Newborn resuscitation auxiliary system based on life parameter collection and evaluation
Through video, a variety of life parameters of the newborn are collected and deep learning technology is used to evaluate muscle tone and skin tone. A personalized resuscitation plan is recommended, which solves the problem of low resuscitation success rate in newborns and improves resuscitation efficiency and accuracy.
Patent Information
- Application Number
- CN202411150850.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-08-07
- Filing Date
- 2024-08-21
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, the differences in resuscitation skills and low training levels of newborns lead to low resuscitation success rate. How to improve the success rate of newborns' resuscitation is an urgent problem.
A neonatal resuscitation assistive system based on life parameter collection and evaluation is adopted to collect multiple life parameters through video, such as skin color, respiratory rate, heart rate, etc., and a multi-layer convolutional neural network is used to process image data, obtain muscle tone characteristics, and analyze pupil light. Stimulate responses, establish a skin color sample library, generate feature vectors and calculate similarity, and quickly recommend the most suitable resuscitation plan.
It provides more accurate data support, improves recovery efficiency, reduces errors caused by clinical experience, improves the success rate of recovery, and reduces the work burden of medical staff. The recommended solutions are only used as reference in the recovery process.
Smart Images

Figure CN118919028A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and more particularly to a neonatal resuscitation assistance system based on life parameter collection and evaluation. Background Art
[0002] It is estimated that approximately 10% of newborns require assistance to initiate breathing at birth, and approximately 1% require intensive resuscitation measures to restore cardiopulmonary function. Neonatal mortality has decreased in recent years from nearly 20 per 1000 live births in the 1960s to approximately 4 per 1000 live births today. The inability of newborns to establish and maintain adequate or spontaneous breathing is an important cause of these early deaths, as well as adverse neurodevelopmental outcomes in survivors. Therefore, effective and timely resuscitation at birth can further improve neonatal outcomes. Successful neonatal resuscitation efforts depend on critical actions that must be taken in rapid succession to maximize the chance of survival.
[0003] However, resuscitation skills may vary between different healthcare providers, especially in some healthcare settings, where lower levels of experience and training may affect resuscitation success rates.
[0004] Therefore, how to improve the success rate of neonatal resuscitation is an urgent problem that needs to be solved at this stage. Summary of the invention
[0005] In view of this, the present invention provides a neonatal resuscitation assistance system based on vital parameter collection and evaluation. By collecting multiple vital parameters such as skin color, respiratory rate, heart rate, etc. through video, it can comprehensively evaluate the health status of the newborn and provide more accurate data support for resuscitation.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] A neonatal resuscitation auxiliary system based on life parameter collection and evaluation comprises: a main control module, a data acquisition module, a simulation prediction module, a database, and a solution recommendation module, wherein the data acquisition module performs video acquisition on the neonate to obtain multi-dimensional parameter information, wherein the multi-dimensional parameter information comprises at least three items of skin color data, amniotic fluid data, muscle tension data, respiratory rate data, birth data, pupil data, and heart rate data; the data acquisition module sends the multi-dimensional parameter information to the simulation prediction module through the main control module for data processing, and sends the processed data to the solution recommendation module through the main control module; the solution recommendation module performs a similarity comparison on the processed data through a preset similarity threshold according to the historical processing data pre-stored in the database and the corresponding solution, and selects the solution corresponding to the historical processing data greater than or equal to the similarity threshold for solution recommendation.
[0008] Preferably, the simulation prediction module includes a muscle tension evaluation module, a skin color evaluation module, and a pupil evaluation module, which respectively evaluate the muscle tension data, skin color data, and pupil data to obtain muscle tension evaluation values, skin color evaluation values, and pupil light reflection evaluation values.
[0009] Preferably, the muscle tension assessment module specifically includes:
[0010] The key frame extraction module collects the video of the contraction amplitude of the lower limbs of newborns during non-invasive electrical stimulation of the foot, and extracts multiple key frame data according to a predetermined sampling frequency;
[0011] A feature tensor extraction module, which obtains a plurality of action feature matrices by respectively passing the plurality of key frame data through a feature extractor, and arranges the plurality of action feature matrices into a three-dimensional feature tensor along a channel dimension;
[0012] An associated feature extraction module, which obtains a first action feature map and a second action feature map respectively by passing the three-dimensional feature tensor through a first-scale convolutional neural network and a second-scale convolutional neural network;
[0013] An evaluation result generation module fuses the first motion feature map and the second motion feature map to obtain a classification feature map, and passes the classification feature map through a classifier to obtain a muscle tension evaluation value of the newborn.
[0014] Preferably, the classification feature graph obtained in the evaluation result generation module specifically includes:
[0015]
[0016] Among them, F 3 is the classification feature map, F 1 is the first action feature map, F 2 is the second action characteristic graph, It represents the addition of the elements at corresponding positions of the first action feature map and the second action feature map, and α and β are weighting parameters used to control the balance between the first action feature map and the second action feature map in the classification feature map.
[0017] Preferably, it also includes a feature distribution correction module, which is used to perform feature distribution correction on the classification feature map to obtain a corrected classification feature map, specifically including:
[0018]
[0019] Among them, f i is the feature value of the i-th position of the classification feature map, f j is the feature value of the jth position of the classification feature map, is the mean of all eigenvalues of the classification feature map, N is the scale of the classification feature map, exp(·) represents the exponential operation of the value, and f i ′ is the feature value of each position of the corrected classification feature map.
[0020] Preferably, the pupil assessment module specifically includes:
[0021] The main key point classification module obtains the key points of the neonatal pupil, obtains the relevance score for the contraction speed and the earliest contraction time of each key point based on the video of the neonatal pupil stimulated by light, sets the relevance score threshold, and classifies the key point as a main key point if the relevance score of the key point is greater than or equal to the relevance score threshold;
[0022] A related point determination module, which obtains key points whose shrinkage amplitude exceeds a preset shrinkage threshold among the main key points as related points, constructs a related point set, and assigns weights to the related points;
[0023] A key frame screening module calculates the change in pupil diameter of the relevant points corresponding to all two adjacent frames in the video, selects the weight of the corresponding relevant point when the change is less than a change threshold, counts the total weight of the relevant points on each frame image in the video, and selects a key frame according to the total weight of the relevant points on each frame image and a set total weight threshold;
[0024] The image detection module obtains the main area of the key frame under different lighting stimulations and sends it to the image classification network for detection and output of the light reflection evaluation value.
[0025] Preferably, the shrinking speed of the key points in the main key point division module is as follows:
[0026]
[0027] Among them, q 1 is the initial frame, q 2 is the number of frames corresponding to the key point reaching the maximum pupil diameter, Δm max is the maximum pupil diameter change of the key point.
[0028] Preferably, the skin color assessment module specifically includes:
[0029] A skin color sample database establishment module is used to obtain historical skin color data and establish a skin color sample database;
[0030] Model training module, which trains the CNN model through historical skin color data to obtain a skin color assessment model;
[0031] The skin color evaluation module inputs the skin color data corresponding to the newborn to be tested into the skin color evaluation model to obtain a skin color evaluation value.
[0032] Preferably, the solution recommendation module includes:
[0033] A feature vector generation module combines the processed data to obtain a feature vector of the newborn to be resuscitated;
[0034] The similarity comparison module calculates multiple similarities based on the feature vector to be resuscitated and the feature vectors corresponding to the multiple historical resuscitation data, wherein the multiple similarities represent the similarity between the feature vector to be resuscitated and the feature vectors corresponding to the multiple historical resuscitation data;
[0035] The recommendation module selects similar feature vectors from the feature vectors corresponding to the historical resuscitation data according to the multiple similarities and recommends corresponding solutions.
[0036] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a neonatal resuscitation assistance system based on life parameter collection and evaluation. By collecting multiple life parameters such as skin color, respiratory rate, heart rate, etc. through video, the health status of the newborn can be comprehensively evaluated, and more accurate data support can be provided for resuscitation. By using the historical processing data in the database and comparing the similarity, the system can quickly recommend the most suitable resuscitation plan for the current newborn situation, thereby improving the resuscitation efficiency. At the same time, by processing the image data through a multi-layer convolutional neural network, the muscle tension characteristics of the newborn can be accurately extracted, thereby improving the accuracy of muscle tension evaluation. By analyzing the pupil's response to light stimulation, reliable physiological parameters can be obtained, which can assist in judging the nervous system condition of the newborn. A skin color sample library is established, and skin color evaluation is performed using a deep learning model, which can provide important physiological indicators for resuscitation and help judge the oxygenation of the newborn. By generating feature vectors and calculating the similarity of multiple resuscitation features, the plan recommendation is more personalized, and a rapid response based on historical data is achieved, thereby reducing the workload of medical staff. The entire system is based on big data and machine learning technology, providing a scientific basis, reducing the errors caused by clinical experience, and improving the success rate of resuscitation. It should be specifically noted that the plan recommendation for similar cases generated by the present invention is only used as a reference for medical staff during the resuscitation process. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0038] Figure 1 A schematic diagram of the structure provided by the present invention;
[0039] Figure 2 This is a structural diagram of the muscle tension assessment module provided by the present invention. DETAILED DESCRIPTION
[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] The embodiment of the present invention discloses a neonatal resuscitation assistance system based on life parameter collection and evaluation, such as Figure 1 As shown, it includes: a main control module, a data acquisition module, a simulation prediction module, a database, and a solution recommendation module. The data acquisition module collects videos of newborns to obtain multi-dimensional parameter information, and the multi-dimensional parameter information includes at least three items of skin color data, amniotic fluid data, muscle tension data, respiratory rate data, production data, pupil data and heart rate data; the data acquisition module sends the multi-dimensional parameter information to the simulation prediction module through the main control module for data processing, and sends the processed data to the solution recommendation module through the main control module; the solution recommendation module compares the processed data with the corresponding solution according to the historical processing data pre-stored in the database through a preset similarity threshold, and selects the solution corresponding to the historical processing data that is greater than or equal to the similarity threshold for solution recommendation.
[0042] In a specific embodiment, the rating criteria for amniotic fluid data processing in the simulation prediction module are: according to the degree of contamination, the amniotic fluid meconium stain can be divided into three degrees, degree I is light green, common fetal chronic hypoxia; degree II is yellow-green and turbid, indicating acute fetal hypoxia; degree III is thick amniotic fluid and brown-yellow in texture, indicating severe fetal hypoxia.
[0043] In a specific embodiment, the simulation prediction module includes a muscle tension evaluation module, a skin color evaluation module, and a pupil evaluation module, which respectively evaluate the muscle tension data, skin color data, and pupil data to obtain muscle tension evaluation values, skin color evaluation values, and pupil light reflection evaluation values.
[0044] In a specific embodiment, Figure 2 As shown in the figure, the muscle tension assessment module specifically includes:
[0045] The key frame extraction module collects the video of the contraction amplitude of the lower limbs of newborns during non-invasive electrical stimulation of the foot, and extracts multiple key frame data according to a predetermined sampling frequency;
[0046] A feature tensor extraction module obtains multiple action feature matrices by passing multiple key frame data through a feature extractor, and arranges the multiple action feature matrices into a three-dimensional feature tensor along the channel dimension;
[0047] Among them, the feature extractor is a convolutional neural network model, and each layer of the convolutional neural network model is used to perform the following operations in the forward pass of the layer: convolution processing is performed on the input data to obtain a convolution feature map;
[0048] The convolution feature map is mean pooled based on the local feature matrix to obtain a pooled feature map; and the pooled feature map is nonlinearly activated to obtain an activated feature map; the output of the last layer of the first convolution neural network model is the action feature matrix, and the input of the first layer of the first convolution neural network model is the key frame data.
[0049] The associated feature extraction module obtains a first action feature map and a second action feature map respectively by passing the three-dimensional feature tensor through a first-scale convolutional neural network and a second-scale convolutional neural network;
[0050] Among them, a first-scale convolutional neural network with a three-dimensional convolution kernel of a first scale is used to perform the following on the input data in the forward pass of the layer: performing three-dimensional convolution processing on the input data based on the three-dimensional convolution kernel with the first scale to obtain a convolution feature map; performing mean pooling processing on the convolution feature map based on a local feature matrix to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first-scale convolutional neural network is a first action feature map, and the input of the first layer is a three-dimensional feature tensor.
[0051] Among them, the second-scale convolutional neural network using the three-dimensional convolution kernel with the second scale performs the following on the input data in the forward pass of the layer: based on the three-dimensional convolution kernel with the second scale, three-dimensional convolution processing is performed on the input data to obtain a convolution feature map; the convolution feature map is subjected to mean pooling processing based on the local feature matrix to obtain a pooled feature map; the pooled feature map is subjected to nonlinear activation to obtain an activation feature map; the output of the last layer of the second-scale convolutional neural network is the second action feature map, and the input of the first layer is the three-dimensional feature tensor.
[0052] The evaluation result generation module fuses the first motion feature map and the second motion feature map to obtain a classification feature map, and passes the classification feature map through a classifier to obtain a muscle tension evaluation value of the newborn.
[0053] In a specific embodiment, obtaining the classification feature graph in the evaluation result generation module specifically includes:
[0054]
[0055] Among them, F3 is the classification feature map, F 1 is the first action feature map, F 2 is the second action characteristic graph, It represents the addition of the elements at corresponding positions of the first action feature map and the second action feature map, and α and β are weighting parameters used to control the balance between the first action feature map and the second action feature map in the classification feature map.
[0056] In a specific embodiment, a feature distribution correction module is further included, which is used to perform feature distribution correction on the classification feature map to obtain a corrected classification feature map, specifically including:
[0057]
[0058] Among them, f i is the feature value of the i-th position of the classification feature map, f j is the feature value of the jth position of the classification feature map, is the mean of all eigenvalues of the classification feature map, N is the scale of the classification feature map, exp(·) represents the exponential operation of the value, and f i ′ is the feature value of each position of the corrected classification feature map.
[0059] In a specific embodiment, the evaluation result generation module includes:
[0060] An expansion unit, which expands each corrected classification feature matrix in the corrected classification feature graph into a one-dimensional feature vector according to a row vector or a column vector and then cascades them to obtain a classification feature vector;
[0061] A fully connected encoding unit is used to use the fully connected layer of the classifier to fully connect the classification feature vector to obtain an encoded classification feature vector; and a classification result generating unit is used to input the encoded classification feature vector into the Softmax classification function of the classifier to obtain a muscle tension evaluation value.
[0062] In a specific embodiment, the muscle tension assessment value may be obtained by referring to the Ashworth Scale or the Modified Ashworth Scale.
[0063] In a specific embodiment, the pupil assessment module specifically includes:
[0064] The main key point classification module obtains the key points of the newborn's pupil, obtains the relevance score of the contraction speed and the earliest contraction time of each key point based on the video of the newborn's pupil stimulated by light, sets the relevance score threshold, and classifies the key point as a main key point if the relevance score of the key point is greater than or equal to the relevance score threshold;
[0065] A related point determination module, which obtains key points whose shrinkage amplitude exceeds a preset shrinkage threshold among the main key points as related points, constructs a set of related points, and assigns weights to the related points;
[0066] Among them, each relevant point is sorted according to the size of the contraction amplitude to obtain the sequence number of each relevant point, and the sequence number is arranged as [1, p]. For each sequence number of the relevant point, the weight is allocated, and the formula is as follows:
[0067]
[0068] Among them, c is the serial number of the current relevant point, A c is the weight value of the current relevant point, It is the sum of all relevant point numbers.
[0069] The key frame screening module calculates the change in pupil diameter of the relevant points corresponding to all two adjacent frames in the video, selects the weight of the corresponding relevant point when the change is less than the change threshold, counts the total weight of the relevant points on each frame image in the video, and selects the key frame according to the total weight of the relevant points of each frame image and the set total weight threshold;
[0070] The image detection module obtains the main area of the key frame under different lighting stimulations and sends it to the image classification network for detection and output of the light reflection evaluation value.
[0071] Among them, according to the obtained video of stimulating the pupil of the newborn through light, the frame corresponding to the first movement of each key point is taken as the initial frame, and the earliest reaction time is determined according to the initial frame; the maximum pupil diameter change of each key point and the frame number corresponding to the maximum pupil diameter change are recorded, and the contraction speed is determined according to the maximum pupil diameter change of the key point.
[0072] In a specific embodiment, the shrinking speed of the key points in the main key point division module is as follows:
[0073]
[0074] Among them, q 1 is the initial frame, q 2 is the number of frames corresponding to the key point reaching the maximum pupil diameter, Δm max is the maximum pupil diameter change of the key point.
[0075] In a specific embodiment, the skin color assessment module specifically includes:
[0076] A skin color sample database establishment module is used to obtain historical skin color data and establish a skin color sample database;
[0077] Model training module, which trains the CNN model through historical skin color data to obtain a skin color assessment model;
[0078] The skin color assessment module inputs the skin color data corresponding to the newborn to be tested into the skin color assessment model to obtain the skin color assessment value.
[0079] The CNN model in skin color assessment specifically includes: after the output of the Xception network, the channel information of the feature map is aggregated using global maximum pooling and global average pooling to generate two different channel features, which are then sent to a multi-layer perceptron, whose outputs are added element by element, and finally activated by the sigmoid function to generate a channel attention map to obtain a channel attention module; after the feature output of the channel attention module is input, the global maximum pooling and global average pooling operations are first applied on the channel axis to obtain two two-dimensional spatial features, which are then spliced into a feature map, and then the dimensionality is reduced by a 7×7 convolution kernel, and finally the sigmoid function is used to generate a channel attention map. Activation produces a channel attention map to obtain a spatial attention module; the output is connected to a global average pooling layer, and then a gender input module is created, with binary production data as input. The binary production data is "0" or "1", premature birth is "1", and normal birth is "0"; the output image feature information and gender feature information are connected in series through a densely connected layer activated by ReLU with 32 neurons, and connected to two densely connected layers, each of which is fed by a ReLU activation layer with 1024 neurons densely connected and a dropout layer. The last layer consists of a densely connected layer with linear activation of 1 neuron to obtain the skin color evaluation value.
[0080] In a specific embodiment, the solution recommendation module includes:
[0081] A feature vector generation module combines the processed data to obtain a feature vector of the newborn to be resuscitated;
[0082] The similarity comparison module calculates multiple similarities based on the feature vector to be resuscitated and the feature vectors corresponding to multiple historical resuscitation data, and the multiple similarities represent the degree of similarity between the feature vector to be resuscitated and the feature vectors corresponding to multiple historical resuscitation data; it should be noted that when the eigenvalues corresponding to the extracted historical resuscitation data are combined into a feature vector, the combination order of the eigenvalues is the same as the combination order of the eigenvalues corresponding to the above-mentioned feature vector to be resuscitated.
[0083] The recommendation module selects similar feature vectors from the feature vectors corresponding to the historical recovery data according to multiple similarities and recommends corresponding solutions.
[0084] The similarity calculation between the feature vector to be resuscitated and the feature vector corresponding to the historical resuscitation data can be calculated using algorithms such as cosine similarity and term frequency-inverse document frequency (TF-IDF). The following takes the cosine similarity algorithm as an example:
[0085] The principle of the cosine similarity algorithm is to treat two eigenvectors as two line segments in space. Both line segments start from the origin and point in different directions, that is, an angle is formed between the two line segments. The smaller the angle, the closer the directions of the two eigenvectors are, that is, the more similar the two vectors are.
[0086] Citing the above example, assuming that the feature vector to be resuscitated is A = [1, 3, 0], and the feature vector corresponding to a certain historical resuscitation data is B = [1, 3, 1]. First, the formula for calculating the cosine value of the angle between the two feature vectors is as follows:
[0087]
[0088] In a specific embodiment, similar solutions are first recommended based on the processed heart rate data, respiratory rate data, and skin color evaluation value. After the processing is completed, a secondary solution recommendation is performed based on the heart rate data collected again at a fixed time sequence.
[0089] In a specific embodiment, a secondary recommendation module is also included. When the scheme recommendation module recommends a scheme, the secondary recommendation module synchronously receives the recommended scheme, collects the heart rate data of the newborn according to a fixed time sequence (1 minute, 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes), and makes a secondary scheme recommendation based on the heart rate data and the relevant schemes in the database. For example: when the heart rate is <100 times / min, check the chest movement, correct the ventilation steps when necessary, and intubate or use a laryngeal mask airway when necessary; when the heart rate is <60 times / min, intubate, coordinate chest compression with positive pressure ventilation, 100% oxygen supply, use 3-lead ECG monitoring, and consider emergency umbilical vein catheterization; if the heart rate is still <60 times / min after re-testing, inject epinephrine intravenously, consider hypovolemia, and consider pneumothorax.
[0090] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0091] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A neonatal resuscitation assistance system based on vital parameter collection and evaluation, characterized in that: include: A main control module, a data acquisition module, a simulation prediction module, a database, and a solution recommendation module. The data acquisition module performs video acquisition on the newborn to obtain multi-dimensional parameter information, and the multi-dimensional parameter information includes at least three items of skin color data, amniotic fluid data, muscle tension data, respiratory rate data, production data, pupil data, and heart rate data; the data acquisition module sends the multi-dimensional parameter information to the simulation prediction module through the main control module for data processing, and sends the processed data to the solution recommendation module through the main control module; the solution recommendation module compares the processed data with the corresponding solution according to the historical processing data pre-stored in the database through a preset similarity threshold, and selects the solution corresponding to the historical processing data greater than or equal to the similarity threshold for solution recommendation.
2. A neonatal resuscitation assistance system based on vital parameter collection and evaluation according to claim 1, characterized in that: The simulation prediction module includes a muscle tension evaluation module, a skin color evaluation module, and a pupil evaluation module, which respectively evaluate the muscle tension data, skin color data, and pupil data to obtain a muscle tension evaluation value, a skin color evaluation value, and a pupil light reflection evaluation value.
3. A neonatal resuscitation assistance system based on vital parameter collection and evaluation according to claim 2, characterized in that: The muscle tension assessment module specifically includes: The key frame extraction module collects the video of the contraction amplitude of the lower limbs of newborns during non-invasive electrical stimulation of the foot, and extracts multiple key frame data according to a predetermined sampling frequency; A feature tensor extraction module, which obtains a plurality of action feature matrices by respectively passing the plurality of key frame data through a feature extractor, and arranges the plurality of action feature matrices into a three-dimensional feature tensor along a channel dimension; An associated feature extraction module, which obtains a first action feature map and a second action feature map respectively by passing the three-dimensional feature tensor through a first-scale convolutional neural network and a second-scale convolutional neural network; An evaluation result generation module fuses the first motion feature map and the second motion feature map to obtain a classification feature map, and passes the classification feature map through a classifier to obtain a muscle tension evaluation value of the newborn.
4. A neonatal resuscitation assistance system based on vital parameter collection and evaluation according to claim 3, characterized in that: The classification feature graph obtained in the evaluation result generation module specifically includes: Among them, F3 is the classification feature map, F1 is the first action feature map, and F2 is the second action feature map. It represents the addition of the elements at corresponding positions of the first action feature map and the second action feature map, and α and β are weighting parameters used to control the balance between the first action feature map and the second action feature map in the classification feature map.
5. A neonatal resuscitation assistance system based on vital parameter collection and evaluation according to claim 3, characterized in that: It also includes a feature distribution correction module, which is used to perform feature distribution correction on the classification feature map to obtain a corrected classification feature map, specifically including: Among them, f i is the feature value of the i-th position of the classification feature map, f j is the feature value of the jth position of the classification feature map, is the mean of all eigenvalues of the classification feature map, N is the scale of the classification feature map, exp(·) represents the exponential operation of the value, and f i ′ is the feature value of each position of the corrected classification feature map.
6. A neonatal resuscitation assistance system based on vital parameter collection and evaluation according to claim 2, characterized in that: The pupil assessment module specifically includes: The main key point classification module obtains the key points of the neonatal pupil, obtains the relevance score for the contraction speed and the earliest contraction time of each key point based on the video of the neonatal pupil stimulated by light, sets the relevance score threshold, and classifies the key point as a main key point if the relevance score of the key point is greater than or equal to the relevance score threshold; A related point determination module, which obtains key points whose shrinkage amplitude exceeds a preset shrinkage threshold among the main key points as related points, constructs a related point set, and assigns weights to the related points; A key frame screening module calculates the change in pupil diameter of the relevant points corresponding to all two adjacent frames in the video, selects the weight of the corresponding relevant point when the change is less than a change threshold, counts the total weight of the relevant points on each frame image in the video, and selects a key frame according to the total weight of the relevant points on each frame image and a set total weight threshold; The image detection module obtains the main area of the key frame under different lighting stimulations and sends it to the image classification network for detection and output of the light reflection evaluation value.
7. A neonatal resuscitation assistance system based on vital parameter collection and evaluation according to claim 6, characterized in that: The shrinking speed of the key points in the main key point division module is as follows: Among them, q1 is the initial frame, q2 is the frame number corresponding to the key point reaching the maximum pupil diameter, Δm max is the maximum pupil diameter change of the key point.
8. A neonatal resuscitation assistance system based on vital parameter collection and evaluation according to claim 2, characterized in that: The skin color assessment module specifically includes: A skin color sample database establishment module is used to obtain historical skin color data and establish a skin color sample database; Model training module, which trains the CNN model through historical skin color data to obtain a skin color assessment model; The skin color evaluation module inputs the skin color data corresponding to the newborn to be tested into the skin color evaluation model to obtain a skin color evaluation value.
9. A neonatal resuscitation assistance system based on vital parameter collection and evaluation according to claim 2, characterized in that: The solution recommendation module includes: A feature vector generation module combines the processed data to obtain a feature vector of the newborn to be resuscitated; The similarity comparison module calculates multiple similarities based on the feature vector to be resuscitated and the feature vectors corresponding to the multiple historical resuscitation data, wherein the multiple similarities represent the similarity between the feature vector to be resuscitated and the feature vectors corresponding to the multiple historical resuscitation data; The recommendation module selects similar feature vectors from the feature vectors corresponding to the historical resuscitation data according to the multiple similarities and recommends corresponding solutions.
Citation Information
Patent Citations
Method and system for personalized guideline-based therapy augmented by imaging information
CN102016859A
Motion detection facial paralysis degree evaluation system based on artificial intelligence
CN112597842A
Deep learning-based children bone age evaluation method and system
CN113298780A
Facial color classification method and device, computer readable program medium and electronic equipment
CN115100723A
Data processing system and method for predicting response ability of old people based on games
CN115862151A
