Pupil image recognition and illness change evaluation system for image recognition

Abstract

The invention discloses a pupil image recognition and disease change evaluation system based on image recognition, and relates to the technical field of monitoring analysis, and the system comprises the steps: obtaining the identity information and historical pupil data of a patient, and building a pupil dynamic database corresponding to the patient; alternately irradiating the eyes of the patient, and collecting a dynamic image sequence corresponding to a pupil light reflex area of the patient; performing spatial-temporal characteristic analysis on the dynamic image sequence, establishing a pupil diameter dynamic change curve, and generating a light reflection sensitivity index of the pupil of the patient according to the pupil diameter dynamic change curve; constructing a pupil dynamic evaluation model according to the pupil diameter dynamic change curve and the light reflex sensitivity index of the patient pupil, and outputting a pupil grade evaluation result based on the pupil dynamic evaluation model; comprehensive analysis is conducted on pupil grade evaluation results, analysis results are uploaded to an intensive care central system, and a grading early warning mechanism is triggered. The application has the effect of reducing the occurrence of illness state delay.

Description

Technical Field

[0001] This application relates to the technical field of monitoring and analysis, and particularly to a pupil image recognition and disease condition change assessment system for image recognition. Background Art

[0002] In the field of medical diagnosis, the timely and accurate assessment of the changes in a patient's condition is crucial. As an important physiological feature of the human body, the changes in the pupil can reflect the physical conditions of the patient's nervous system, cardiovascular system, etc. to a certain extent.

[0003] In the related art, traditional pupil observation mainly relies on medical staff to manually observe. By observing the characteristics such as the size, shape, and light reflex of the pupil with the naked eye to judge the condition. However, the above methods have many limitations. On the one hand, the subjectivity of manual observation is relatively strong, and the judgment criteria of different medical staff may vary, resulting in inaccurate and unreliable results. On the other hand, it is difficult to achieve real-time and dynamic monitoring of pupil changes. For some patients with rapidly changing conditions, the subtle changes in the pupil may not be captured in time, thus delaying the diagnosis and treatment of the condition, and there is room for improvement. Summary of the Invention

[0004] In view of the deficiencies of the prior art, this application provides a pupil image recognition and disease condition change assessment system for image recognition.

[0005] In a first aspect, this application provides a method for pupil image recognition and disease condition change assessment for image recognition, including the following steps: Step S1: Scan the QR code on the patient's wristband through a medical terminal device to obtain the patient's identity information and historical pupil data, and establish a pupil dynamic database corresponding to the patient; Step S2: Use a multi-spectral pupil imaging device to alternately irradiate the patient's both eyes, and synchronously collect a dynamic image sequence corresponding to the pupil light reflex area of the patient; Step S3: Analyze the spatio-temporal features of the dynamic image sequence, establish a dynamic change curve of the pupil diameter, and generate a pupil light reflex sensitivity index for the patient according to the dynamic change curve of the pupil diameter; Step S4: Construct a pupil dynamic assessment model according to the dynamic change curve of the pupil diameter and the pupil light reflex sensitivity index of the patient, and then output a pupil grade assessment result based on the pupil dynamic assessment model; Step S5: Comprehensively analyze the pupil grade assessment result and upload the analysis result to the intensive care central system to trigger a grading warning mechanism.

[0006] Preferably, step S3 includes the following steps: Step S31: Perform multi-frame pupil area segmentation and temporal alignment processing on the dynamic image sequence corresponding to the patient's pupillary light reflex area, and extract the spatio-temporal feature data of the pupil contour. Step S32: Divide the spatio-temporal feature data into pupil contraction-dilation phases, and construct a dynamic change curve of the pupil diameter. Step S33: Track the iris texture displacement vector by the optical flow method, and then calculate the contraction rate of the pupillary sphincter based on the iris texture displacement vector. Step S34: Generate a pupillary light reflex sensitivity index according to the dynamic change curve of the pupil diameter and the contraction rate of the pupillary sphincter.

[0007] Preferably, step S4 includes the following steps: Step S41: Input the dynamic change curve of the pupil diameter into a pre-trained deep residual network to extract multi-scale spatio-temporal feature data. Step S42: Feature-weight the iris texture displacement vector based on the multi-scale spatio-temporal feature data and the attention mechanism, and then generate a sphincter movement feature vector. Step S43: Concatenate the pupillary light reflex sensitivity index and the sphincter movement feature vector, input them into a bidirectional LSTM network for temporal modeling, and then construct a pupil dynamic evaluation model, and then output a pupil grade evaluation result based on the pupil dynamic evaluation model.

[0008] Preferably, step S5 includes the following steps: Step S51: Dynamically compare the pupil grade evaluation result with the corresponding historical pupil data in the patient's pupil dynamic database, and calculate the pupil parameter change gradient data. Step S52: Construct a brain function injury probability prediction model according to the pupil parameter change gradient data, and then generate a disease deterioration risk coefficient. Step S53: Upload the disease deterioration risk coefficient and the pupil grade evaluation result to the intensive care central system through blockchain encryption technology, and trigger a grading warning mechanism.

[0009] Preferably, step S52 includes the following steps: Step S521: Perform dynamic temporal analysis and feature extraction on the pupil parameter change gradient data to obtain pupil response feature data. Step S522: Based on the pupil response feature data, simulate the brain function degenerative pathological mechanism through a neurophysiological injury association model to generate brain function injury mechanism mapping data. Step S523: Integrate the brain function injury mechanism mapping data and multi-modal brain imaging feature data to construct a brain function injury probability prediction model, and output brain function injury probability distribution data. Step S524: Generate a disease deterioration risk coefficient based on the brain function injury probability distribution data and in combination with the clinical deterioration path quantification algorithm.

[0010] Preferably, trigger a grading early warning mechanism, specifically including: Obtain the disease deterioration risk coefficient corresponding to the patient, and compare the disease deterioration risk coefficient corresponding to the patient with a preset risk threshold; If the disease deterioration risk coefficient corresponding to the patient is between the first risk threshold and the second risk threshold, trigger a primary warning signal, trigger a yellow flashing identifier on the electronic medical record interface of the intensive care central system, automatically generate a pupil grade assessment result report and push it to the nurse's mobile terminal; If the disease deterioration risk coefficient corresponding to the patient is between the second risk threshold and the third risk threshold, trigger an intermediate warning signal, and simultaneously activate the sound and light signals of the bedside alarm in the patient's ward, and send a warning message to the attending physician; If the disease deterioration risk coefficient corresponding to the patient exceeds the third risk threshold, trigger a high-level warning signal, start the hospital-wide emergency broadcast system, automatically retrieve the electronic signature of the preoperative informed consent form to verify the identity, and link to reserve an emergency scanning channel in the CT room. At the same time, transmit the pupil grade assessment result report and the corresponding pupil dynamic database of the patient to the remote consultation center in real time through the 5G private network. When no medical confirmation feedback is received for 3 consecutive minutes, automatically trigger the standby emergency response protocol.

[0011] In a second aspect, the present application provides a pupil image recognition and disease condition change assessment system for image recognition, including: A data recognition module, configured to scan the QR code on the patient's wristband through a medical terminal device, obtain the patient's identity information and historical pupil data, and establish a pupil dynamic database corresponding to the patient; A data acquisition module, configured to alternately irradiate the patient's both eyes with a multi-spectral pupil imaging device, and synchronously acquire a dynamic image sequence corresponding to the pupil light reflection area of the patient; An analysis module, configured to perform spatio-temporal feature analysis on the dynamic image sequence, establish a dynamic change curve of the pupil diameter, and generate a pupil light reflex sensitivity index of the patient according to the dynamic change curve of the pupil diameter; An evaluation module, configured to construct a pupil dynamic evaluation model according to the dynamic change curve of the pupil diameter and the pupil light reflex sensitivity index of the patient, and then output a pupil grade evaluation result based on the pupil dynamic evaluation model; An early warning module, configured to comprehensively analyze the pupil grade evaluation result, and upload the analysis result to the intensive care central system to trigger a grading early warning mechanism.

[0012] In a third aspect, the present application provides a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to execute a method for pupil image recognition and disease condition change assessment in image recognition as described in any one of the above.

[0013] In summary, the present application includes the following beneficial technical effects: The present application provides a method for pupil image recognition and disease condition change assessment in image recognition. By collecting and analyzing a dynamic image sequence corresponding to the pupil light reflection area of a patient, a dynamic change curve of the pupil diameter is established, and a pupil light reflex sensitivity index of the patient is generated based on the dynamic change curve of the pupil diameter. A pupil dynamic assessment model is constructed based on the dynamic change curve of the pupil diameter and the pupil light reflex sensitivity index of the patient. Then, a pupil grade assessment result is output based on the pupil dynamic assessment model, the pupil grade assessment result is comprehensively analyzed, and the analysis result is uploaded to the intensive care central system to trigger a grading warning mechanism. This reduces the occurrence of inaccurate results caused by the strong subjectivity of manual observation, and real-time and dynamic monitoring of the pupil changes of the patient is carried out to capture the subtle changes of the patient's pupils in a timely manner, thereby reducing the occurrence of delayed disease diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0015] Figure 1 is a flowchart of a method for pupil image recognition and disease condition change assessment in image recognition according to an embodiment of the present application.

[0016] Figure 2 is a schematic diagram of a system for pupil image recognition and disease condition change assessment in image recognition according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The following will further elaborate on the present application Figure 1-2 in further detail.

[0018] Embodiment 1 The embodiment of the present application discloses a method for pupil image recognition and disease condition change assessment in image recognition.

[0019] Refer to Figure 1 , a method for pupil image recognition and disease condition change assessment in image recognition, includes the following steps: Step S1: Scan the QR code on the patient's wristband through the medical terminal device to obtain the patient's identity information and historical pupil data, and establish a corresponding pupil dynamic database for the patient; Step S2: Use a multi-spectral pupil imaging device to alternately irradiate the patient's both eyes, and synchronously collect a dynamic image sequence corresponding to the pupil light reflex area of the patient; Step S3: Analyze the spatio-temporal features of the dynamic image sequence, establish a dynamic change curve of the pupil diameter, and generate a pupil light reflex sensitivity index for the patient according to the dynamic change curve of the pupil diameter; Step S4: Construct a pupil dynamic evaluation model based on the dynamic change curve of the pupil diameter and the pupil light reflex sensitivity index of the patient, and then output a pupil grade evaluation result based on the pupil dynamic evaluation model; Step S5: Comprehensively analyze the pupil grade evaluation result, and upload the analysis result to the intensive care central system to trigger a grading warning mechanism.

[0020] It should be noted that Step S3 includes the following steps: Step S31: Perform multi-frame pupil area segmentation and temporal alignment processing on the dynamic image sequence corresponding to the patient's pupil light reflex area, and extract the spatio-temporal feature data of the pupil contour; Step S32: Divide the spatio-temporal feature data into pupil contraction-dilation phases, and construct a dynamic change curve of the pupil diameter; Step S33: Trace the iris texture displacement vector by the optical flow method, and then calculate the contraction rate of the pupil sphincter based on the iris texture displacement vector; Step S34: Generate a pupil light reflex sensitivity index according to the dynamic change curve of the pupil diameter and the contraction rate of the pupil sphincter.

[0021] Step S31: Perform multi-frame pupil area segmentation and temporal alignment processing on the dynamic image sequence corresponding to the patient's pupillary light reflex area, and extract the spatio-temporal feature data of the pupil contour. In actual operation, use a high-resolution image acquisition device, such as a specialized ophthalmic camera, to continuously capture the pupillary light reflex area of the patient during light stimulation, obtaining a series of dynamic images. Exemplarily, when giving the patient a certain intensity of flash stimulation, the camera captures at a speed of 30 frames per second, recording the changes of the pupil during the light stimulation process. Apply image segmentation algorithms, such as threshold segmentation, edge detection, or deep learning-based segmentation methods, to accurately segment the pupil area in each frame of the image and perform temporal alignment processing. Due to factors such as slight time deviation or device jitter in the acquisition of different frames of images, calibrate the images on the time axis through the algorithm to ensure that it can accurately reflect the change process of the pupil over time. After completing the above processing, extract the spatio-temporal feature data of the pupil contour, including information such as the shape, size, and position of the pupil at different times. Exemplarily, record the edge coordinates of the pupil contour in each frame of the image, as well as the position change of the pupil center, etc.; Step S32: Divide the spatio-temporal feature data into pupil contraction-dilation phases and construct a dynamic change curve of the pupil diameter. According to the spatio-temporal feature data of the pupil contour extracted in Step S31, analyze the contraction and dilation processes of the pupil under light stimulation, and divide the above processes into different phases, such as the initial light stimulation response phase, the rapid pupil contraction phase, the stable phase after contraction, and the subsequent pupil dilation phase, etc. By analyzing the pupil contour in each phase, calculate the diameter size of the pupil at different times. For example, in a certain frame of the image, according to the edge coordinates of the pupil contour, use geometric calculation methods to obtain that the diameter of the pupil is 4 millimeters. Connect the above calculated pupil diameter values in chronological order to construct a dynamic change curve of the pupil diameter. The dynamic change curve of the pupil diameter intuitively shows the change of the pupil diameter over time during the light stimulation process; Step S33: Track the iris texture displacement vector by the optical flow method, and then calculate the contraction rate of the pupillary sphincter based on the iris texture displacement vector. The optical flow method is an algorithm for analyzing the motion of objects in images, and it is used to track the displacement of iris texture between different frame images. Since the contraction and dilation of the pupil are caused by the motion of the pupillary sphincter, and the displacement of the iris texture is closely related to the motion of the pupillary sphincter. For example, when the pupillary sphincter contracts, the iris moves towards the center, resulting in a change in the position of the iris texture. Calculate the displacement vector of the iris texture between adjacent frame images through the optical flow algorithm. The vector contains information about the direction and magnitude of the displacement. According to the displacement vector and the time interval, calculate the contraction rate of the pupillary sphincter. Assuming that the time interval between two adjacent frame images is 0.03 seconds (taking the acquisition speed of 30 frames per second as an example), and the displacement of the iris texture in a certain direction is 0.5 millimeters, then the contraction rate of the pupillary sphincter in this direction can be calculated to be approximately 16.7 millimeters per second; Step S34: Generate the pupillary light reflex sensitivity index based on the dynamic change curve of the pupil diameter and the contraction rate of the pupillary sphincter. Comprehensively consider the dynamic change curve of the pupil diameter obtained in step S32 and the contraction rate of the pupillary sphincter calculated in step S33, and use a mathematical model or algorithm to generate the pupillary light reflex sensitivity index. Exemplarily, set a formula that incorporates factors such as the amplitude of the change in the pupil diameter, the speed of the change, and the contraction rate of the pupillary sphincter. Assume that the amplitude of the change in the pupil diameter before and after light stimulation is 2 millimeters, the average speed of the change in the pupil diameter is 0.5 millimeters per second, and the contraction rate of the pupillary sphincter is 15 millimeters per second. Calculate the pupillary light reflex sensitivity index as a specific value through the formula. The above value can reflect the sensitivity of the patient's pupillary light reflex. The larger the value, the more sensitive the pupillary light reflex.

[0022] It should be noted that step S4 includes the following steps: Step S41: Input the dynamic change curve of the pupil diameter into a pre-trained deep residual network to extract multi-scale spatio-temporal feature data; Step S42: Perform feature weighting on the iris texture displacement vector based on the multi-scale spatio-temporal feature data and the attention mechanism, and then generate a sphincter motion feature vector; Step S43: Concatenate the pupillary light reflex sensitivity index and the sphincter motion feature vector, input them into a bidirectional LSTM network for temporal modeling, and then construct a pupil dynamic evaluation model. Then, based on the pupil dynamic evaluation model, output the pupil grade evaluation result.

[0023] Specifically, in step S41: The dynamic change curve of pupil diameter is input into a pre-trained deep residual network to extract multi-scale spatio-temporal feature data. The deep residual network is a powerful deep learning model. Because it can effectively handle the vanishing gradient problem in deep networks, it is widely used in feature extraction tasks. In the above step, the dynamic change curve of pupil diameter is obtained and input into the deep residual network that has been pre-trained on a large amount of relevant data. The deep residual network processes the input curve through operations such as multi-layer convolution and pooling. In different network layers, spatio-temporal features of different scales will be extracted. Exemplarily, shallower networks may capture local detail features of pupil diameter changes, such as minute fluctuations in a short period of time; while deeper networks may extract more macroscopic spatio-temporal features, such as the overall change trend of pupil diameter during the entire light stimulation process. Through the above operations, multi-scale spatio-temporal feature data is output from the deep residual network. The multi-scale spatio-temporal feature data contains rich information about pupil dynamic changes; Step S42: Feature-weight the iris texture displacement vector based on the multi-scale spatio-temporal feature data and the attention mechanism, and then generate a sphincter movement feature vector. After obtaining the multi-scale spatio-temporal feature data, further analysis is carried out in combination with the iris texture displacement vector. The iris texture displacement vector is obtained by tracking the displacement of iris texture between different frame images through the optical flow method. It reflects the influence of pupil sphincter movement on iris texture. The attention mechanism is a method that enables the model to focus on important information. In the above step, using the multi-scale spatio-temporal feature data as a guide, the attention mechanism analyzes which parts of the iris texture displacement vector are more relevant to the dynamic changes of the pupil, that is, which displacement vectors are more important for describing the movement of the pupil sphincter. For example, if the multi-scale spatio-temporal feature data shows a rapid contraction change in pupil diameter during a certain period, then the attention mechanism will pay more attention to the corresponding iris texture displacement vector during the above period. According to the analysis results, the iris texture displacement vector is weighted, giving higher weights to important displacement vectors and lower weights to unimportant displacement vectors. The weighted iris texture displacement vectors are integrated to generate a sphincter movement feature vector. The sphincter movement feature vector combines multi-scale spatio-temporal features and iris texture displacement information and can more accurately describe the movement characteristics of the pupil sphincter; Step S43: Concatenate the pupillary light reflex sensitivity index with the sphincter movement feature vector, and input it into a bidirectional LSTM network for temporal modeling, thereby constructing a pupillary dynamic assessment model. Then, based on the pupillary dynamic assessment model, output the pupillary grade assessment result. The pupillary light reflex sensitivity index is previously obtained through comprehensive calculations of factors such as the dynamic change curve of the pupil diameter and the contraction rate of the pupillary sphincter, which reflects the sensitivity of the pupil to light stimulation. Concatenate the pupillary light reflex sensitivity index with the sphincter movement feature vector generated in step S42 to form a feature vector containing more dimensional information. The bidirectional LSTM network is a recurrent neural network suitable for processing time series data, which can consider both the forward and backward information of the time series. Input the concatenated feature vector into the bidirectional LSTM network, and the network will perform temporal modeling on the dynamic change process of the pupil. By learning a large amount of pupillary dynamic data, the bidirectional LSTM network can capture the dependence relationship and change rules between different time points of the pupil. Exemplarily, it can learn how the change of the pupillary light reflex sensitivity index affects the movement of the pupillary sphincter, and how the interaction reflects the overall dynamic condition of the pupil. Based on the trained bidirectional LSTM network, construct a pupillary dynamic assessment model. When new pupil-related feature data is input, the model will evaluate the pupil condition according to the learned rules and output the pupillary grade assessment result. The pupillary grade assessment result can be a classification label, such as "normal", "mild abnormality", "severe abnormality", etc., or a numerical value representing the health degree of the pupil, and the higher the value, the better the pupil condition.

[0024] It should be noted that step S5 includes the following steps: Step S51: Dynamically compare the pupillary grade assessment result with the corresponding historical pupil data in the patient's pupillary dynamic database, and calculate the pupillary parameter change gradient data; Step S52: Construct a brain function damage probability prediction model based on the pupillary parameter change gradient data, and then generate a disease deterioration risk coefficient; Step S53: Upload the disease deterioration risk coefficient and the pupillary grade assessment result to the intensive care central system through blockchain encryption technology, triggering a hierarchical warning mechanism.

[0025] Specifically, in step S51: The pupil grade evaluation result is dynamically compared with the corresponding historical pupil data in the patient's pupil dynamic database to calculate the change gradient data of pupil parameters. First, after the evaluation of the patient's pupils is completed, a pupil grade evaluation result will be obtained. The pupil grade evaluation result can be a classification of the pupil health status (such as normal, mildly abnormal, severely abnormal, etc.) or a specific numerical score. At the same time, the patient's pupil dynamic database stores the historical pupil data of the patient at different time points. The historical pupil data includes various parameter information such as pupil diameter, pupil contraction rate, and pupil light reflex sensitivity index. For example, in previous examinations of a certain patient at different times, the change values of the pupil diameter under different light stimuli and the corresponding pupil light reflex sensitivity index were recorded. The current pupil grade evaluation result is dynamically compared with the historical pupil data. For pupil parameters, such as pupil diameter, the current measured value is compared with the historical measured value, and the change amount within a certain time interval is calculated. Exemplarily, the pupil diameter of this patient was 4 mm under a certain light stimulus during the previous examination, and it is 3.6 mm during the current examination, and the time interval is one week. Then the change amount of the pupil diameter is 4 - 3.6 = 0.4 mm. Then, the change gradient is calculated according to the time interval, that is, 0.4 mm / 7 days ≈ 0.06 mm / day. Similar comparisons and calculations are also performed on other pupil parameters (such as pupil contraction rate, pupil light reflex sensitivity index, etc.), so as to obtain the complete change gradient data of pupil parameters. The change gradient data of pupil parameters reflects the change trend of the patient's pupil parameters over time; In step S52: A brain function injury probability prediction model is constructed based on the change gradient data of pupil parameters, and then a disease deterioration risk coefficient is generated. Based on the change gradient data of pupil parameters obtained in step S51, combined with medical knowledge and clinical experience, the key parameter change gradients related to brain function injury are screened out. For example, research shows that a rapid decrease in the pupil light reflex sensitivity index and a change gradient of pupil diameter exceeding the normal range may be closely related to brain function injury. The above key pupil parameter change gradients are used as input variables, and a mathematical model or machine learning algorithm is used to construct a brain function injury probability prediction model. Exemplarily, the logistic regression algorithm is adopted. By training a large number of case data with known brain function injury conditions, the relationship between the input variables and the probability of brain function injury is determined. And during the training process, the parameters of the model are adjusted to enable the model to accurately predict the probability of brain function injury according to the change gradient of pupil parameters. When new change gradient data of pupil parameters is input, the model will output the corresponding probability of brain function injury. According to the probability of brain function injury and other relevant factors (such as the patient's underlying diseases, current vital signs, etc.), the disease deterioration risk coefficient is comprehensively calculated. The larger the value of the disease deterioration risk coefficient, the higher the risk of disease deterioration. For example, if the probability of brain function injury is high and the patient has other serious underlying diseases, then the disease deterioration risk coefficient will increase accordingly.

[0026] Step S53: Upload the risk coefficient of disease deterioration and the pupil level assessment result to the intensive care central system through blockchain encryption technology, triggering the grading warning mechanism. Blockchain encryption technology has the characteristics of decentralization, immutability, security and reliability. Encrypt the risk coefficient of disease deterioration calculated in step S52 and the pupil level assessment result obtained in step S51 using the blockchain encryption algorithm. The encrypted data is uploaded to the intensive care central system. The intensive care central system is the core platform for hospital intensive care management, storing and managing important information of numerous patients. When the data is uploaded to the intensive care central system, the intensive care central system will trigger the grading warning mechanism according to the size of the risk coefficient of disease deterioration.

[0027] Further, step S52 includes the following steps: Step S521: Conduct dynamic time series analysis and feature extraction on the pupil parameter change gradient data to obtain pupil response feature data; Step S522: Based on the pupil response feature data, simulate the brain function degenerative pathological mechanism through the neurophysiological injury association model to generate brain function injury mechanism mapping data; Step S523: Integrate the brain function injury mechanism mapping data and multi-modal brain image feature data to construct a brain function injury probability prediction model and output brain function injury probability distribution data; Step S524: Generate the risk coefficient of disease deterioration according to the brain function injury probability distribution data and in combination with the clinical deterioration path quantification algorithm.

[0028] Specifically, step S521: Conduct dynamic time series analysis and feature extraction on the pupil parameter change gradient data to obtain pupil response feature data. The pupil parameter change gradient data includes the change conditions of parameters such as pupil diameter, pupil contraction rate, and pupil light reflex sensitivity index at different time points. Conduct dynamic time series analysis on the pupil parameter change gradient data, arrange the change gradients of the above parameters in chronological order, and observe their change trends over time. For example, for the pupil diameter change gradient, analyze whether it gradually increases, decreases, or remains stable over a period of time. Use the feature extraction method in machine learning to extract the key features that can reflect the pupil response characteristics from the above change gradient data to obtain the pupil response feature data. The pupil response feature data includes the response patterns and change rules of the pupil to various stimuli; Step S522: Based on the pupil response characteristic data, simulate the neurodegenerative pathological mechanism through the neurophysiological injury association model to generate brain function injury mechanism mapping data. The neurophysiological injury association model is established based on the principles of neurophysiology and a large number of clinical studies, which describes the potential connection between pupil response and brain function injury. Input the pupil response characteristic data obtained in step S521 into the above model, and then, according to existing knowledge and algorithms, simulate the neurodegenerative pathological mechanism when the brain function is damaged. Exemplarily, it is known that abnormal changes in the light reflex sensitivity index of the pupil may be related to the injury of the brain nerve conduction pathway. The model will infer the possible impacts on the brain nerve conduction pathway based on the changes in this index in the pupil response characteristic data, such as the degree of nerve fiber injury and abnormal neurotransmitter transmission. Through the above simulation, generate brain function injury mechanism mapping data. The brain function injury mechanism mapping data corresponds the pupil response characteristics to the specific brain function injury pathological mechanism, providing an important basis for subsequent analysis; Step S523: Integrate the brain function injury mechanism mapping data with multi-modal brain image feature data to construct a brain function injury probability prediction model and output brain function injury probability distribution data. The multi-modal brain image feature data includes brain structure and function information obtained from different imaging examinations such as CT, such as the density, morphology, and metabolic activities of brain tissues. Integrate the brain function injury mechanism mapping data with the multi-modal brain image feature data, and combine machine learning or deep learning algorithms to construct a brain function injury probability prediction model. For example, use a convolutional neural network to extract features from brain image data, and then combine the extracted features with the brain function injury mechanism mapping data and input them into the neural network for training. By training a large number of case data with known brain function injury conditions, adjust the parameters of the model to enable the model to accurately predict the probability of brain function injury according to the input data. When new patient data is input, the model will output brain function injury probability distribution data. The brain function injury probability distribution data represents the probability situation of the patient under different degrees of brain function injury, such as the probability of mild injury, moderate injury, and severe injury, etc.; Step S524: Generate a disease deterioration risk coefficient based on the brain function injury probability distribution data and in combination with the clinical deterioration path quantification algorithm. The clinical deterioration path quantification algorithm is summarized based on clinical experience and research and is used to quantify the possibility of a patient's condition deteriorating from the current state to a worse direction. Input the brain function injury probability distribution data obtained in step S523 into the clinical deterioration path quantification algorithm. The clinical deterioration path quantification algorithm will comprehensively consider factors such as the degree of brain function injury, the patient's underlying diseases, age, gender, etc., and calculate the disease deterioration risk coefficient. Exemplarily, if the brain function injury probability distribution of a patient shows a relatively high probability of severe injury and the patient itself has other serious underlying diseases, such as cardiovascular disease or diabetes, then according to the clinical deterioration path quantification algorithm, the disease deterioration risk coefficient will increase accordingly. The larger the value of the disease deterioration risk coefficient, the higher the risk of disease deterioration, which provides an important reference basis for medical staff to judge the development trend of the patient's condition and formulate treatment plans.

[0029] It should be noted that triggering the grading warning mechanism specifically includes: Obtain the disease deterioration risk coefficient corresponding to the patient and compare the disease deterioration risk coefficient corresponding to the patient with a preset risk threshold; If the disease deterioration risk coefficient corresponding to the patient is between the first risk threshold and the second risk threshold, then trigger a primary warning signal, trigger a yellow flashing identifier on the electronic medical record interface of the intensive care central system, automatically generate a pupil grade assessment result report form and push it to the nurse's mobile terminal; If the disease deterioration risk coefficient corresponding to the patient is between the second risk threshold and the third risk threshold, then trigger an intermediate warning signal, and simultaneously activate the sound and light signals of the bedside alarm in the patient's ward and send a warning message to the attending physician; If the disease deterioration risk coefficient corresponding to the patient exceeds the third risk threshold, then trigger a high-level warning signal, start the hospital-wide emergency broadcast system, automatically retrieve the electronic signature of the preoperative informed consent form to verify the identity, and link to reserve an emergency scanning channel in the CT room. At the same time, transmit the pupil grade assessment result report form and the corresponding pupil dynamic database of the patient to the remote consultation center in real time through the 5G private network. When no medical staff confirmation feedback is received for 3 consecutive minutes, automatically trigger the standby emergency response protocol.

[0030] Specifically, the hierarchical early warning mechanism enables medical staff to reasonably allocate medical resources and arrange nursing work according to the different degrees of the risk of disease deterioration. For patients with low risk, medical staff can appropriately reduce the examination frequency and focus more on high-risk patients; for medium-risk patients, the treatment plan can be adjusted in a timely manner, and the changes in the condition can be closely monitored to prevent the condition from deteriorating further; for high-risk patients, emergency measures can be taken immediately, and expert consultations can be organized to save the patient's life to the greatest extent and improve the success rate of treatment. At the same time, it also helps the patient's family members understand the severity of the condition, make psychological preparations and cooperate with the treatment.

[0031] Embodiment 2 The embodiment of the present application also discloses a pupil image recognition and disease condition change evaluation system for image recognition.

[0032] Referring to Figure 2 , a pupil image recognition and disease condition change evaluation system for image recognition, includes: A data recognition module, configured to scan the QR code on the patient's wristband through a medical terminal device, obtain the patient's identity information and historical pupil data, and establish a pupil dynamic database corresponding to the patient; A data acquisition module, configured to alternately irradiate the patient's both eyes with a multi-spectral pupil imaging device, and synchronously acquire a dynamic image sequence corresponding to the pupil light reflection area of the patient; An analysis module, configured to perform spatio-temporal feature analysis on the dynamic image sequence, establish a dynamic change curve of the pupil diameter, and generate a pupil light reflex sensitivity index of the patient according to the dynamic change curve of the pupil diameter; An evaluation module, configured to construct a pupil dynamic evaluation model according to the dynamic change curve of the pupil diameter and the pupil light reflex sensitivity index of the patient, and then output a pupil grade evaluation result based on the pupil dynamic evaluation model; An early warning module, configured to comprehensively analyze the pupil grade evaluation result, and upload the analysis result to the intensive care central system to trigger a hierarchical early warning mechanism.

[0033] The above content is only an example and illustration of the concept of the present invention. Those skilled in the art of the present technology make various modifications or supplements to the described specific embodiments or use similar methods to replace them. As long as they do not deviate from the concept of the invention, they should all fall within the protection scope of the present invention.

[0034] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples.

[0035] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to only the specific implementation manners. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the present invention, so that those skilled in the relevant technical fields can well understand and utilize the present invention.

Claims

1. A method for pupil image recognition and disease condition change assessment in image recognition, characterized in that It includes the following steps: Step S1: Scan the QR code on the patient's wristband through the medical terminal device to obtain the patient's identity information and historical pupil data, and establish a pupil dynamic database corresponding to the patient; Step S2: Use a multispectral pupil imaging device to alternately irradiate the patient's both eyes, and synchronously collect a dynamic image sequence corresponding to the patient's pupil light reflex area; Step S3: Analyze the spatio-temporal features of the dynamic image sequence, establish a dynamic change curve of the pupil diameter, and generate a pupil light reflex sensitivity index for the patient according to the dynamic change curve of the pupil diameter; Step S4: Construct a pupil dynamic evaluation model based on the dynamic change curve of the pupil diameter and the pupil light reflex sensitivity index of the patient, and then output a pupil grade evaluation result based on the pupil dynamic evaluation model; Step S5: Comprehensively analyze the pupil grade evaluation result, and upload the analysis result to the intensive care central system to trigger a grading warning mechanism.

2. The pupil image recognition and disease condition change assessment method for image recognition according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Perform multi-frame pupil area segmentation and temporal alignment processing on the dynamic image sequence corresponding to the patient's pupil light reflex area, and extract the spatio-temporal feature data of the pupil contour; Step S32: Divide the spatio-temporal feature data into pupil contraction-dilation phases, and construct a dynamic change curve of the pupil diameter; Step S33: Track the iris texture displacement vector by the optical flow method, and then calculate the contraction rate of the pupil sphincter based on the iris texture displacement vector; Step S34: Generate a pupil light reflex sensitivity index according to the dynamic change curve of the pupil diameter and the contraction rate of the pupil sphincter.

3. The pupil image recognition and disease condition change assessment method for image recognition according to claim 2, characterized in that Step S4 includes the following steps: Step S41: Input the dynamic change curve of the pupil diameter into a pre-trained deep residual network to extract multi-scale spatio-temporal feature data; Step S42: Feature-weight the iris texture displacement vector based on the multi-scale spatio-temporal feature data and the attention mechanism, and then generate a sphincter movement feature vector; Step S43: Concatenate the pupil light reflex sensitivity index and the sphincter movement feature vector, input them into a bidirectional LSTM network for temporal modeling, and then construct a pupil dynamic evaluation model, and then output a pupil grade evaluation result based on the pupil dynamic evaluation model.

4. A pupil image recognition and disease condition change assessment method for image recognition according to claim 1, characterized in that, Step S5 includes the following steps: Step S51: Dynamically compare the pupil grade evaluation result with the corresponding historical pupil data in the patient's pupil dynamic database, and calculate the pupil parameter change gradient data; Step S52: Construct a brain function injury probability prediction model according to the pupil parameter change gradient data, and then generate a disease deterioration risk coefficient; Step S53: Upload the disease deterioration risk coefficient and the pupil grade evaluation result to the intensive care central system through blockchain encryption technology to trigger a grading warning mechanism.

5. The method for pupil image recognition and disease condition change assessment in image recognition according to claim 4, wherein, Step S52 includes the following steps: Step S521: Perform dynamic temporal analysis and feature extraction on the pupil parameter change gradient data to obtain pupil response feature data; Step S522: Based on the pupil response feature data, simulate the brain function degenerative pathological mechanism through a neurophysiological injury association model to generate brain function injury mechanism mapping data; Step S523: Integrate the brain function damage mechanism mapping data and multi-modal brain image feature data to construct a brain function damage probability prediction model and output brain function damage probability distribution data; Step S524: Generate a disease deterioration risk coefficient based on the brain function damage probability distribution data and in combination with the clinical deterioration path quantification algorithm.

6. The pupil image recognition and disease condition change evaluation method for image recognition according to claim 5, characterized in that, Trigger a hierarchical warning mechanism, specifically including: Obtain the disease deterioration risk coefficient corresponding to the patient, and compare the disease deterioration risk coefficient corresponding to the patient with a preset risk threshold; If the disease deterioration risk coefficient corresponding to the patient is between the first risk threshold and the second risk threshold, trigger a primary warning signal, trigger a yellow flashing identifier on the electronic medical record interface of the intensive care central system, automatically generate a pupil grade assessment result report and push it to the nurse's mobile terminal; If the disease deterioration risk coefficient corresponding to the patient is between the second risk threshold and the third risk threshold, trigger an intermediate warning signal, and simultaneously activate the sound and light signals of the bedside alarm in the patient's ward to send a warning message to the attending physician; If the disease deterioration risk coefficient corresponding to the patient exceeds the third risk threshold, trigger a high-level warning signal, start the hospital-wide emergency broadcast system, automatically retrieve the electronic signature of the preoperative informed consent form for identity verification, and link to reserve an emergency scanning channel in the CT room. At the same time, transmit the pupil grade assessment result report and the corresponding pupil dynamic database of the patient to the remote consultation center in real time through the 5G private network. When no medical confirmation feedback is received for 3 consecutive minutes, automatically trigger the standby emergency response protocol.

7. An iris image recognition and disease condition change assessment system for image recognition, which is applied to an iris image recognition and disease condition change assessment method described in any one of the above claims 1-6, and is characterized in that, Including: A data identification module for scanning the QR code on the patient's wristband through a medical terminal device to obtain the patient's identity information and historical pupil data, and establishing a corresponding pupil dynamic database for the patient; A data acquisition module for alternately irradiating the patient's both eyes with a multi-spectral pupil imaging device and synchronously acquiring a dynamic image sequence corresponding to the pupil light reflection area of the patient; An analysis module for performing spatio-temporal feature analysis on the dynamic image sequence, establishing a dynamic change curve of the pupil diameter, and generating a pupil light reflex sensitivity index for the patient based on the dynamic change curve of the pupil diameter; An evaluation module for constructing a pupil dynamic evaluation model based on the dynamic change curve of the pupil diameter and the pupil light reflex sensitivity index of the patient, and then outputting a pupil grade assessment result based on the pupil dynamic evaluation model; A warning module for comprehensively analyzing the pupil grade assessment result and uploading the analysis result to the intensive care central system to trigger a hierarchical warning mechanism.

8. A computer-readable storage medium, characterized in that: Stored with instructions, when the instructions are run on a computer, the computer is caused to execute a pupil image recognition and disease condition change assessment method according to any one of claims 1 to 6.

Images