A method and system for auxiliary fundus screening management of premature infants

Abstract

The present invention belongs to the technical field of auxiliary fundus screening for premature infants, and specifically relates to a method and system for managing auxiliary fundus screening for premature infants. The method comprises the following steps: obtaining various data of existing premature infants to train a prediction model, predicting fundus abnormalities of premature infants to be screened, screening premature infants who need fundus screening from the prediction results, collecting fundus images of premature infants to be screened, performing image analysis to determine whether there are fundus abnormalities, and outputting the results. Combined with the doctor's judgment, the existing fundus screening process for premature infants is improved, and premature infants who need fundus screening are screened in advance, so that all premature infants do not need to undergo eye treatment and then obtain fundus images, because the process of eye treatment and fundus photography is very painful for premature infants, and there are certain fundus screening problems. Image processing and analysis are used to assist in the judgment of fundus abnormalities, thereby improving the efficiency and accuracy of fundus screening.

Description

Technical Field

[0001] The present invention belongs to the technical field of auxiliary fundus screening for premature infants, and in particular relates to a method and system for auxiliary fundus screening management for premature infants. Background Art

[0002] A premature baby is a baby born before 37 weeks of gestation. Medically, premature babies are divided into extremely premature babies (gestational age less than 28 weeks), very premature babies (gestational age between 28 and 32 weeks) and premature babies (gestational age between 32 and 37 weeks).

[0003] Fundus screening is a very important examination for premature infants. Premature infants have a shorter gestational period and are more likely to develop retinopathy of prematurity. Severe retinopathy of prematurity can lead to blindness if not treated promptly. Early and timely treatment can preserve some useful vision in most cases. The younger the gestational age and the lighter the birth weight, the more likely the infant will develop retinopathy of prematurity. The purpose of fundus screening is to detect retinopathy of prematurity in a timely manner and provide appropriate treatment.

[0004] Currently, in clinical practice, doctors almost always recommend that parents of premature babies undergo fundus screening to promptly screen premature babies for retinal lesions or congenital abnormalities and other lesions of the fundus, so that appropriate treatment can be given in a timely manner. However, in clinical practice, not all premature babies have fundus abnormalities. A large proportion of premature babies do not have fundus abnormalities. However, in order to rule out fundus abnormalities, all premature babies will be screened for fundus. Fundus screening for all premature babies will increase the economic cost for the parents of premature babies and increase the pain of premature babies during fundus screening. In addition, during the fundus screening process for premature babies, the doctor uses an ophthalmoscope to analyze the fundus of the premature baby and determine whether there are fundus abnormalities. This method relies on the doctor's experience and has low efficiency and accuracy.

[0005] Therefore, how to improve the existing fundus screening process for premature infants, screen out premature infants who need fundus screening in advance, and use image processing and analysis to assist in the judgment of fundus abnormalities, thereby improving the efficiency and accuracy of fundus screening, is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for auxiliary fundus screening management of premature infants, so as to improve the existing fundus screening process of premature infants, screen out premature infants who need fundus screening in advance, and use image processing and analysis to assist in the judgment of fundus abnormalities, thereby improving the efficiency and accuracy of fundus screening.

[0007] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0008] In a first aspect, a method for auxiliary fundus screening management of premature infants is provided, comprising the following steps:

[0009] S1: Obtain basic data of premature infants and fundus screening data based on big data, and classify them to obtain text data and image data;

[0010] S2: Preprocess the text data and image data respectively, and perform text feature extraction and image feature extraction on the preprocessed text data and image data respectively;

[0011] S3: Create a fundus abnormality prediction model, input the extracted text features and image features into the fundus abnormality prediction model to train, test and verify the model;

[0012] S4: basic data of the premature infants to be screened are obtained and text features and image features are extracted. The extracted text features and image features are input into a fundus abnormality prediction model. The fundus abnormality prediction model predicts fundus abnormalities of the premature infants to be screened. The doctor determines the premature infants who need fundus screening based on the prediction results and executes step S5;

[0013] S5: The fundus images of the premature infants to be screened are collected and preprocessed, and input into the fundus abnormality prediction model to identify fundus abnormalities. The doctor makes a fundus abnormality judgment based on the processed fundus images and abnormality identification results, and gives a fundus abnormality result;

[0014] S6: Inputting the fundus abnormality results and their corresponding basic data into the fundus abnormality prediction model, and continuously optimizing and updating the model.

[0015] Preferably, the basic data include gestational period, gender, height and weight at birth, and various examination data during pregnancy.

[0016] Preferably, the specific process of classifying the basic data and fundus screening data to obtain text data and image data in step S1 is as follows:

[0017] S11: Create a data type recognition model, input the labeled text data and image data into the data type recognition model for model training, evaluate the data type recognition model performance, and if it meets the preset performance indicators, execute step S12, otherwise re-train the model;

[0018] S12: Inputting the basic data of the premature infant and the fundus screening data into the trained data type recognition model, wherein the data type recognition model performs data type recognition on the input data;

[0019] S13: The data type recognition model adds corresponding data type labels to the output data according to the data type recognition result.

[0020] Preferably, the specific process of preprocessing the text data in step S2 is as follows:

[0021] S21: extracting characters from the text data to form a text string, and splitting the text string into phrases;

[0022] S22: constructing a phrase list based on the phrases, establishing a mapping relationship between phrases and numbers based on the phrases and their frequencies in the phrase list, and then converting the phrase list into a digital index sequence.

[0023] Preferably, the specific process of preprocessing the image data in step S2 is as follows:

[0024] S23: Create a filter kernel and determine its size and standard deviation. Generate a two-dimensional matrix based on the size and standard deviation of the filter kernel. The elements in the two-dimensional matrix are generated by a specified two-dimensional function. The two-dimensional function is expressed as follows:

[0025] ;

[0026] in,( x , y ) is the coordinate of the pixel point of the filter kernel, ( x 0, y 0) is the coordinate of the center pixel point in the filter kernel, σ is the standard deviation;

[0027] S24: performing normalization processing on the two-dimensional matrix so that the sum of all matrix elements in the two-dimensional matrix is 1;

[0028] S25: Perform a kernel convolution operation on the image to be preprocessed and the normalized two-dimensional matrix. For each pixel in the preprocessed image, obtain its neighborhood with the same size as the two-dimensional matrix, multiply the pixel value in the neighborhood with the element at the corresponding position of the Gaussian kernel, and add the products. The result obtained is used as the new pixel value of the pixel.

[0029] Preferably, step S5 also includes fundus screening pain management for premature infants, and the specific process is as follows:

[0030] S51: Real-time collection of designated physiological indicators and behavioral data of premature infants undergoing fundus screening, wherein the physiological indicators include heart rate, blood pressure, and blood oxygen saturation, and the behavioral data include facial expressions, crying frequency and duration, and body movements;

[0031] S52: Based on the physiological indicators and behavioral data and using a preset pain scoring strategy, a pain score is performed on the premature infant undergoing fundus screening, and adjustments are made during the fundus screening process based on the pain scoring results.

[0032] In a second aspect, a premature infant fundus auxiliary screening management system is provided, which is used to implement the premature infant fundus auxiliary screening management method, including a data acquisition module, a data classification module, a preprocessing module, a feature extraction module, a fundus abnormality prediction model and a fundus image acquisition module, wherein the data acquisition module is connected to the data classification module, the data classification module is connected to the preprocessing module, the preprocessing module is connected to the feature extraction module, the feature extraction module is connected to the fundus abnormality prediction model, and the fundus abnormality prediction model is connected to the fundus image acquisition module;

[0033] The data acquisition module is used to acquire basic data and fundus screening data of premature infants based on big data;

[0034] The data classification module is used to classify the basic data and fundus screening data to obtain text data and image data;

[0035] The preprocessing module is used to preprocess the text data and the image data respectively;

[0036] The feature extraction module is used to perform text feature extraction and image feature extraction on the preprocessed text data and image data respectively;

[0037] The fundus abnormality prediction model is used to predict fundus abnormalities in premature infants to be screened;

[0038] The fundus image acquisition module is used to acquire fundus images of premature infants to be screened.

[0039] Preferably, the fundus abnormality prediction model in step S3 is a neural network model including an input layer, a processing layer and an output layer, wherein the input layer is provided with dual input channels, and the processing layer includes an input integration layer, a feature extraction layer, a prediction result generation layer and a feedback link layer;

[0040] The input integration layer is used to perform preliminary integration on the input data received from the two input channels;

[0041] The feature extraction layer is used to extract different features of the integrated data using different neurons;

[0042] The prediction result generation layer is used to generate prediction results of fundus abnormalities in premature infants;

[0043] The feedback connection layer is used to dynamically adjust the data processing process of the processing layer based on the input and output.

[0044] Preferably, the processing layer is provided with a number of neurons, each of which is provided with a specified activation function and learning strategy. The neurons adjust the learning strategy of the neurons by maximizing long-term rewards and use a stochastic gradient descent algorithm to update the neuron parameters to achieve end-to-end control optimization.

[0045] The beneficial effects of the present invention include:

[0046] The present invention provides a method and system for auxiliary fundus screening management of premature infants. After obtaining various data of existing premature infants to train the prediction model, the fundus abnormalities of the premature infants to be screened are predicted, and the premature infants who need to undergo fundus screening are screened from the prediction results. The fundus images of the premature infants to be screened are collected for image analysis to determine whether there are fundus abnormalities and the identification results are output. Combined with the doctor's judgment, the existing fundus screening process for premature infants is improved, and premature infants who need fundus screening are screened in advance, so that all premature infants do not need to undergo eye treatment and then obtain fundus images, because the process of eye treatment and fundus photography is very painful for premature infants, and there are certain fundus screening. The fundus abnormalities are determined with the help of image processing and analysis to improve the efficiency and accuracy of fundus screening.

[0047] First, the model is trained by obtaining basic data of premature infants and fundus screening data based on big data, so that the fundus abnormality prediction model can learn the relationship between the basic data of premature infants and fundus abnormalities. Then, the fundus abnormality prediction model is used to predict fundus abnormalities in premature infants to be screened, so that there is no need to perform fundus screening on all premature infants, saving medical resources, reducing the economic costs of parents of premature infants, and avoiding the pain of premature infants during fundus screening.

[0048] Secondly, the model can be used to identify fundus abnormalities in fundus images, thereby improving the efficiency of fundus screening. Combined with the doctor's diagnosis, it can avoid misscreening and improve the accuracy of fundus screening for premature infants.

[0049] Again, the fundus abnormality prediction model is a neural network model including an input layer, a processing layer and an output layer. The input layer is provided with dual input channels. The fundus abnormality prediction model has dual input channels, and the processing layer is provided with an input integration layer, a feature extraction layer, a prediction result generation layer and a feedback link layer. It can be optimized and updated through each fundus image recognition, so that the model has better generalization and higher accuracy, which provides for fundus screening of premature infants. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of the method for auxiliary fundus screening and management of premature infants of the present invention.

[0051] Figure 2 Schematic diagram of the architecture of the fundus abnormality prediction model of the present invention.

[0052] Figure 3 This is a schematic diagram of the fundus screening pain management process of the present invention. DETAILED DESCRIPTION

[0053] The following is combined with Figures 1 to 3 The present invention is described in further detail:

[0054] Example 1

[0055] See attached Figure 1 As shown, a method for auxiliary fundus screening and management of premature infants includes the following steps:

[0056] S1: Based on big data, basic data and fundus screening data of premature infants are obtained, and classified to obtain text data and image data. The basic data include gestational period, gender, height and weight at birth, and various examination data during pregnancy. Since there is a certain relationship between the various developments of premature infants, and the various developments of the body, the present invention obtains a large amount of data of premature infants who have undergone fundus screening, including basic data and fundus screening data, and creates a neural network model to capture the relationship between basic data and fundus screening data, and then predicts fundus abnormalities for subsequent premature infants to be screened, and determines whether fundus screening is needed based on the prediction results, so as to avoid premature infants who do not need fundus screening from suffering the physical pain caused by fundus screening.

[0057] S2: Preprocess the text data and image data separately, and perform text feature extraction and image feature extraction on the preprocessed text data and image data respectively. The data preprocessing process is based on the previous data classification, and preprocessing and feature extraction are performed based on the classification results, improving the efficiency of preprocessing and feature extraction.

[0058] S3: Create a fundus abnormality prediction model and input the extracted text and image features into the model for training, testing, and verification. By pre-extracting text and image features from a large amount of data from premature infants who have undergone fundus screening to train the model, the model can directly capture the relationship between text and image features and fundus abnormalities, improving model training efficiency.

[0059] S4: Obtain the basic data of the premature infants to be screened and extract text features and image features, input the extracted text features and image features into the fundus abnormality prediction model, the fundus abnormality prediction model predicts the fundus abnormalities of the premature infants to be screened, and the doctor determines the premature infants who need to undergo fundus screening based on the prediction results, and executes step S5. When the doctor determines whether the premature infants to be screened need fundus screening, he first obtains the prediction results of the fundus abnormality prediction model for the fundus abnormalities of the premature infants to be screened, and then combines the various data of the premature infants to finally determine whether fundus screening is needed. For premature infants who are in the critical condition of needing screening or not needing screening, or are close to the critical condition, it is determined that fundus screening is necessary, so as to avoid the misjudgment of the presence of fundus abnormalities and not undergoing fundus screening, and ensure that premature infants who are judged not to need fundus screening do not need screening.

[0060] S5: The fundus images of the premature infants to be screened are collected and preprocessed, and input into the fundus abnormality prediction model for fundus abnormality identification. The doctor makes a fundus abnormality judgment based on the processed fundus images and abnormality identification results, and gives the fundus abnormality result.

[0061] S6: Inputting the fundus abnormality results and their corresponding basic data into the fundus abnormality prediction model, and continuously optimizing and updating the model.

[0062] Retinopathy of prematurity is a developmental disorder of the retinal blood vessels in premature infants, with a high incidence rate. Fundus screening at designated times after birth is a globally preventable cause of childhood blindness. Therefore, in current clinical practice, nearly all premature infants, regardless of their preterm age, are screened to avoid the possibility of missing out on timely treatment for fundus abnormalities that could be detected through screening. However, a significant proportion of premature infants, particularly those born between 32 and 37 weeks of gestation, have normal fundus development. Screening these infants would waste medical resources and increase the cost to their parents. Most importantly, the screening process is painful for these infants, as it involves placing eye drops to dilate the pupil, holding the eyelids open, and then photographing or observing the fundus with an ophthalmoscope to identify abnormalities.

[0063] After obtaining various data of existing premature infants to train the prediction model, the fundus abnormalities of premature infants to be screened are predicted, and premature infants who need fundus screening are screened from the prediction results. By collecting fundus images of premature infants to be screened, image analysis is performed to determine whether there are fundus abnormalities and the identification results are output. Combined with the doctor's judgment, the existing fundus screening process for premature infants is improved, and premature infants who need fundus screening are screened in advance, so that all premature infants no longer need to undergo eye treatment and then obtain fundus images, because the process of eye treatment and fundus photography is very painful for premature infants, and there are certain fundus screening. Image processing and analysis are used to assist in the judgment of fundus abnormalities, thereby improving the efficiency and accuracy of fundus screening.

[0064] Example 2

[0065] On the basis of Example 1, the specific process of classifying the basic data and fundus screening data to obtain text data and image data in step S1 is as follows:

[0066] S11: Create a data type recognition model, input the labeled text data and image data into the data type recognition model for model training, evaluate the data type recognition model performance, and if it meets the preset performance indicators, proceed to step S12; otherwise, retrain the model. The data type recognition model is a neural network model structure. By using the labeled data for model training, the model has the ability to quickly recognize data types with high accuracy.

[0067] S12: Inputting the basic data of the premature infant and the fundus screening data into the trained data type recognition model, wherein the data type recognition model performs data type recognition on the input data;

[0068] S13: The data type recognition model adds corresponding data type labels to the output data according to the data type recognition results. By adding data type labels to different data types, when the data is subsequently input into the fundus abnormality prediction model, it is automatically input into the input channel of the corresponding type according to the data type label. The text data is input into the text data channel of the fundus abnormality prediction model, and the image data is input into the image data channel of the fundus abnormality prediction model.

[0069] The data type recognition model is a dual-branch hybrid neural network structure consisting of a text branch and an image branch. The text branch is based on a variant of the BERT pre-trained model and is used to process text data in electronic medical records, including gestational age, birth weight, oxygen inhalation duration, and other data. The image branch is a lightweight MobileNetV3 network that processes fundus images, including JPEG / DICOM format files collected by RetCam. The data type recognition model also includes a fusion layer: a multi-head self-attention mechanism is used to align text and image feature vectors, outputting a unified data type discrimination result.

[0070] The training process for the data type recognition model begins with training data preparation. This training data uses a labeled dataset consisting of over 10,000 annotated text samples, covering both structured and unstructured text such as laboratory reports and medical records. Over 8,000 annotated fundus images, including samples of normal fundus, ROP lesions at various stages, and artifacts, are also included. Image data enhancement processing is performed, including random rotation (±15°), brightness adjustment (±20%), and Gaussian noise injection (σ≤0.05), to improve model robustness.

[0071] The trained data type recognition model was optimized and validated, with the optimization objective set to minimize the cross-entropy loss function. The Adam optimizer was used, with an initial learning rate of 1e-4 and a batch size of 32. The model was considered qualified when the validation set accuracy was ≥98% and the F1 score was ≥0.97. If this was not achieved, a dynamic curriculum learning strategy was adopted: training was prioritized on easily categorized samples, such as clear fundus images, and complex cases, such as mixed hemorrhagic / exudative lesions, were gradually added.

[0072] Data is classified and preprocessed, with text data divided into unstructured text and numerical data. Unstructured text, including doctors' handwritten notes, is subjected to OCR recognition and keyword extraction, using the regular expression "GA < 28 weeks" to match. Numerical data, including birth weight, is normalized and mapped to the [0, 1] interval. Image data is uniformly scaled to 512×512 pixels, and histogram equalization is applied to eliminate imaging differences between devices.

[0073] For inference and classification of the data type recognition model, the preprocessed data is fed into the trained data type recognition model: 128-dimensional feature vectors are generated for text data through the text branch, and 256-dimensional feature vectors are generated for image data through the image branch. The cosine similarity of the two types of data is calculated through the fusion layer. If the similarity is greater than 0.9, the data is considered to be of the same type. Otherwise, an anomaly alarm is triggered, such as mistaking text for an image.

[0074] Label each data item and assign channels to it. Add text data labels to text data, add text_meta, metadata such as "birth weight_1500g" or text_desc, descriptive text such as "oxygen inhalation history_positive".

[0075] Add img_raw (raw image) or img_processed (pre-processed image) to the image data tag, along with the resolution and acquisition device identifier.

[0076] Multi-channel input integration is used to build a dynamic routing module. Text data is encapsulated in JSON format and fed into the text embedding layer of the fundus abnormality prediction model. Image data is converted to HDF5 format and fed into the model's convolutional feature extraction layer. If unlabeled data types, such as video streams, are detected, data isolation is automatically triggered and administrators are notified.

[0077] In actual hospital tests, the text / image classification error rate dropped from 5.3% of traditional rule engines to 0.8%, and the time required to process single-case data was shortened from 2 minutes of manual operation to 0.5 seconds. It supports concurrent processing of 200+ cases per second, and is compatible with more than 15 medical data formats through a dynamic routing mechanism to meet the needs of multi-center collaboration.

[0078] In this embodiment, the specific process of preprocessing the text data in step S2 is as follows:

[0079] S21: Convert the text data into a specified encoding format, remove the noise data, and concatenate the remaining text in the original order to form a continuous text string, which is then processed using the specified string operation function. Normalize the text string, convert all letters to uppercase or lowercase, unify the representation of numbers, etc., to reduce the diversity of the text and facilitate subsequent model analysis and processing. Verify the generated text string to check for garbled characters, typos, and other problems, and make corresponding corrections. Finally, split the text string into phrases;

[0080] S22: Construct a phrase list based on the phrases, establish a mapping relationship between phrases and numbers based on the phrases and their frequencies in the phrase list, and then convert the phrase list into a numerical index sequence. For high-frequency words, a fixed index is assigned, such as 1-10. For medium- and low-frequency words, an index is assigned in descending order of frequency, starting at 11. Unregistered words, i.e., newly appeared words, are uniformly marked as 0 to establish a special index. The cleaned phrases are converted into a numerical sequence according to the mapping table, and the numerical sequence is normalized and truncated / padded to a fixed length.

[0081] The specific process of preprocessing the image data in step S2 is as follows:

[0082] S23: Create a filter kernel and determine its size and standard deviation. Generate a two-dimensional matrix based on the size and standard deviation of the filter kernel. The elements in the two-dimensional matrix are generated by a specified two-dimensional function. The two-dimensional function is expressed as follows:

[0083] ;

[0084] in,( x , y ) is the coordinate of the pixel point of the filter kernel, G( x , y ) is the pixel point in the two-dimensional matrix ( x , y )’s two-dimensional function value, ( x 0, y 0) is the coordinate of the center pixel point in the filter kernel, σ The standard deviation is adjusted dynamically according to the noise level of the fundus image. σ It is 1.5, and increases to 2.5 when the noise is serious;

[0085] The size of the filter kernel is determined by k size = 2* int (3σ) + 1 determines, e.g. σ =1.5 generates a 7×7 kernel, i.e. 3 σ =4.5→int takes 4→core width = 9→corrected to 7 to ensure its symmetry.

[0086] S24: performing normalization processing on the two-dimensional matrix so that the sum of all matrix elements in the two-dimensional matrix is 1;

[0087] Calculate the sum of all elements of the current kernel matrix. For example, the initial sum of a 5×5 kernel is 0.4787. Normalize each element:

[0088] ;

[0089] in, is the pixel point in the two-dimensional matrix The normalized value of the two-dimensional function value is the sum of the normalized kernel matrix elements is strictly equal to 1, ensuring that the overall brightness of the image remains unchanged after convolution.

[0090] S25: Perform kernel convolution operation on the preprocessed image using the normalized two-dimensional matrix, and traverse each pixel in the image in the specified order of the pixels. For each pixel in the preprocessed image, obtain its neighborhood with the same size as the two-dimensional matrix, multiply the pixel value in the neighborhood with the element at the corresponding position of the Gaussian kernel, and add the products to obtain the result as the new pixel value of the pixel. When performing the kernel convolution operation, add one or more layers of pixel points with a pixel value of 0 outside the edge of the image so that a complete kernel convolution operation can be performed on all pixels.

[0091] Edge zero padding strategy‌, for the input image, assuming the size is H×W, add pad width = k size / 2 layers of zero pixels, for example, a 7×7 kernel needs to be padded with 3 layers, and the image size after padding is: (H+6)×(W+6).

[0092] Sliding window convolution, traversing each pixel of the image in raster scan order ( i , j ), extract ( i , j ) is the center k size × k size Neighborhood window, calculate the weighted sum:

[0093] ;

[0094] in, is the weighted sum of the convolution of the pixels in the neighborhood window, is the pixel point in the two-dimensional matrix The normalized value of the two-dimensional function value is Pixel The convolution value of .

[0095] Image optimization is achieved by using SIMD instructions to accelerate matrix multiplication and addition operations, achieving a single-channel processing speed of 120 frames per second for 512×512 pixel grayscale images. In a test set of fundus images of premature infants, the signal-to-noise ratio was improved from 18.7dB in the original image to 24.3dB, and the clarity of vascular boundaries was increased by 35%. Compared to traditional mean filtering, the processing speed of images with a 7×7 kernel was increased by 2.3 times. Support σ The value is adjusted in real time (0.5<σ<3.0) to adapt to the noise characteristics of different imaging devices.

[0096] Example 3

[0097] On the basis of Example 1 or Example 2, see Figure 2The fundus abnormality prediction model in step S3 is a neural network model, which is an improved DD-DC neural network model, including an input layer, a processing layer and an output layer. The input layer is provided with dual input channels. The fundus abnormality prediction model has dual input channels, one of which is used to input text data or text features. The input format of the text data is: 128-dimensional structured text vector and 768-dimensional unstructured text embedding vector.

[0098] Another input channel is used to input image data or image features, and the input format is a standardized fundus image of 512×512 pixels.

[0099] The processing layer includes an input integration layer, a feature extraction layer, a prediction result generation layer, and a feedback connection layer. The input integration layer is used to perform preliminary integration of data received from two input channels. The feature extraction layer is used to use different neurons to extract different features of the integrated data. The prediction result generation layer is used to generate prediction results for premature infant fundus abnormalities. The feedback connection layer is used to dynamically adjust the data processing process of the processing layer based on the input and output.

[0100] The processing layer also includes two processing units connected via an information exchange layer. Each processing unit is connected to two input channels and is used to process text data or image data, respectively. The processing units for processing text data and image data exchange information during their processing, respectively. This means that the different processing units are not completely independent in the text data and image data processing processes, and in fact, there is a process of information exchange.

[0101] The core components of the processing layer include the input integration layer, which uses the cross-attention mechanism to fuse bimodal features. The specific formula is as follows:

[0102] ;

[0103] in, Q text The query vector is generated by linear transformation of text features. K image and V image The key-value pairs generated for image features, d k is the dimension size of the key vector.

[0104] The processing layer sets up a dual processing unit interaction mechanism. The "text processing unit" contains a 3-layer BiLSTM network with 512 hidden units in each layer. The "image processing unit" contains 4 residual convolution blocks, and the number of channels of the residual convolution block increases from 64 to 256.

[0105] Set up a phased training protocol‌, where phase 1 is the first 50 rounds, freeze the image channel, and train only the text-related module. Phase 2 is 51-150 rounds, perform dual-channel joint training, and reduce the learning rate to 1e -5 ,After 151 rounds in stage 3, the feedback connection layer is enabled and gradient clipping is added.

[0106] Each neuron has a number of neurons, each with a specified activation function and learning strategy. Each neuron generates feedback control based on the input data and its own characteristics as a feedback controller. Each neuron may have an independent control strategy that maps input to output, and there are feedback connections between neurons. Neurons in the later processing chain can feedback and influence neurons in the earlier processing chain.

[0107] In another implementation of this embodiment, neurons adjust their learning strategies by maximizing long-term rewards and update their parameters using a stochastic gradient descent algorithm, achieving end-to-end control optimization. First, the neuron parameters are initialized, i.e., each neuron parameter is assigned an initial value, which is randomly determined within a specified range. Input data is propagated through the neurons. Each layer's input data is matrix-multiplied by its weights and then biased. The output of each layer is obtained using a preset activation function, and the output of that layer serves as the input for the next layer until the final output is obtained. After the model output is obtained, a loss function is calculated based on the model output and the true label to measure the difference between the model output and the true result. Then, starting from the output layer, the gradient of the loss function is calculated for each layer's parameters. The chain rule is used to calculate the partial derivative of the loss function with respect to each parameter, and these partial derivatives form the gradient. Through backpropagation, the gradient information is gradually propagated from the output layer to the input layer, ultimately obtaining the gradient of each parameter. A specified function is set based on the gradient of each parameter to update the parameter, achieving model optimization.

[0108] Step S5 also includes fundus screening and pain management for premature infants, see Figure 3 The specific process is as follows:

[0109] S51: Real-time collection of designated physiological indicators and behavioral data of premature infants undergoing fundus screening. The physiological indicators include heart rate, blood pressure, and blood oxygen saturation; the behavioral data include facial expressions, crying frequency and duration, and body movements. During fundus screening, premature infants need to endure great pain, as the various operations of fundus screening will cause pain and stress to premature infants. Since premature infants are young and cannot express themselves verbally, it is necessary to collect relevant data during fundus screening to facilitate subsequent pain assessment of premature infants based on the collected data.

[0110] S52: Based on the physiological indicators and behavioral data and using a preset pain scoring strategy, a pain score is performed on the premature infant undergoing fundus screening, and adjustments are made to the fundus screening process based on the pain score results. By performing a pain assessment based on various data of the premature infant during the fundus screening process, the premature infant's pain information is understood, and adjustments are made during the execution of the operation. When the pain score reaches a threshold, the relevant operation needs to be suspended, and the corresponding operation can be performed again after the premature infant's reaction is alleviated to a certain extent.

[0111] A premature infant fundus assisted screening management system, for implementing the aforementioned premature infant fundus assisted screening management method, comprises a data acquisition module, a data classification module, a preprocessing module, a feature extraction module, a fundus abnormality prediction model, and a fundus image acquisition module. The data acquisition module is connected to the data classification module, which is in turn connected to the preprocessing module, which is in turn connected to the feature extraction module, which is in turn connected to the fundus abnormality prediction model, which is in turn connected to the fundus image acquisition module. The data acquisition module is configured to acquire basic data and fundus screening data of premature infants based on large data sets. The data classification module is configured to classify the basic data and fundus screening data to produce text data and image data. The preprocessing module is configured to preprocess the text data and image data, respectively. The feature extraction module is configured to extract text features and image features from the preprocessed text data and image data, respectively. The fundus abnormality prediction model is configured to predict fundus abnormalities in premature infants to be screened. The fundus image acquisition module is configured to acquire fundus images of premature infants to be screened.

[0112] In summary, the method and system for auxiliary fundus screening management of premature infants provided by the present invention obtain various data of existing premature infants to train the prediction model, and then predict the fundus abnormalities of premature infants to be screened, screen premature infants who need fundus screening from the prediction results, and collect fundus images of premature infants to be screened to perform image analysis to determine whether there are fundus abnormalities and output the identification results. Combined with the doctor's judgment, the existing fundus screening process for premature infants is improved, and premature infants who need fundus screening are screened in advance, so that all premature infants no longer need to undergo eye treatment and then obtain fundus images, because the process of eye treatment and fundus photography is very painful for premature infants, and there are certain fundus screening. The fundus abnormalities are assisted by image processing and analysis to improve the efficiency and accuracy of fundus screening.

Claims

1. A method for auxiliary fundus screening and management of premature infants, characterized in that: The following steps are involved: S1: Obtain basic data of premature infants and fundus screening data based on big data, and classify them to obtain text data and image data; S2: Preprocess the text data and image data respectively, and perform text feature extraction and image feature extraction on the preprocessed text data and image data respectively; S3: Create a fundus abnormality prediction model, input the extracted text features and image features into the fundus abnormality prediction model to train, test and verify the model; S4: basic data of the premature infants to be screened are obtained and text features and image features are extracted. The extracted text features and image features are input into a fundus abnormality prediction model. The fundus abnormality prediction model predicts fundus abnormalities of the premature infants to be screened. The doctor determines the premature infants who need fundus screening based on the prediction results and executes step S5; S5: The fundus images of the premature infants to be screened are collected and preprocessed, and input into the fundus abnormality prediction model to identify fundus abnormalities. The doctor makes a fundus abnormality judgment based on the processed fundus images and abnormality identification results, and gives a fundus abnormality result; S6: Inputting the fundus abnormality results and the corresponding basic data into the fundus abnormality prediction model, and continuously optimizing and updating the model; Step S5 also includes fundus screening pain management for premature infants, and the specific process is as follows: S51: Real-time collection of designated physiological indicators and behavioral data of premature infants undergoing fundus screening, wherein the physiological indicators include heart rate, blood pressure, and blood oxygen saturation, and the behavioral data include facial expressions, crying frequency and duration, and body movements; S52: Based on the physiological indicators and behavioral data and using a preset pain scoring strategy, a pain score is performed on the premature infant undergoing fundus screening, and the fundus screening process is adjusted based on the pain scoring result.

2. A method for auxiliary fundus screening and management of premature infants according to claim 1, characterized in that: The basic data include gestational period, gender, height and weight at birth, and various examination data during pregnancy.

3. A method for auxiliary fundus screening and management of premature infants according to claim 1, characterized in that: The specific process of classifying the basic data and fundus screening data to obtain text data and image data in step S1 is as follows: S11: Create a data type recognition model, input the labeled text data and image data into the data type recognition model for model training, evaluate the data type recognition model performance, and if it meets the preset performance indicators, execute step S12, otherwise re-train the model; S12: Inputting the basic data of the premature infant and the fundus screening data into the trained data type recognition model, wherein the data type recognition model performs data type recognition on the input data; S13: The data type recognition model adds corresponding data type labels to the output data according to the data type recognition result.

4. A method for auxiliary fundus screening and management of premature infants according to claim 3, characterized in that: The specific process of preprocessing the text data in step S2 is as follows: S21: extracting characters from the text data to form a text string, and splitting the text string into phrases; S22: constructing a phrase list based on the phrases, establishing a mapping relationship between phrases and numbers based on the phrases and their frequencies in the phrase list, and then converting the phrase list into a digital index sequence.

5. A method for auxiliary fundus screening and management of premature infants according to claim 3, characterized in that: The specific process of preprocessing the image data in step S2 is as follows: S23: creating a filter kernel and determining its size and standard deviation, generating a two-dimensional matrix according to the size and standard deviation of the filter kernel, wherein elements in the two-dimensional matrix are generated by a specified two-dimensional function; S24: performing normalization processing on the two-dimensional matrix so that the sum of all matrix elements in the two-dimensional matrix is 1; S25: Perform a kernel convolution operation on the image to be preprocessed and the normalized two-dimensional matrix. For each pixel in the preprocessed image, obtain its neighborhood with the same size as the two-dimensional matrix, multiply the pixel value in the neighborhood with the element at the corresponding position of the Gaussian kernel, and add the products. The result obtained is used as the new pixel value of the pixel.

6. A premature infant fundus auxiliary screening management system, used to implement the premature infant fundus auxiliary screening management method according to any one of claims 1 to 5, characterized in that: It includes a data acquisition module, a data classification module, a preprocessing module, a feature extraction module, a fundus abnormality prediction model and a fundus image acquisition module, wherein the data acquisition module is connected to the data classification module, the data classification module is connected to the preprocessing module, the preprocessing module is connected to the feature extraction module, the feature extraction module is connected to the fundus abnormality prediction model, and the fundus abnormality prediction model is connected to the fundus image acquisition module; The data acquisition module is used to acquire basic data and fundus screening data of premature infants based on big data; The data classification module is used to classify the basic data and fundus screening data to obtain text data and image data; The preprocessing module is used to preprocess the text data and the image data respectively; The feature extraction module is used to perform text feature extraction and image feature extraction on the preprocessed text data and image data respectively; The fundus abnormality prediction model is used to predict fundus abnormalities in premature infants to be screened; The fundus image acquisition module is used to acquire fundus images of premature infants to be screened.

7. The premature infant fundus auxiliary screening and management system according to claim 6, characterized in that: The fundus abnormality prediction model in step S3 is a neural network model including an input layer, a processing layer and an output layer, wherein the input layer is provided with dual input channels, and the processing layer includes an input integration layer, a feature extraction layer, a prediction result generation layer and a feedback link layer; The input integration layer is used to perform preliminary integration on the input data received from the two input channels; The feature extraction layer is used to extract different features of the integrated data using different neurons; The prediction result generation layer is used to generate prediction results of fundus abnormalities in premature infants; The feedback connection layer is used to dynamically adjust the data processing process of the processing layer based on the input and output.

8. The premature infant fundus auxiliary screening and management system according to claim 7, characterized in that: The processing layer is provided with a number of neurons, each of which is provided with a specified activation function and a learning strategy. The neurons adjust the learning strategy of the neurons by maximizing long-term rewards and update the neuron parameters using a stochastic gradient descent algorithm.

Images