombus area, thrombus location, and historical patient medication records; Denoise and normalize the thrombus dataset to obtain a standard thrombus dataset, and use a machine learning regression model for training, with the output being the first recommended medication dose ; Obtain the individual data of the patient and process it using the trained machine learning regression model to obtain the first recommended medication dose for the patient. Determine whether it is within the safe range based on the historical dose data. If it is greater than the maximum medication dose, the value of the first recommended medication dose is the maximum medication dose.

[0052] Furthermore, the second recommendation unit further includes: Obtain the historical data of the patient and process it to obtain the individual data of the patient. Calculate the second recommended medication dose based on the individual data of the patient. The formula is: Where, represents the second recommended medication dose, represents the first recommended medication dose, represents the patient's weight, represents the standard weight, represents the age adjustment coefficient, represents the medical history adjustment coefficient.

[0053] Furthermore, the recommended medication dose further includes: Obtain the individual data of the patient and obtain the first recommended medication dose for the patient through the first recommendation unit, and obtain the second recommended medication dose through the second recommendation unit. Based on the first recommended medication dose, the second recommended medication dose, and the expert-recommended medication dose, obtain the recommended medication dose. The formula is: ; Where, represents the recommended medication dose, represents the first recommended medication dose 's weight coefficient, represents the weight coefficient of the second recommended medication dose, represents the patient's weight, represents the standard weight, represents the age adjustment coefficient, represents the medical history adjustment coefficient, represents the weight coefficient of the expert-recommended medication dose as shown in Table 3 are the recommended medication doses for some patients.

[0054] The efficacy analysis module is used to collect treatment data, including blood flow recovery speed, vascular patency, imaging contrast data, and patient symptoms, and generate a report; construct an efficacy score based on the treatment data for assisting in decision-making.

[0055] Furthermore, the calculation formula for the efficacy score in the efficacy analysis module is: ; wherein, represents the efficacy score, represents the weight coefficient of the imaging index score ; represents the blood flow recovery index ; represents the symptom improvement score ; represents the complication score ;

[0056] Through a comprehensive data processing and analysis system for super-selective ophthalmic artery thrombolysis treatment provided by the present invention, data collection, processing, analysis, and feedback of the whole process of super-selective ophthalmic artery thrombolysis treatment are realized, providing comprehensive, accurate, and real-time decision-making assistance for doctors and improving doctors' work efficiency.

[0057] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A comprehensive data processing and analysis system for superselective ophthalmic artery thrombolysis therapy, characterized in that: include: The thrombus localization module includes an image preprocessing unit and a thrombus identification unit, wherein the thrombus identification unit includes a thrombus segmentation model for identifying the thrombus location; the thrombus segmentation model includes a thrombus feature extraction layer, a thrombus multimodal fusion layer, a thrombus segmentation layer and a thrombus output layer; The medication location optimization module is used to determine whether the patient is suitable for super-selective ophthalmic artery thrombolysis therapy according to the thrombus location, eyeball structure data, orbital structure data and arterial structure data. If suitable, the medication location optimization algorithm is used to analyze and obtain the first medication location data and the second medication location data. The medication location data includes the insertion point coordinates, the release point coordinates, the insertion angle and the insertion depth. The medication dosage recommendation module includes a first recommendation unit, a second recommendation unit and a comprehensive recommendation unit, wherein the first recommendation unit processes the patient's historical thrombosis data through a machine learning algorithm to obtain a first recommended medication dosage; The second recommendation unit obtains a second recommended medication dosage based on the patient's personal physical data and historical medication data; The comprehensive recommendation unit obtains a recommended medication dosage according to the first recommended medication dosage, the second recommended medication dosage and the expert recommended medication dosage; The efficacy analysis module is used to collect treatment data, including blood flow recovery rate, vascular patency, imaging comparison data and patient symptoms, generate reports, and construct efficacy scores based on treatment data to provide auxiliary decision-making.

2. A superselective ophthalmic artery thrombolysis treatment comprehensive data processing and analysis system according to claim 1, characterized in that: The image preprocessing unit comprises: Acquiring a thrombus multimodal image with the same resolution, wherein the thrombus multimodal image includes a DSA image, an OCTA image, and a CDU image; The thrombus multimodal image is subjected to data alignment based on a timestamp synchronization technology to obtain an aligned thrombus multimodal image; the aligned thrombus multimodal image is subjected to spatial alignment using an image registration algorithm to obtain a registered thrombus multimodal image; the registered thrombus multimodal image is subjected to data enhancement to obtain an enhanced thrombus multimodal image.

3. A superselective ophthalmic artery thrombolysis treatment comprehensive data processing and analysis system according to claim 1, characterized in that: The thrombus segmentation model further includes: The thrombus feature extraction layer includes a first branch layer, a second branch layer and a third branch layer, wherein the first branch layer includes The convolution layer, BatchNorm layer and ReLU layer are used to process DSA images to obtain large blood vessel structure feature maps; the second branch layer includes a deep convolution layer and a hole convolution layer, which are used to process OCTA images to obtain fine blood vessel structure feature maps; the third branch layer includes a 1D convolution layer, which is used to process CDU images to obtain blood flow feature maps; The thrombus feature fusion layer includes a pixel-level fusion layer, a feature-level fusion layer, and an output layer, wherein the pixel-level fusion layer combines the large vessel structure feature map, the fine vessel structure feature map, and the blood flow feature map through linear weighted average to obtain a first comprehensive feature; the feature-level fusion layer uses a Transformer module to process the global association between different feature maps to obtain a second comprehensive feature; the output layer obtains and processes the first comprehensive feature and the second comprehensive feature to obtain a multimodal feature map for thrombus region segmentation; The thrombus segmentation layer includes an encoder layer, a skip connection layer, a decoder layer, and a dynamic segmentation head layer. The encoder layer includes 4 convolution blocks, each of which contains a convolution layer, a BatchNorm layer, a ReLU layer, and a maximum pooling operation layer. The skip connection layer uses an attention U-Net structure to filter the information of the skip connection. The decoder layer includes an upsampling layer and a multi-scale fusion layer. The dynamic segmentation head layer combines dynamic convolution to adjust the weights of different modal features and outputs a thrombus segmentation mask. The thrombus output layer outputs a binary image of thrombus segmentation and marks the thrombus area.

4. A superselective ophthalmic artery thrombolysis treatment comprehensive data processing and analysis system according to claim 1, characterized in that: The medication location optimization module also includes: The formula for calculating the suitability score for superselective ophthalmic artery thrombolysis is: ； in, represents the suitability score of the superselective ophthalmic artery thrombolysis, Indicates the depth of thrombus location The weight coefficient of Indicates the complexity of the eyeball structure The weight coefficient of Indicates the complexity of orbital structure The weight coefficient of Indicates the complexity of arterial structure The weight coefficient of If the super-selective ophthalmic artery thrombolysis suitability score is greater than the preset suitability threshold, super-selective ophthalmic artery thrombolysis treatment is not suitable; If the super-selective ophthalmic artery thrombolysis suitability score is not greater than the preset suitability threshold, the medication location optimization algorithm is entered, including: obtaining an obstacle avoidance node set based on the eyeball structure data, the orbital structure data and the arterial structure data, constructing an obstacle avoidance area based on the maximum diameter of the obstacle avoidance node to obtain an obstacle avoidance area set, and removing the obstacle avoidance area set from the three-dimensional space model through Boolean operations to obtain a feasible path space; using an octree algorithm to segment the feasible path space, and using a Dijkstra algorithm to obtain the first medication location data; removing the insertion point coordinates of the first medication location data and recalculating to obtain the second medication location data.

5. The comprehensive data processing and analysis system for superselective ophthalmic artery thrombolysis therapy according to claim 1, characterized in that: The first recommendation unit further includes: constructing a thrombus data set based on historical patient data, wherein the thrombus data set includes vessel diameter, blood flow, thrombus area, thrombus location, and historical patient medication records; The thrombus dataset is denoised and normalized to obtain a standard thrombus dataset, which is then trained using a machine learning regression model and output as the first recommended medication dose. ; The patient's individual data is obtained and processed using a trained machine learning regression model to obtain the patient's first recommended dosage. Based on the historical dosage data, it is determined whether it is within the safety range. If it is greater than the maximum dosage, the first recommended dosage is taken as the maximum dosage.

6. A superselective ophthalmic artery thrombolysis treatment comprehensive data processing and analysis system according to claim 1, characterized in that: The second recommendation unit further includes: Obtain and process the patient's historical data to obtain the patient's individual data, and calculate the second recommended medication dose based on the patient's individual data. The formula is: ; in, Indicates the second recommended dosage. Indicates the first recommended dosage. represents the patient’s weight, Indicates standard weight, represents the age adjustment factor, represents the medical history adjustment factor.

7. The comprehensive data processing and analysis system for superselective ophthalmic artery thrombolysis therapy according to claim 1, characterized in that: The recommended dosage also includes: The individual data of the patient is obtained, and the first recommended medication dose of the patient is obtained through the first recommendation unit, and the second recommended medication dose is obtained through the second recommendation unit. The recommended medication dose is obtained according to the first recommended medication dose, the second recommended medication dose and the expert recommended medication dose, and the formula is: ； in, Indicates the recommended dosage, Indicates the first recommended dose The weight coefficient of Represents the weight coefficient of the second recommended dosage, represents the patient’s weight, Indicates standard weight, represents the age adjustment factor, represents the medical history adjustment factor, Indicates the dosage recommended by experts The weight coefficient of .

8. The comprehensive data processing and analysis system for superselective ophthalmic artery thrombolysis therapy according to claim 1, characterized in that: The calculation formula of the efficacy score in the efficacy analysis module is: ； in, represents the efficacy score, Indicates the image index score The weight coefficient of Blood flow restoration index The weight coefficient of Symptom improvement score The weight coefficient of Complication score The weight coefficient of .

Images