Nursing auxiliary system in first-aid transfer process
By utilizing image classification technology with inquiry prompting module, information collection module and convolutional neural network model during emergency transport, a rapid and accurate treatment plan is provided, solving the problem of lost emergency opportunities due to the lack of professional doctors in ambulances, and realizing low-cost and efficient emergency support.
Patent Information
- Application Number
- CN202510057685.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-14
- Publication Date
- 2026-01-13
AI Technical Summary
During emergency rescue, the lack of professional doctors on ambulances and reliance on the experience of medical staff can easily lead to missed treatment opportunities due to traffic jams and other problems. Existing big data models are costly and have limited application scenarios, making them difficult to promote widely.
Design a nursing assistance system for emergency transport, utilizing an inquiry prompt module, an information collection module, an auxiliary diagnosis module, and a display unit, combined with a convolutional neural network model and image classification technology, to quickly match and provide treatment plans, reducing costs and improving accuracy.
It enables the rapid and accurate provision of treatment plans during emergency care, reduces software development costs, and improves the accuracy and timeliness of nursing care during emergency transport.
Smart Images
Figure CN121331418A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of medical equipment based on digital information technology, and particularly relates to an emergency transfer assisting system for assisting medical staff by using big data. BACKGROUND
[0002] With the development of society and the acceleration of life rhythm, various accidents also come. In the face of unexpected situations, first aid and rescue are good means for people to save themselves and protect others. For first aid and rescue, an ambulance is undoubtedly a necessary part. Although the ambulance is equipped with rescue personnel with rescue experience, they are not professional doctors. In the process of transporting patients for first aid, they can only perform some basic bandaging and simple nursing. Once encountering traffic jams and other problems, it is easy to cause the patient to miss the opportunity for treatment in the process of first aid, resulting in irreparable damage or even loss of life.
[0003] At present, with the development of information technology, more and more auxiliary systems based on patient data are applied to the patient diagnosis process. However, the use of big data in the process of first aid is still relatively lacking. First aid still mainly relies on the experience and technology of medical staff. It is difficult to obtain corresponding financial support due to the high cost of developing a big data model specifically for first aid and the limited application scenarios. SUMMARY
[0004] In view of the above problems, the present application hopes to provide a low-cost nursing system based on digital information technology, which can provide better rescue suggestions for ambulance personnel during the transportation of patients by the ambulance, and thus better provide treatment for patients.
[0005] Specifically, the present application provides a nursing assisting system in the process of first aid, characterized in that the system comprises: an inquiry prompting module, an information acquisition module, an auxiliary diagnosis module, an emergency database and a display unit,
[0006] The inquiry prompting module is used to generate pre-questions required in the process of first aid,
[0007] The information acquisition module is used to acquire the response information of the patient or the patient's family to the pre-questions and to acquire the current physiological parameter data of the patient monitored by the physiological monitoring system,
[0008] The auxiliary diagnosis module is used to obtain the cause information of the patient's this time first aid based on the response information of the patient's family to the pre-questions, and to retrieve the patient's past medical history data. The auxiliary diagnosis module retrieves the approximate processing cases from the emergency database based on the physiological parameter data, the cause information and the past medical history data by using a convolutional neural network model, and generates reference processing opinions according to the approximate processing cases. The display unit receives and displays the reference processing opinions.
[0009] The auxiliary diagnosis module has a trained neural network model for image classification built-in. The information collection module maps the obtained physiological parameter data, previous medical history data and emergency cause information to different regions of a template image. The template image includes an emergency cause region, a physiological parameter region and a previous medical history region. (1) Different emergency causes are assigned different pixel blocks. For the current emergency cause, the emergency personnel input the corresponding severity level. Each severity level is assigned a different pixel value for the pixel block. (2) For different physiological parameters, different pixel blocks of the physiological parameters are used. The pixel values in the pixel blocks are used to represent the normalized parameter values of the physiological parameters. (3) For different types of previous medical history related to emergency, different pixel blocks of the previous medical history are assigned. For each type of previous medical history, the more serious the block is to life, the greater the pixel gray value is.
[0010] The auxiliary diagnosis module is configured to: after the information of the current patient is mapped to the template image to obtain a current patient information image, use the trained neural network model to classify the patient information image of the current patient, and find the existing patient information that is most similar to the classification of the patient information image of the current patient and the emergency treatment information of the existing patient.
[0011] Preferably, the emergency cause region is located in the center region of the image, the physiological parameter region surrounds the periphery of the emergency cause region, and the previous medical history region surrounds the periphery of the physiological parameter region.
[0012] Preferably, the physiological parameter region further includes an expression region for different injured parts of the patient.
[0013] The neural network model has a multi-layer structure, including two or three convolution layers, two or three pooling layers, one or two fully connected layers and a softmax regression layer.
[0014] Preferably, the size of the patient information image is 256*256.
[0015] Preferably, the pixel coordinates [96, 96], [96, 159], [159, 159] and [159, 96] are used as four corner points to form a square region as the emergency cause region. The emergency cause region is divided into multiple blocks, each block corresponding to a different attribute category.
[0016] The pixel coordinates [64, 64], [64, 191], [191, 64], [191, 191] as four corner points of the square region surrounded by the outside, the pixel coordinates [0, 0], [0, 255], [255, 0], [255, 255] as four corner points of the square region surrounded by the inside as the past medical history area, and each 32*32 pixel area as a type of past medical history representation area, the different types of past medical history areas are divided.
[0017] The pixel coordinates [64, 64], [64, 191], [191, 64], [191, 191] as four corner points of the square region surrounded by the outside, the pixel coordinates [0, 0], [0, 255], [255, 0], [255, 255] as four corner points of the square region surrounded by the inside as the past medical history area, and each 32*32 pixel area as a type of past medical history representation area, the different types of past medical history areas are divided.
[0018] Preferably, the method further comprises constructing a weight template image, the weight template image is divided in the same way as the template image, each partition is weighted according to the influence degree of the corresponding partition on the emergency process, and before classification, the corresponding region pixel value of each patient information image is respectively multiplied with the weight value in the weight template image to obtain a new patient information image.
[0019] Preferably, the method further comprises respectively classifying patients according to the core partition, the core partition plus the first peripheral partition and the complete partition image of the template image, respectively, to realize classification of different fine granularities.
[0020] Beneficial effects
[0021] The present application ingeniously converts the patient information in the emergency process into an image, uses the developed image data classification method to match and classify the image, and provides a similar patient rescue scheme as a reference for the nursing staff in the current emergency through the matching of patient data, better solves the problem of difficult determination of a rescue scheme in an emergency. The software development cost is reduced, and a classification model with high accuracy can be quickly constructed, which provides technical support for improving the accuracy and timeliness of nursing staff treatment in the future emergency transfer process.
[0022] In addition, the present application adopts an image mapping manner with gradually decreasing importance from the center to the periphery, so that the image classification can be gradually expanded from the inner ring to the outside. Although the image mapping in the present application is performed in three partitions, those skilled in the art can adopt more partitions, so that for patients with different emergency levels or different severity, the range of patient information image compared in the matching process can be gradually expanded from the core area image at the center. For example, in one model or one comparison, the core partition is input into the model for classification, then the core partition and the first peripheral partition are input into the model, and then the complete partition image is input into the model for classification, so as to realize classification with different fine granularity. In another implementation manner, in order to increase the accuracy of image classification, the classification results obtained by three times of classification can be compared, and only the existing patient information with the same classification results obtained by three levels as the feedback result. BRIEF DESCRIPTION OF DRAWINGS
[0023] Figure 1 Flow chart of patient data matching of the nursing system of the present application;
[0024] Figure 2 Structure schematic diagram of the nursing auxiliary system of the present application;
[0025] Figure 3 Schematic diagram of an image template according to an embodiment of the present application;
[0026] Figure 4 Schematic diagram of a neural network architecture according to an embodiment of the present application. DETAILED DESCRIPTION
[0027] The present application will be further described in detail below in combination with embodiments and drawings, but the embodiments of the present application are not limited thereto.
[0028] As Figure 1 shown is a structure schematic diagram of the nursing auxiliary system in the emergency transfer process of the present embodiment.
[0029] The nursing auxiliary system in the present embodiment comprises an inquiry prompting module 1, an information collecting module 2, an auxiliary diagnosis module 3, an emergency database 4 and a display unit 5.
[0030] The inquiry prompting module 1 is used to generate the pre-questions of the required inquiries in the first aid process, and after the ambulance arrives at the scene, the rescuer can inquire the patient or the patient's family according to the experience or according to the inquiry prompting module.
[0031] The information acquisition module 2 includes a voice recognition device and / or an input device (such as a keyboard, a touchpad), which is used to acquire the response information of the patient or the patient's family to the pre-questions, and the information acquisition module 2 receives the current physiological parameter data of the patient from the physiological monitoring system such as a sphygmomanometer, an electrocardiograph, an oximeter and the like carried on the ambulance.
[0032] The auxiliary diagnosis module is used to acquire the cause information of the patient's current first aid based on the response information of the patient's family to the pre-questions, and to retrieve the patient's past medical history data, and the auxiliary diagnosis module retrieves the approximate processing cases from the first aid database based on the physiological parameter data, the cause information and the past medical history data, and generates the reference processing opinions according to the approximate processing cases, and the display unit receives and displays the reference processing opinions.
[0033] The auxiliary diagnosis module in the present application is a fast auxiliary diagnosis module based on a CNN network model, and the CNN network model for image classification is built in the auxiliary diagnosis module, which is used to find the first aid cases matching the current symptoms of the patient from the first aid database.
[0034] The information acquisition module maps the obtained physiological parameter data, past medical history data and first aid cause information to different regions of the template image, and the template image includes a first aid cause region, a physiological parameter region and a past medical history region (an image mapping mode with gradually decreasing importance from the center to the periphery), the first aid cause region is located in the center region of the image, the physiological parameter region surrounds the periphery of the first aid cause region, and the past medical history region surrounds the periphery of the physiological parameter region, different first aid causes are allocated to different pixel blocks, the corresponding severity level is input by the rescuer, each severity level is assigned to a different pixel block, and the different injured parts of the patient are taken into account as part of the physiological parameter region, for example, head injury is assigned to B1 region, and the assigned value is proportional to the damage degree, and for physiological parameters such as heart rate, oxygen saturation and blood pressure, the corresponding physiological parameter blocks are allocated, and each physiological block reflects different physiological parameters through the gray value of the pixel and the change of the gray value.
[0035] This yields a patient information map representing the patient's emergency condition and physiological status. Image similarity judgment and classification already have relatively mature classification models, allowing the classification of the patient's emergency condition to be converted into image classification, significantly reducing the cost of rebuilding an emergency-based classification model. In this invention, a CNN model is used to find cases and emergency treatment plans most similar to the current patient's emergency condition, thereby providing rescue personnel with a reference for rescue efforts.
[0036] Construct a CNN network model with an SVM classifier (or use an existing mature CNN network model with image classification capabilities).
[0037] In this embodiment, a multi-layered CNN network model is constructed, including two or three convolutional layers, two or three pooling layers, one or two fully connected layers, and a softmax regression layer. The neural network model has a multi-layered structure, including two or three convolutional layers, two or three pooling layers, one or two fully connected layers, and a softmax regression layer. Figure 3 The diagram shows the network architecture in one implementation of the present invention.
[0038] In this embodiment, as Figure 4 As shown, the description takes the construction of a 256*256 patient information image as an example, dividing the image into three large areas. Those skilled in the art will understand that other partitioning methods can also be used, but circular partitioning is sufficient.
[0039] The square pixel area enclosed by the four corner points [96,96], [96,159], [159,159], [159,96] is designated as the emergency treatment cause area. This area is further divided into multiple blocks, each corresponding to a different attribute category. For simplicity, it is divided into four blocks, each occupying one-quarter of the pixel area. These blocks could be categorized as accident, sudden illness, worsening of chronic illness, or other. Different accident or illness categories are assigned to different grayscale value ranges within each area. The start and end points are each represented as a single pixel within this area. Horizontally, [96,96] to [159,96] contains 64 pixel coordinates, and vertically, [96,96] to [96,159] contains 64 pixel coordinates. The assignment of pixel coordinates in the following sections follows the same method and will not be repeated.
[0040] The area defined by the pixel coordinates [64,64], [64,191], [191,64], [191,191] is the inner side of the square region enclosed by the four corner points, and the area defined by the coordinates [96,96], [96,159], [159,159], [159,96] is the outer side of the square region enclosed by the four corner points. This area is designated as the physiological parameter area. The physiological parameter area is further divided into regions, with each 32*32 pixel region serving as a representation area for a physiological parameter indicator. If the current physiological indicator has more sub-indicators, the region is further divided. The value of each physiological parameter is normalized and converted into the corresponding grayscale value.
[0041] The region defined by the pixel coordinates [64,64], [64,191], [191,64], [191,191] as the outer side of the square region enclosed by the four corner points, and the region defined by the pixel coordinates [0,0], [0,255], [255,0], [255,255] as the inner side of the square region enclosed by the four corner points, is designated as the past medical history region. Each 32*32 pixel region is designated as a type of past medical history representation region. Different types of past medical history regions are divided, such as urinary system diseases, respiratory system diseases, hematological system diseases, digestive system diseases, etc. Unassigned representation regions are left blank.
[0042] After converting the patient's emergency care and physical condition information into a corresponding patient information map, a CNN network can be used to classify patients and find similar patients. However, before this, the CNN network model needs to be trained using existing data. Considering the limited training data, the applicant randomly selected emergency care data and past medical history data from 100 emergency patients, and cross-substituted different data from different patients to form simulated augmented data, obtaining 2000 sets of cross-augmented patient data for testing and validation.
[0043] 1600 sets of data were used for model training, and 400 sets of data were used for model testing.
[0044] Image data is first fed into the first convolutional layer for feature vector extraction. The output of the first convolutional layer is fed into the first pooling layer for dimensionality reduction. The output of the first pooling layer is fed into the second convolutional layer for feature extraction. The output of the second convolutional layer is fed into the second pooling layer. The output of the second pooling layer is fed into the third convolutional layer. The output of the third convolutional layer is fed into the third pooling layer. The output of the third pooling layer is fed into the fully connected layer to output feature vectors and the softmax regression layer.
[0045] Finally, a softmax regression layer is used to back-train the CNN, with the cross-entropy function as the error cost function. The fully connected layer outputs a feature vector, which is then used as input to an SVM classifier for image classification. Dropout is implemented to avoid overfitting. The model is repeatedly trained using the training dataset. After training, the model is loaded into the emergency medical services system.
[0046] In use, patient data is converted into images and input into the model as described above. Convolution and pooling are performed on the patient data to obtain the output feature vector of the fully connected layer. The model retrieves the training data with the highest similarity to the feature vector of the patient data, and the emergency treatment methods of the training data are output to the rescue personnel for their reference, providing them with the emergency treatment plan for the patient whose situation is most similar to that of the patient.
[0047] The trained model was tested using 400 sets of test data. The test results showed that the accuracy of classifying the test data was as high as 98.3%.
[0048] In a preferred implementation, diagnostic data of reference patients are selected from the same category, and the vector distance between the fully connected layer output feature vector of the reference patient and the patient to be tested is calculated to find the reference patient with the smallest vector distance and use it as a rescue reference.
[0049] In another implementation, diagnostic data from several reference patients in the same category are selected, and the treatment plans for these reference patients are displayed in a list format, providing more comparative references for emergency personnel.
[0050] In another preferred implementation, the scope of patient information images compared during the matching process is gradually expanded from the central core region image. Models are constructed for the core partition, the core partition plus the first outer partition, and the complete three-part partition. During comparison, the core partition is input into the core partition classification model for classification. Then, the core partition plus the first outer partition is input into the second model, and finally, the complete partition image is input into the complete model for classification. The three models use the same or different model architectures, but the training images used are patient information images with different coverage areas to achieve different levels of fine-grained classification.
[0051] To increase the accuracy of image classification, the classification results obtained from three classifications are compared, and only existing patient information that matches the classification results obtained from all three levels is output as feedback.
[0052] This approach can further ensure the accuracy of classification and avoid misclassification.
[0053] Although the principles of the present invention have been described in detail above with reference to preferred embodiments, those skilled in the art should understand that the above embodiments are merely illustrative explanations of the implementation of the present invention and are not intended to limit the scope of the present invention. The details in the embodiments do not constitute a limitation on the scope of the present invention. Any obvious changes, such as equivalent transformations or simple substitutions, based on the technical solutions of the present invention without departing from the spirit and scope of the present invention fall within the protection scope of the present invention.
Claims
1. A nursing support system for emergency transport, characterized in that, The system includes: an inquiry and prompting module, an information collection module, an auxiliary diagnosis module, an emergency database, and a display unit. The question prompting module is used to generate preliminary questions that need to be asked during the emergency rescue process. The information acquisition module is used to collect the patient's or their family's responses to pre-existing questions and to collect the patient's current physiological parameter data monitored by the physiological monitoring system. The auxiliary diagnostic module is used to obtain the cause information of the patient's emergency treatment based on the response information of the patient's family to the pre-existing questions, and to retrieve the patient's past medical history data. Based on physiological parameter data, cause information, and past medical history data, the auxiliary diagnostic module uses a convolutional neural network model to retrieve approximate treatment cases from the emergency database and generates reference treatment opinions based on the approximate treatment cases. The display unit receives and displays the reference treatment opinions. The auxiliary diagnosis module has a built-in trained neural network model for image classification. The information acquisition module maps the obtained physiological parameter data, past medical history data and emergency cause information to different regions of the template image. The template image includes an emergency cause area, a physiological parameter area and a past medical history area. (1) Different emergency causes are assigned different pixel blocks. For the current emergency cause, the emergency personnel input the corresponding severity level. Each severity level is assigned a different pixel value to the pixel block. (2) For different physiological parameters, different physiological parameter pixel blocks are assigned. The pixel values in the pixel blocks represent the normalized parameter values of the physiological parameters. (3) For different types of past medical history related to emergency care, different past medical history pixel blocks are assigned. For each type of past medical history, the more life-threatening the block, the larger the pixel gray value. The auxiliary diagnostic module is used to: after mapping the current patient's information to a template image to obtain the current patient information image, use a trained neural network model to classify the current patient's information image, and find the existing patient information and emergency treatment information of the existing patient that are most similar in classification to the current patient's information image.
2. The nursing support system for emergency transport according to claim 1, characterized in that, The emergency cause area is located in the center of the image, the physiological parameter area surrounds the emergency cause area, and the past medical history area surrounds the physiological parameter area.
3. The nursing support system for emergency transport according to claim 1, characterized in that, The physiological parameter area also includes expression areas for different injured sites of the patient.
4. The nursing support system for emergency transport according to claim 1, characterized in that, The neural network model has a multi-layer structure, including two or three convolutional layers, two or three pooling layers, one or two fully connected layers, and a softmax regression layer.
5. The nursing support system for emergency transport according to claim 1, characterized in that, The patient information image has a size of 256*256.
6. The nursing support system for emergency transport according to claim 5, characterized in that, The square area enclosed by the four corner points [96,96],[96,159],[159,159],[159,96] is used as the emergency cause area. The emergency cause area is divided into multiple blocks, and each block corresponds to a different attribute category. The area within the square formed by the four corner points is defined as the area with pixel coordinates [64,64], [64,191], [191,64], [191,191]. The area outside the square formed by the four corner points is defined as the physiological parameter area. The physiological parameter area is divided into regions, with each 32*32 pixel region serving as a representation area for a physiological parameter index. The regions with pixel coordinates [64,64], [64,191], [191,64], [191,191] are considered as the outer side of the square region enclosed by the four corner points, while the regions with pixel coordinates [0,0], [0,255], [255,0], [255,255] are considered as the inner side of the square region enclosed by the four corner points. These regions are designated as the past medical history regions, with each 32*32 pixel region representing a different type of past medical history region. This classification is used to distinguish between different types of past medical history regions.
7. The nursing support system for emergency transport according to claim 6, characterized in that, The method further includes constructing a weighted template image, which is partitioned in the same way as the template image. Each partition is weighted according to the degree of influence of the corresponding partition on the emergency process. Before classification, the pixel values of the corresponding area of each patient information image are multiplied by the weight values in the weighted template image to obtain a new patient information image.
8. The nursing support system for emergency transport according to claim 6, characterized in that, The method further includes classifying patients according to the core partition of the template image, the core partition plus the first outer partition, and the complete partition image, respectively, to achieve classification with different fine granularities.