Stroke patient transfer auxiliary decision system based on multi-modal images

Abstract

The application relates to the technical field of transport auxiliary decision-making, and provides a stroke patient transport auxiliary decision-making system based on multi-modal images, which comprises an image data acquisition module, an image data processing module, an information analysis module, an auxiliary decision-making module and a user interface. The image data acquisition module is used for acquiring multi-modal image data of a patient and transmitting the multi-modal image data to the image data processing module; the image data processing module performs feature extraction and fusion on the received image data, generates fused multi-modal image feature information, and transmits the fused multi-modal image feature information to the information analysis module; the information analysis module analyzes the fused image feature information by using a deep learning model, generates image feature quantitative indexes and image feature index evaluation values, and transmits the image feature quantitative indexes and the image feature index evaluation values to the auxiliary decision-making module; and the auxiliary decision-making module generates a hospital allocation scheme based on the received quantitative indexes, scores, a hospital resource database and geographical position information, combines a genetic algorithm, and transmits the result to the user interface.

Description

Technical Field

[0001] This invention relates to the field of transport assistance decision-making technology, and more specifically, to a stroke patient transport assistance decision-making system based on multimodal imaging. Background Technology

[0002] Stroke is an acute disease that seriously threatens human health, characterized by rapid progression and life-threatening consequences. For stroke patients, time is of the essence; whether they receive timely and appropriate treatment within the first few hours of symptom onset directly impacts their prognosis. In the current stroke care system, patients are typically first transferred to the nearest medical facility for initial assessment and examination. However, due to the complexity and diversity of stroke conditions, further specialized evaluation and advanced treatment may be required after the initial diagnosis, necessitating inter-hospital transfer. Given the highly time-sensitive nature of stroke treatment, patient transfer decisions become crucial.

[0003] Existing stroke patient transport methods have several shortcomings. First, transport decisions often rely on the experience and judgment of medical staff or simply consider hospital distance, lacking standardized and objective evaluation criteria. This high degree of subjectivity easily leads to inconsistencies in decision-making. Second, the utilization of multimodal imaging data is severely insufficient; imaging data from different modalities, such as computed tomography (CT) and MRI, are often analyzed separately, lacking effective integration and failing to fully extract the comprehensive information within the multimodal imaging data. Furthermore, traditional transport decisions typically lack a comprehensive consideration of real-time hospital resources, ignoring current bed occupancy and the workload of medical staff, making it difficult to accurately match the specific needs of patients with the professional capabilities of medical institutions at all levels, resulting in inefficient allocation of regional medical resources.

[0004] With the development of medical imaging technology and artificial intelligence, stroke patient transport auxiliary decision-making systems based on multimodal images have become a research hotspot. Among existing published literature, Reference 1 (Multimodal medical image fusion algorithm in the era of big data. Neural Computing and Applications, 2020, 1-21) proposed an algorithm based on multimodal image fusion, using a pulse-coupled neural network and an energy attribute fusion strategy to perform image data fusion in the non-sampling Shearlet transform domain. Simulation studies showed that the algorithm exhibited high accuracy and stability in several medical diagnostic problems. Reference 2 (Abstract TP391: Working Within a Stroke System of Care to Decrease Door to Transfer Times From Spoketo Hub. Stroke, 2016) investigated how to reduce transport time from regional hospitals to central hospitals by improving the transport process in stroke networks. The results showed that by optimizing the coordination process and simplifying steps, the transport time for stroke patients can be significantly reduced, thereby improving treatment efficiency. However, these studies still have some limitations. For example, Reference 2 mainly focuses on optimizing processes and coordinating steps, lacking in-depth fusion analysis of multimodal image data, and cannot fully utilize image data to assist in transfer decisions, especially in complex cases, where the accuracy of decision-making still needs to be improved.

[0005] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of existing technologies, this invention provides a stroke patient transfer auxiliary decision-making system based on multimodal imaging. Through image data acquisition, processing, and deep learning model analysis, quantitative indicators of the patient's image features are obtained. Combined with the hospital's real-time resource database and geographical location information, a hospital allocation plan is generated using a genetic algorithm. This overcomes the problems of insufficient utilization of multimodal image data, lack of standardized evaluation, and neglect of the real-time status of hospital resources in existing technologies, thereby improving the accuracy and efficiency of stroke patient transfer decisions.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A stroke patient transport auxiliary decision-making system based on multimodal imaging includes an image data acquisition module, an image data processing module, an information analysis module, an auxiliary decision-making module, and a user interface. The image data acquisition module is connected to the image data processing module, the image data processing module is connected to the information analysis module, the information analysis module is connected to the auxiliary decision-making module, and the auxiliary decision-making module is connected to the user interface. The image data acquisition module acquires multimodal image data of the patient and transmits the acquired image data to the image data processing module. The image data processing module preprocesses, extracts, and fuses the received image data to obtain fused multimodal image feature information, and transmits the fused multimodal image feature information to the information analysis module. The information analysis module analyzes the received fused multimodal image feature information based on a preset deep learning model to obtain quantitative indicators of the patient's image features. These quantitative indicators are then input into an image feature evaluation model to evaluate the image feature index. The calculated image feature index evaluation value and the quantitative image feature index are transmitted to the auxiliary decision-making module. The formula for the image feature evaluation model is:

[0009]

[0010] In the formula, P is the image feature index evaluation value, and V b For the volume of high-density regions with CT values ​​greater than 60 HU, L v A is the ratio of ventricular volume to total intracranial volume. p P represents the area of ​​a low-density region with a CT value of 10-20 HU. f Mean cerebral blood flow;

[0011] The auxiliary decision-making module is used to generate a hospital allocation plan based on the received quantitative indicators and image feature index evaluation values, combined with a preset hospital resource database and geographical location information, and a genetic algorithm, and transmit the results to the user interface. The hospital resource database includes the number of beds, the number of doctors, and the number of patients in the current department.

[0012] As a further embodiment of the present invention, the image data processing module includes an image preprocessing unit, a feature extraction unit, and a feature fusion unit. The image preprocessing unit is used to perform denoising and enhancement preprocessing operations on the received image data. The feature extraction unit is used to extract high-density regions with CT values ​​greater than 60 HU and low-density regions with CT values ​​of 10-20 HU from computed tomography images, as well as the average cerebral blood flow, ventricular volume, and total intracranial volume from magnetic resonance images. The feature fusion unit is used to fuse the extracted image feature information to obtain fused multimodal image feature information.

[0013] As a further aspect of the present invention, the feature fusion unit employs a preset multimodal image feature fusion model to calculate the fused multimodal image feature information, the formula of which is:

[0014]

[0015] Where Q represents the fused multimodal image feature information, T represents the time span of the image data, λ represents the time decay coefficient, t represents the time point of the image data, w1 and w2 represent the weights of the computed tomography (CT) image feature information and the magnetic resonance imaging (MRI) image feature information, respectively, and w1 + w2 = 1, and f1(t) and f2(t) represent the feature values ​​of the CT image feature information and the MRI image feature information at time point t, respectively.

[0016] As a further embodiment of the present invention, the information analysis module includes a feature input unit for receiving fused multimodal image feature information transmitted by the image data processing module; an analysis model unit for calling a preset deep learning model to analyze the received fused multimodal image feature information to obtain the patient's image feature quantification indicators, including the volume of high-density areas with CT values ​​greater than 60 HU, the ratio of ventricular volume to total intracranial volume, the area of ​​low-density areas with CT values ​​of 10-20 HU, and the average cerebral blood flow; and an indicator output unit for outputting the image feature quantification indicators and image feature index evaluation values ​​obtained by the analysis model unit to the auxiliary decision-making module.

[0017] As a further aspect of the present invention, the analysis model unit uses a convolutional neural network model to analyze the received fused multimodal image feature information to obtain the volume of high-density regions with CT values ​​greater than 60 HU, the ratio of ventricular volume to total intracranial volume, the area of ​​low-density regions with CT values ​​of 10-20 HU, and the average cerebral blood flow.

[0018] As a further embodiment of the present invention, the auxiliary decision-making module includes an information receiving unit for receiving the patient's image feature quantification indicators and image feature index evaluation values ​​transmitted by the information analysis module, as well as a preset hospital resource database and geographical location information. The hospital resource database includes the number of beds, the number of doctors, and the number of patients in the current department.

[0019] As a further embodiment of the present invention, the auxiliary decision-making module includes a distance calculation unit and a matching degree evaluation unit. The distance calculation unit is used to calculate the distance from the patient to each hospital based on the patient's location and the geographical location information of each hospital. The matching degree evaluation unit is used to generate a hospital allocation scheme using a genetic algorithm, including the following steps:

[0020] Step S1: Randomly generate an initial solution set, where each solution represents a possible hospital allocation scheme;

[0021] Step S2: Encode the solution using integer encoding, where each gene represents a hospital number assigned to a patient, and set the population size;

[0022] Step S3: For each individual, randomly assign a hospital number to each patient; repeat this process until a number of different individuals equal to the preset population size are generated, thus forming the initial population;

[0023] Step S4: Calculate the fitness value for each solution based on the fitness function, where the formula for the fitness function is:

[0024]

[0025] In the formula, F is the fitness value, P is the image feature index assessment value, D is the distance from the patient to the hospital, and Pt is the hospital's reception capacity threshold.

[0026] The formula for calculating the threshold of a hospital's patient capacity is:

[0027]

[0028] In the formula, P t denoted as the threshold for the hospital's capacity to receive patients, D as the number of hospital beds, E as the number of hospital doctors, C as the current number of patients in the hospital's departments, and a1, a2, and a3 as the weighting coefficients for the number of beds, the number of doctors, and the current number of patients in the departments, respectively.

[0029] Step S5: Calculate the selection probability based on the fitness value of each solution, and use the roulette wheel method to select a number of individuals equal to the preset population size as parents;

[0030] Step S6: Perform a crossover operation, randomly select two parent solutions, and when the preset crossover probability is met, randomly select a crossover point between the two parent solutions to exchange some of the patients' hospital allocation schemes.

[0031] Step S7: Based on the hospital's geographical location and reception capacity threshold, the hospital allocation for each patient in each solution is mutated with a preset probability. Hospitals within a preset range and whose reception capacity threshold differs from the patient's image feature index assessment value by less than a preset threshold are selected for reassignment.

[0032] Step S8: Repeat steps S4 to S7 until the termination condition is met.

[0033] Step S9: Select the individual with the highest fitness value from the last generation of the population, decode the individual into a specific hospital allocation scheme, and output the final hospital allocation scheme.

[0034] As a further aspect of the present invention, the user interface is used to display the reception capacity threshold of each hospital and the hospital allocation scheme.

[0035] Compared with the prior art, the beneficial effects of the stroke patient transport auxiliary decision-making system based on multimodal imaging of the present invention are as follows:

[0036] This invention provides a more accurate and objective assessment of image features through the effective integration of multimodal image data and deep learning analysis. Secondly, the system takes into account the real-time resource status of the hospital and uses a genetic algorithm to generate an optimized transport plan, thereby achieving a match between patient needs and medical resources. This intelligent decision support overcomes the shortcomings of existing technologies that rely too much on subjective judgment and ignore dynamic changes in resources, thus improving the accuracy, efficiency, and utilization rate of transport decisions and medical resources. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the structure of a stroke patient transport auxiliary decision-making system based on multimodal imaging according to the present invention. Detailed Implementation

[0038] The technical solutions of this embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0039] Example 1

[0040] A stroke patient transport auxiliary decision-making system based on multimodal imaging includes an image data acquisition module, an image data processing module, an information analysis module, an auxiliary decision-making module, and a user interface.

[0041] In this embodiment of the invention, the image data acquisition module is connected to the image data processing module, the image data processing module is connected to the information analysis module, the information analysis module is connected to the decision support module, and the decision support module is connected to the user interface.

[0042] The image data acquisition module is used to acquire multimodal image data of patients and transmit the acquired image data to the image data processing module.

[0043] In this embodiment of the invention, the image data acquisition module uses an imaging technology operator to perform a whole-brain scan with a CT scanner and a multi-sequence scan with an MRI scanner, including T1-weighted imaging, T2-weighted imaging, diffusion-weighted imaging, and perfusion-weighted imaging. The acquired image data is saved in DICOM format and transmitted in real time to the image data processing module through the hospital's existing PACS network. The combination of multimodal image data provides more comprehensive information about the patient's condition, which is helpful for subsequent accurate analysis; the standardized acquisition process ensures the comparability and repeatability of the data.

[0044] The image data processing module is used to preprocess, extract features, and fuse the received image data to obtain fused multimodal image feature information, and then transmit the fused multimodal image feature information to the information analysis module.

[0045] The image data processing module in this embodiment of the invention includes an image preprocessing unit, a feature extraction unit, and a feature fusion unit. The image preprocessing unit is used to perform noise reduction and enhancement preprocessing operations on the received image data. The feature extraction unit is used to extract high-density regions with CT values ​​greater than 60 HU and low-density regions with CT values ​​of 10-20 HU from computed tomography images, as well as average cerebral blood flow, ventricular volume, and total intracranial volume from magnetic resonance images. The feature fusion unit is used to fuse the extracted image feature information to obtain fused multimodal image feature information.

[0046] In this embodiment of the invention, the feature fusion unit uses a preset multimodal image feature fusion model to calculate the fused multimodal image feature information, and the formula is as follows:

[0047]

[0048] Where Q represents the fused multimodal image feature information, T represents the time span of the image data, λ represents the time decay coefficient, t represents the time point of the image data, w1 and w2 represent the weights of the computed tomography (CT) image feature information and the magnetic resonance imaging (MRI) image feature information, respectively, and w1 + w2 = 1, and f1(t) and f2(t) represent the feature values ​​of the CT image feature information and the MRI image feature information at time point t, respectively.

[0049] The multimodal image feature fusion model in this embodiment of the invention provides a time integral result of the fused multimodal image feature information, reflecting the comprehensive features over the entire observation period. The total time span of the image data is typically from the patient's admission to the last imaging examination. A time decay coefficient, e, is introduced. -λtThe term ensures that the influence of earlier time points on the final result gradually decreases over time, which aligns with the principle in clinical practice that more recent data is given greater importance; t represents a specific time point, varying within the range of 0 to T; w1 and w2 represent the weights of the feature information from computed tomography (CT) and magnetic resonance imaging (MRI), respectively, and their sum of 1 ensures the rationality of the weight allocation. The weight design allows for adjustments to the importance of the two imaging modalities based on different situations; f 1(t) and f 2(t) These represent the feature values ​​of computed tomography (CT) and magnetic resonance imaging (MRI) images at time point t, including lesion volume, density, or signal intensity indices. The model can capture the dynamic changes in disease progression while balancing information from different time points and different imaging modalities. It considers both early imaging features that may reflect initial damage and later changes that may indicate disease progression or treatment response.

[0050] In this embodiment of the invention, the image preprocessing unit uses a Gaussian filter to denoise computed tomography (CT) images and magnetic resonance imaging (MRI) images, and applies histogram equalization for image enhancement. The feature extraction unit extracts high-density regions with CT values ​​greater than 60 HU and low-density regions with CT values ​​between 10 and 20 HU from CT images using a threshold segmentation method. Simultaneously, it uses a segmentation tool to extract ventricles and intracranial structures from MRI images, calculates ventricular volume and total intracranial volume, and calculates average cerebral blood flow using dynamic magnetic susceptibility contrast-enhanced perfusion imaging sequences. The feature fusion unit calculates the fused multimodal image feature information using functions, and the time decay coefficient and weights are obtained through cross-validation. The extraction and fusion of multimodal features provide more comprehensive image feature information, which helps improve the accuracy of the evaluation results. The time-weighted fusion model considers the importance of image data at different time points and provides a data foundation for subsequent auxiliary decision-making by integrating multimodal image information.

[0051] The information analysis module includes a feature input unit for receiving fused multimodal image feature information transmitted from the image data processing module; an analysis model unit for calling a preset deep learning model to analyze the received fused multimodal image feature information and obtain the patient's image feature quantification indicators, including the volume of high-density areas with CT values ​​greater than 60 HU, the ratio of ventricular volume to total intracranial volume, the area of ​​low-density areas with CT values ​​of 10-20 HU, and the average cerebral blood flow; and an indicator output unit for outputting the image feature quantification indicators and image feature index evaluation values ​​obtained by the analysis model unit to the auxiliary decision module.

[0052] In this embodiment of the invention, the feature input unit receives fused multimodal image feature information from the image data processing module via a data interface. This information is stored and transmitted in array form. The analysis model unit uses a pre-trained convolutional neural network as a deep learning model. The model receives the fused feature information as input and, through processing by convolutional layers and fully connected layers, outputs quantitative indicators of the patient's image features, including the volume of high-density areas with CT values ​​greater than 60 HU, the ratio of ventricular volume to total intracranial volume, the area of ​​low-density areas with CT values ​​between 10 and 20 HU, and the average cerebral blood flow. The indicator output unit integrates these quantitative indicators and the image feature index evaluation values ​​obtained through function calculation into a structured format and transmits them to the auxiliary decision-making module through a predefined interface. The information analysis module uses deep learning technology to automatically extract and analyze complex image features, transforming complex image information into specific quantitative indicators and evaluation values, providing doctors with more intuitive and objective information.

[0053] In this embodiment of the invention, the information analysis module is used to analyze the received fused multimodal image feature information based on a preset deep learning model, obtain the patient's image feature quantification index, input the image feature quantification index into the image feature evaluation model, evaluate the image feature index, and transmit the calculated image feature index evaluation value and image feature quantification index to the auxiliary decision-making module. The formula of the image feature evaluation model is:

[0054]

[0055] In the formula, P is the image feature index evaluation value, and V b For the volume of high-density regions with CT values ​​greater than 60 HU, L v A is the ratio of ventricular volume to total intracranial volume. p P represents the area of ​​a low-density region with a CT value of 10-20 HU. f This represents the average cerebral blood flow.

[0056] In the image feature assessment model formula of this invention, the volume of high-density areas with CT values ​​greater than 60 HU is usually associated with acute hemorrhagic stroke; the larger the volume, the more severe the hemorrhage. Using its logarithmic form can maintain sensitivity while avoiding excessive influence of large-volume hemorrhage on the assessment value. The ratio of ventricular volume to total intracranial volume reflects the degree of hydrocephalus or brain atrophy. Taking the exponent of this ratio can highlight the influence of abnormalities in the ventricular system. The area of ​​low-density areas with CT values ​​of 10-20 HU is usually associated with the infarct core area of ​​ischemic stroke. Taking its square root can balance the influence of small and large infarcts. Mean cerebral blood flow is an important indicator for assessing the perfusion status of brain tissue. Using its logarithmic form can provide sensitivity over a wide range of blood flow. The assessment model has good adaptability to both hemorrhagic and ischemic strokes, and takes into account factors of hydrocephalus and brain perfusion, providing a comprehensive assessment indicator.

[0057] In this embodiment of the invention, the analysis model unit uses a convolutional neural network model to analyze the received fused multimodal image feature information to obtain the volume of high-density regions with CT values ​​greater than 60 HU, the ratio of ventricular volume to total intracranial volume, the area of ​​low-density regions with CT values ​​of 10-20 HU, and the average cerebral blood flow.

[0058] In this embodiment of the invention, the input layer of the convolutional neural network model is designed to receive fused feature maps from computed tomography (CT) scans and magnetic resonance imaging (MRI), organizing the feature maps into multi-channel three-dimensional tensors. The network structure includes convolutional layers, pooling layers, batch normalization layers, and fully connected layers. The convolutional kernel size is 3x3x3 with a stride of 1 and padding of 1 to maintain spatial dimensions. Each convolutional layer is followed by an activation function and a batch normalization layer to improve the model's nonlinear expressive power and training stability. At the end of the network, a global average pooling layer is used to compress the feature maps, and then a fully connected layer outputs four target metrics: the volume of high-density regions with CT values ​​greater than 60 HU, the ratio of ventricular volume to total intracranial volume, the area of ​​low-density regions with CT values ​​between 10 and 20 HU, and average cerebral blood flow. The model is trained using the Adam optimizer with an initial learning rate of 0.001. The model is pre-trained on a stroke imaging dataset and then fine-tuned on a dataset from a target hospital to adapt to a specific clinical environment. This approach can automatically extract key features from complex multimodal image data, reducing the need for manual feature engineering. Furthermore, through end-to-end learning, the model can capture subtle patterns that human experts might overlook.

[0059] The auxiliary decision-making module is used to generate a hospital allocation plan based on the received quantitative indicators and image feature index evaluation values, combined with a preset hospital resource database and geographical location information, and a genetic algorithm, and transmit the results to the user interface. The hospital resource database includes the number of beds, the number of doctors, and the number of patients in the current department.

[0060] The auxiliary decision-making module in this embodiment of the invention includes an information receiving unit, which is used to receive the patient's image feature quantification indicators and image feature index evaluation values ​​transmitted by the information analysis module, as well as a preset hospital resource database and geographical location information. The hospital resource database includes the number of beds, the number of doctors, and the number of patients in the current department.

[0061] The auxiliary decision-making module in this embodiment of the invention includes a distance calculation unit and a matching degree evaluation unit. The distance calculation unit is used to calculate the distance from the patient to each hospital based on the patient's location and the geographical location information of each hospital. The matching degree evaluation unit is used to generate a hospital allocation scheme using a genetic algorithm, including the following steps:

[0062] Step S1: Randomly generate an initial solution set, where each solution represents a possible hospital allocation scheme;

[0063] Step S2: Encode the solution using integer encoding, where each gene represents a hospital number assigned to a patient, and set the population size;

[0064] Step S3: For each individual, randomly assign a hospital number to each patient; repeat this process until a number of different individuals equal to the preset population size are generated, thus forming the initial population;

[0065] Step S4: Calculate the fitness value for each solution based on the fitness function, where the formula for the fitness function is:

[0066]

[0067] In the formula, F is the fitness value, P is the image feature index assessment value, D is the distance from the patient to the hospital, and Pt is the hospital's reception capacity threshold.

[0068] The formula for calculating the threshold of a hospital's patient capacity is:

[0069]

[0070] In the formula, P t denoted as the threshold for the hospital's capacity to receive patients, D as the number of hospital beds, E as the number of hospital doctors, C as the current number of patients in the hospital's departments, and a1, a2, and a3 as the weighting coefficients for the number of beds, the number of doctors, and the current number of patients in the departments, respectively.

[0071] Step S5: Calculate the selection probability based on the fitness value of each solution, and use the roulette wheel method to select a number of individuals equal to the preset population size as parents;

[0072] Step S6: Perform a crossover operation, randomly select two parent solutions, and when the preset crossover probability is met, randomly select a crossover point between the two parent solutions to exchange some of the patients' hospital allocation schemes.

[0073] Step S7: Based on the hospital's geographical location and reception capacity threshold, the hospital allocation for each patient in each solution is mutated with a preset probability. Hospitals within a preset range and whose reception capacity threshold differs from the patient's image feature index assessment value by less than a preset threshold are selected for reassignment.

[0074] Step S8: Repeat steps S4 to S7 until the termination condition is met.

[0075] Step S9: Select the individual with the highest fitness value from the last generation of the population, decode the individual into a specific hospital allocation scheme, and output the final hospital allocation scheme.

[0076] Below is a Python code example that uses a genetic algorithm to obtain a hospital allocation scheme. Please note that this example is only a starting point and may need to be adjusted according to the actual situation and device interface in real applications.

[0077]

[0078]

[0079]

[0080] This code is for illustrative purposes only. In actual applications, it needs to be modified and adjusted according to specific circumstances.

[0081] The fitness function in this embodiment of the invention employs a variant of the Sigmoid function, incorporating multiple key factors to evaluate the fitness of patient allocation schemes. The function integrates the image feature index assessment value, the patient's distance to the hospital, and the hospital's capacity threshold into a mathematical model. The image feature index assessment value reflects the patient's urgency; a higher score may indicate a need for more urgent medical intervention. The patient's distance to the hospital considers geographical location; a shorter distance means a faster response time. The hospital's capacity threshold represents the hospital's resource constraints; exceeding this threshold may lead to excessive strain on medical resources. The exponential term in the function comprehensively evaluates the rationality of the allocation scheme by comparing the difference between the image feature index assessment value and the hospital's capacity threshold, while also considering distance. When the image feature index assessment value is significantly higher than the hospital's capacity threshold plus the distance factor, the exponential term approaches positive infinity, making the fitness value close to 1, indicating a highly suitable allocation scheme. The coefficients in the function are used to adjust the steepness of the curve and can be fine-tuned according to actual conditions to better reflect the weights of different factors. The fitness function is designed to smoothly handle various possible scenarios, taking into account both the need for nearby treatment in specific situations and avoiding the over-concentration of medical resources. Furthermore, the continuity of the function allows it to perform well in complex optimization algorithms, which is beneficial for finding the global optimum.

[0082] The user interface is used to display the patient capacity thresholds of each hospital and the hospital allocation plan.

[0083] The user interface in this embodiment of the invention displays the patient reception capacity threshold in real time, allowing users to immediately identify hospitals that are nearing saturation and those with the capacity to receive more patients, thereby making more timely and accurate resource allocation decisions. The visual display of the hospital allocation plan helps managers to fully understand the patient diversion situation and promotes the balanced use of medical resources. By clearly displaying the load of each hospital, it is possible to avoid situations where some hospitals are overcrowded while other hospitals have idle resources.

[0084] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0085] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A stroke patient transport auxiliary decision-making system based on multimodal imaging, comprising an image data acquisition module, an image data processing module, an information analysis module, an auxiliary decision-making module, and a user interface, characterized in that, The image data acquisition module is connected to the image data processing module, which in turn is connected to the information analysis module. The information analysis module is connected to the decision support module, which is also connected to the user interface. The image data acquisition module acquires multimodal image data from the patient and transmits the acquired image data to the image data processing module. The image data processing module preprocesses, extracts, and fuses the received image data to obtain fused multimodal image feature information, which is then transmitted to the information analysis module. The information analysis module analyzes the received fused multimodal image feature information based on a preset deep learning model to obtain quantitative indicators of the patient's image features. These quantitative indicators are then input into an image feature evaluation model to evaluate the image feature index. The calculated image feature index evaluation value and the quantitative image feature index are transmitted to the decision support module. The formula for the image feature evaluation model is: In the formula, P is the image feature index evaluation value, and V b For the volume of high-density regions with CT values ​​greater than 60 HU, L v A is the ratio of ventricular volume to total intracranial volume. p P represents the area of ​​a low-density region with a CT value of 10-20 HU. f Mean cerebral blood flow; The auxiliary decision-making module is used to generate a hospital allocation plan based on the received quantitative indicators and image feature index evaluation values, combined with a preset hospital resource database and geographical location information, and a genetic algorithm, and transmit the results to the user interface. The hospital resource database includes the number of beds, the number of doctors, and the number of patients in the current department.

2. The stroke patient transport auxiliary decision-making system based on multimodal imaging according to claim 1, characterized in that, The image data processing module includes an image preprocessing unit, a feature extraction unit, and a feature fusion unit. The image preprocessing unit is used to perform noise reduction and enhancement preprocessing operations on the received image data. The feature extraction unit is used to extract high-density regions with CT values ​​greater than 60 HU and low-density regions with CT values ​​of 10-20 HU from computed tomography images, as well as average cerebral blood flow, ventricular volume, and total intracranial volume from magnetic resonance images. The feature fusion unit is used to fuse the extracted image feature information to obtain fused multimodal image feature information.

3. The stroke patient transport auxiliary decision-making system based on multimodal imaging according to claim 2, characterized in that, The feature fusion unit uses a preset multimodal image feature fusion model to calculate the fused multimodal image feature information, and the formula is as follows: Where Q represents the fused multimodal image feature information, T represents the time span of the image data, λ represents the time decay coefficient, t represents the time point of the image data, w1 and w2 represent the weights of the computed tomography (CT) image feature information and the magnetic resonance imaging (MRI) image feature information, respectively, and w1 + w2 = 1, and f1(t) and f2(t) represent the feature values ​​of the CT image feature information and the MRI image feature information at time point t, respectively.

4. The stroke patient transport auxiliary decision-making system based on multimodal imaging according to claim 1, characterized in that, The information analysis module includes a feature input unit for receiving fused multimodal image feature information transmitted by the image data processing module; an analysis model unit for calling a preset deep learning model to analyze the received fused multimodal image feature information and obtain the patient's image feature quantification indicators, including the volume of high-density areas with CT values ​​greater than 60 HU, the ratio of ventricular volume to total intracranial volume, the area of ​​low-density areas with CT values ​​of 10-20 HU, and the average cerebral blood flow; and an indicator output unit for outputting the image feature quantification indicators and image feature index evaluation values ​​obtained by the analysis model unit to the auxiliary decision module.

5. A stroke patient transport auxiliary decision-making system based on multimodal imaging according to claim 4, characterized in that, The analysis model unit uses a convolutional neural network model to analyze the received fused multimodal image feature information to obtain the volume of high-density areas with CT values ​​greater than 60 HU, the ratio of ventricular volume to total intracranial volume, the area of ​​low-density areas with CT values ​​of 10-20 HU, and the average cerebral blood flow.

6. The stroke patient transport auxiliary decision-making system based on multimodal imaging according to claim 1, characterized in that, The decision support module includes an information receiving unit, which receives the patient's image feature quantification indicators and image feature index evaluation values ​​transmitted by the information analysis module, as well as a preset hospital resource database and geographical location information. The hospital resource database includes the number of beds, the number of doctors, and the number of patients in the current department.

7. A stroke patient transport auxiliary decision-making system based on multimodal imaging according to claim 1, characterized in that, The decision support module includes a distance calculation unit and a matching degree evaluation unit. The distance calculation unit is used to calculate the distance from the patient to each hospital based on the patient's location and the geographical location information of each hospital. The matching degree evaluation unit is used to generate a hospital allocation scheme using a genetic algorithm, including the following steps: Step S1: Randomly generate an initial solution set, where each solution represents a possible hospital allocation scheme; Step S2: Encode the solution using integer encoding, where each gene represents a hospital number assigned to a patient, and set the population size; Step S3: For each individual, randomly assign a hospital number to each patient; repeat this process until a number of different individuals equal to the preset population size are generated, thus forming the initial population; Step S4: Calculate the fitness value for each solution based on the fitness function, where the formula for the fitness function is: In the formula, F is the fitness value, P is the image feature index assessment value, D is the distance from the patient to the hospital, and Pt is the hospital's reception capacity threshold. The formula for calculating the threshold of a hospital's patient capacity is: In the formula, P t , where D is the threshold for the hospital's capacity to receive patients, E is the number of beds in the hospital, C is the number of doctors in the hospital, and a1, a2, and a3 are the weighting coefficients for the number of beds, the number of doctors, and the number of patients in the current department, respectively. Step S5: Calculate the selection probability based on the fitness value of each solution, and use the roulette wheel method to select a number of individuals equal to the preset population size as parents; Step S6: Perform a crossover operation, randomly select two parent solutions, and when the preset crossover probability is met, randomly select a crossover point between the two parent solutions to exchange some of the patients' hospital allocation schemes. Step S7: Based on the hospital's geographical location and reception capacity threshold, the hospital allocation for each patient in each solution is mutated with a preset probability. Hospitals within a preset range and whose reception capacity threshold differs from the patient's image feature index assessment value by less than a preset threshold are selected for reassignment. Step S8: Repeat steps S4 to S7 until the termination condition is met. Step S9: Select the individual with the highest fitness value from the last generation of the population, decode the individual into a specific hospital allocation scheme, and output the final hospital allocation scheme.

8. A stroke patient transport auxiliary decision-making system based on multimodal imaging according to claim 1, characterized in that, The user interface is used to display the patient capacity thresholds of each hospital and the hospital allocation plan.

Images