Nasal allergy prevention and management system

Abstract

The invention, which relates to the technical field of medical data processing, discloses a nasal allergy prevention and management system comprising a data acquisition module, a data analysis module, an allergy prediction module and a terminal management module. The data acquisition module is used for acquiring a personal data set, a nose ultrasonic Doppler image and a nose near-infrared image of a tester; the data analysis module is used for performing quantitative coding processing on the personal data set of the testee, and performing feature extraction on the nose ultrasonic Doppler image and the nose near-infrared image through a first convolutional neural network path and a second convolutional neural network path respectively; the allergy prediction module is used for performing three-phase feature fusion on the personal data feature vector, the ultrasonic Doppler output feature map and the near-infrared output feature map to obtain output fusion features, and performing prediction classification on the output fusion features to obtain a nose allergy prevention result; and the terminal management module is used for providing a display interface for a user, receiving a user operation instruction, establishing a personal file of a tester and storing data.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to a nasal allergy prevention and management system. Background Art

[0002] In recent years, thanks to improved living standards and widespread access to health knowledge, public health awareness in my country has rapidly increased, and awareness of allergies has also surged. Nasal allergies are common allergic conditions. Although relatively mild in severity, they can cause persistent and adverse impacts on patients' daily lives, work, study, and sleep quality. Due to the environmental impacts of urbanization and industrialization, the incidence of nasal allergies in my country has been increasing year by year. On the one hand, some people live in highly urbanized environments, away from the natural environment. Exposure to plant allergens such as pollen can easily lead to an overreaction of the immune system, resulting in nasal allergy symptoms. On the other hand, pollutants emitted during industrial production, chemical components in interior decoration products, and additives in daily necessities have increased the variety of allergens, contributing to the increased prevalence of nasal allergies. Currently, mainstream serum tests only detect specific IgE antibodies when nasal allergies have progressed to a severe stage. Furthermore, they can only detect a limited range of allergens and are not suitable for the early stages of nasal allergies. Therefore, developing management systems for the early prevention of nasal allergies is a promising research direction.

[0003] Currently, a Chinese invention patent application with application number 202311760428.1 discloses a modeling method for an allergic rhinitis prediction model. The application specifically includes: obtaining medical data of allergic rhinitis cases as a sample set for the neural network model; preprocessing the sample set to obtain a training set for the neural network model; establishing an allergic rhinitis prediction model; using the training set to train the allergic rhinitis prediction model, the allergic rhinitis prediction model includes a first prediction model and a second prediction model, the first prediction model is used to predict allergic rhinitis for subjects aged 0-7 years old, and the second prediction model is used to predict allergic rhinitis for subjects aged 7 years and above; dividing patients into two groups according to age: toddlers and non-toddlers, and establishing two sub-models for these two age groups, and using different training sets to train the two sub-models, thereby improving the scientific nature of the allergic rhinitis model establishment. However, the patient data collected and analyzed in this application only included seven common symptoms, including nasal congestion, nasal itching, sneezing, runny nose, bilateral subungual edema, coughing, and throat itching. The data dimension is insufficient, and the prediction results obtained by analysis are greatly affected by the subjectivity of the test subjects. Summary of the Invention

[0004] The technical problem solved by the present invention is that when traditional methods are used to prevent nasal allergies, the patient data collected and analyzed only include common symptoms, the collected data dimensions are insufficient, and the prediction results obtained by analysis are greatly interfered by the subjectivity of the testers.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: A nasal allergy prevention and management system, comprising: a data acquisition module, a data analysis module, an allergy prediction module and a terminal management module; The data acquisition module includes a personal information acquisition unit, an ultrasound acquisition unit, and an infrared acquisition unit, which are used to collect statistics of the patient's information to obtain a personal data set, obtain an ultrasound Doppler image of the tester's nose, and collect a near-infrared image of the tester's nose through a near-infrared imager; The data analysis module includes a personal information feature extraction unit, an ultrasonic Doppler image feature extraction unit, and a near-infrared image feature extraction unit, which are used to perform quantization encoding processing on the tester's personal data set to obtain a personal data feature vector, perform feature extraction on the nasal ultrasonic Doppler image through a first convolutional neural network path to obtain an ultrasonic Doppler output feature map, and perform feature extraction on the nasal near-infrared image through a second convolutional neural network path to obtain a near-infrared output feature map; The allergy prediction module includes a feature fusion unit and a classification prediction unit, which are used to perform three-phase feature fusion on the personal data feature vector, the ultrasound Doppler output feature map, and the near-infrared output feature map to obtain output fusion features, and then predict and classify the output fusion features to obtain nasal allergy prevention results; The terminal management module includes a data management unit and a terminal interaction unit, which is used to provide a display interface for users and receive user operation instructions, establish personal files of all testers and store data.

[0006] As a preferred solution of the nasal allergy prevention and management system described in the present invention, the personal information collection unit is used to collect statistics on the patient's information to obtain a personal data set, and the personal data set includes: age, gender, nasal allergy history, family allergy history, allergen environment, sneezing, nasal congestion, itchy nose, runny nose, itchy eyes, red and swollen eyes, itchy throat, cough and chest tightness.

[0007] As a preferred embodiment of the nasal allergy prevention and management system described in the present invention, the ultrasonic acquisition unit is used to obtain an ultrasonic Doppler image of the test subject's nose, and the ultrasonic diagnostic instrument is set based on preset two-dimensional ultrasonic scanning parameters and ultrasonic Doppler parameters. The ultrasonic linear array probe is used to perform a two-dimensional ultrasonic scan of the nasal mucosa and superficial blood vessels on both sides of the test subject's nose, and the ultrasonic Doppler function is fixedly activated to obtain an ultrasonic Doppler image of the test subject's nose. The preset two-dimensional ultrasonic scanning parameters include: ultrasonic frequency and scanning depth, the ultrasonic frequency is set to 12 MHz, and the scanning depth is set to 0.5 cm; The ultrasound Doppler parameters include Doppler gain and velocity range, and the velocity range is set to 2 cm / s.

[0008] As a preferred embodiment of the nasal allergy prevention and management system described in the present invention, the infrared acquisition unit is configured to acquire a near-infrared image of the subject's nose using a near-infrared imager, perform a resting adjustment judgment on the subject, and, if the resting adjustment judgment indicates that the subject is suitable for acquisition, illuminate the subject with a light source having preset near-infrared acquisition parameters, and acquire a near-infrared image of the subject's nose by vertically aligning the near-infrared imager lens with the base of the subject's nose. The resting adaptation operation includes: having the test subject sit still for T time, measuring the test subject's pulse and body temperature, and determining that the test subject's pulse and body temperature are suitable for collection when the test subject's pulse is less than the resting pulse threshold and the test subject's body temperature is less than the resting body temperature threshold; The preset near-infrared acquisition parameters include: the wavelength of the near-infrared light source and the near-infrared imager are set to 850 nm, and the distance between the near-infrared imager lens and the tester's nose is set to 30 cm.

[0009] As a preferred solution of the nasal allergy prevention and management system described in the present invention, the personal information feature extraction unit is used to perform quantization coding processing on the tester's personal data set to obtain a personal data feature vector. The logic of the quantization coding processing includes: The age data in the personal data set is encoded by assigning 0 to the age group of 0-12 years old, 1 to the age group of 13-35 years old, 2 to the age group of 36-50 years old, and 3 to the age group of 50 years old and above to obtain the first quantitative code; The gender data in the personal data set is encoded by assigning 0 to male and 1 to female to obtain the second quantitative code; The nasal allergy history, family allergy history, and allergen environmental data in the personal data set are coded in sequence according to the value of 0 if they exist and 1 if they do not exist, to obtain the third quantitative code; The data of sneezing, nasal congestion, nasal itching, runny nose, itchy eyes, red and swollen eyes, itchy throat, cough, and chest tightness in the personal data set are coded in sequence according to the value of 0 for no symptoms, 1 for mild symptoms, and 2 for obvious symptoms to obtain the fourth quantitative code; The first quantization code, the second quantization code, the third quantization code and the fourth quantization code are sequentially connected in series to obtain a personal data feature vector.

[0010] As a preferred embodiment of the nasal allergy prevention and management system described in the present invention, the ultrasonic Doppler image feature extraction unit is configured to extract features from the nasal ultrasonic Doppler image through a first convolutional neural network path to obtain an ultrasonic Doppler output feature map. The processing logic for extracting features from the nasal ultrasonic Doppler image through the first convolutional neural network path includes: The nasal ultrasound Doppler image is input and deep denoised using a DRCNN module consisting of five dilated convolutional layers to obtain an ultrasound denoised feature map. The size of the five dilated convolutional layers in the DRCNN module is set to 3×3, and the dilation rates are set to 1, 2, 4, 2, and 1, respectively. The ultrasonic denoising feature map is extracted through the ResNeXt-50 network. The pixel values ​​of the ultrasonic denoising feature map are subjected to sliding convolution calculation through the convolution layer to obtain the weighted sum of the pixels corresponding to each window. The weighted sum data of the pixels corresponding to each window are nonlinearly mapped through the ReLU activation function to obtain the ultrasonic local feature map. The ultrasonic local feature map is subjected to maximum pooling calculation through a 2*2 pooling window in the pooling layer to obtain the downsampled feature map. The ultrasonic denoising feature map and the downsampled feature map are subjected to residual connection calculation through the residual block to obtain the ultrasonic Doppler output feature map.

[0011] As a preferred embodiment of the nasal allergy prevention and management system described in the present invention, the near-infrared image feature extraction unit is configured to extract features from the nasal near-infrared image through a second convolutional neural network path to obtain a near-infrared output feature map. The processing logic for extracting features from the nasal near-infrared image by the second convolutional neural network path includes: The nose near-infrared image is denoised by Gaussian filtering to obtain a denoised near-infrared image. The denoised near-infrared image is histogram equalized by the CLAHE algorithm to obtain an enhanced near-infrared image. The enhanced near-infrared image is preliminarily processed by a standard convolution layer of size 3x3, and a preliminary infrared feature map is obtained by batch normalization operation and ReLU function activation operation. The depth-wise separable convolution is used to convolve each channel in the preliminary infrared feature map to obtain a deep infrared feature map. The deep infrared feature map is subjected to global average pooling and Sigmoid function activation by the SE module, and is combined and calculated by a 1x1 point convolution kernel to obtain an abstract infrared feature map. The channel attention weight and spatial attention weight of the abstract infrared feature map are calculated by the CBAM attention module. The abstract infrared feature map is element-wise multiplied based on the channel attention weight and spatial attention weight to obtain the near-infrared output feature map.

[0012] As a preferred solution of the nasal allergy prevention and management system described in the present invention, the feature fusion unit is used to perform three-phase feature fusion on the personal data feature vector, the ultrasound Doppler output feature map, and the near-infrared output feature map to obtain an output fusion feature. The processing logic for performing the three-phase feature fusion includes: Based on the channel number of the ultrasound Doppler output feature map and the near-infrared output feature map, the personal data feature vector is linearly transformed and expanded to obtain a two-dimensional tensor of personal data. A bidirectional feature pyramid network (BiFPN) is established, which assigns the two-dimensional tensor of personal data to the top level, the ultrasound Doppler output feature map to the middle level, and the near-infrared output feature map to the bottom level. According to the order from the bottom layer to the highest layer in BiFPN, each layer is upsampled in turn by bilinear interpolation to obtain the upsampled features corresponding to each layer. According to the order from the highest layer to the bottom layer in BiFPN, each layer is downsampled in turn by maximum pooling to obtain the downsampled features corresponding to each layer. The upsampled features and upsampled features of each layer are vector-concatenated to obtain the intermediate fusion features corresponding to each layer. The first convolution pooling module in the AOA attention is used to perform an attention calculation on the intermediate fusion features corresponding to each level to obtain the first fusion attention weight corresponding to each level. The first fusion attention weight corresponding to each level and the intermediate fusion features are multiplied element by element to obtain the first attention features corresponding to each level. The second convolution pooling module in the AOA attention is used to perform a secondary attention calculation on the first attention features corresponding to each level to obtain the second fusion attention weight corresponding to each level. The second fusion attention weight corresponding to each level and the first attention features are multiplied element by element to obtain the final fusion features corresponding to each level. The final fusion features corresponding to all levels are weighted summed and processed by maximum pooling to obtain the output fusion features.

[0013] As a preferred solution of the nasal allergy prevention and management system described in the present invention, the classification prediction unit is used to predict and classify the output fusion features to obtain nasal allergy prevention results. The processing logic for the prediction and classification includes: The output fusion feature is input into the fully connected layer module of the two-layer structure, the output fusion feature is forward propagated through the first fully connected layer and the second fully connected layer module, the output fusion feature is multiplied by the first weight matrix in the first fully connected layer to obtain the first product, the first product is added by the first bias term, and a nonlinear transformation is performed through the ReLU activation function to obtain the transition output score; The number of neuron nodes in the first fully connected layer is N, and the number of neuron nodes in the second fully connected layer is 2; The output result of the first node is multiplied by the second weight matrix in the second fully connected layer to obtain a second product, the second product is added by the second bias term, and a nonlinear transformation is performed by the Softmax activation function to obtain the final output score result; Compare the two probability values ​​in the final output score result to obtain the larger probability value, and obtain the nasal allergy prevention judgment result based on the neuron node label corresponding to the larger probability value; The nasal allergy prevention judgment results include low prevention urgency and high prevention urgency.

[0014] As a preferred solution of the nasal allergy prevention and management system described in the present invention, the terminal interaction unit is used to provide a display interface for the user and receive user operation instructions, allowing the user to adjust and set parameters of the data acquisition module, data analysis module and allergy prediction module; The data management unit is used to establish personal files of all testers through the database, and to store and update the output data of the data acquisition module, data analysis module and allergy prediction module.

[0015] Beneficial effects of the present invention: In addition to collecting personal data, the present application also collects near-infrared images and ultrasonic Doppler images of the tester's nose for analysis, which can obtain information such as the condition of the nasal mucosa and the blood circulation condition of the nasal cavity, which is conducive to adding objective information in addition to the patient's personal data for nasal allergy prevention analysis. When acquiring near-infrared images, a resting adjustment operation is added, which is conducive to reducing the error caused by the tester's body temperature. At the same time, the personal data is encoded to generate a feature vector, and is assigned to the highest level of the bidirectional pyramid network during feature fusion, which is conducive to obtaining an abstract expression of the personal data and fully maintaining the key position of the personal data in the fusion process. Feature extraction of the nasal ultrasonic Doppler image and the nasal near-infrared image is performed respectively through the first convolutional neural network path and the second convolutional neural network path, which is conducive to targeted noise suppression and image feature extraction of the two image modal characteristics, and is conducive to obtaining abstract features that improve complementary information. Cross-level information interaction and feature fusion are achieved through the BiFPN architecture, and the AOA attention mechanism is introduced. Through two attention calculations, the key information in the fused features is accurately focused, which further improves the fusion effect, helps to improve the accuracy of subsequent classification predictions, and achieves early prevention of nasal allergies, thereby reducing the risk of nasal allergies. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of the basic flow of a nasal allergy prevention and management system provided by one embodiment of the present invention.

[0017] Figure 2 An ultrasonic Doppler image of the nose provided by one embodiment of the present invention; Figure 3 A near-infrared image of the nose provided by one embodiment of the present invention; Figure 4 An ultrasonic Doppler output characteristic diagram provided by one embodiment of the present invention; Figure 5 An ROC curve diagram provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0019] Reference Figure 1 , as one embodiment of the present invention, provides a nasal allergy prevention and management system, including: a data acquisition module, a data analysis module, an allergy prediction module and a terminal management module; The data acquisition module includes a personal information acquisition unit, an ultrasound acquisition unit, and an infrared acquisition unit, which are used to collect statistics of the patient's information to obtain a personal data set, obtain an ultrasound Doppler image of the tester's nose, and collect a near-infrared image of the tester's nose through a near-infrared imager; The data analysis module includes a personal information feature extraction unit, an ultrasonic Doppler image feature extraction unit, and a near-infrared image feature extraction unit, which are used to perform quantization encoding processing on the tester's personal data set to obtain a personal data feature vector, perform feature extraction on the nasal ultrasonic Doppler image through a first convolutional neural network path to obtain an ultrasonic Doppler output feature map, and perform feature extraction on the nasal near-infrared image through a second convolutional neural network path to obtain a near-infrared output feature map; The allergy prediction module includes a feature fusion unit and a classification prediction unit, which are used to perform three-phase feature fusion on the personal data feature vector, the ultrasound Doppler output feature map, and the near-infrared output feature map to obtain output fusion features, and then predict and classify the output fusion features to obtain nasal allergy prevention results; The terminal management module includes a data management unit and a terminal interaction unit, which is used to provide a display interface for users and receive user operation instructions, establish personal files of all testers and store data.

[0020] In this embodiment, the personal information collection unit is used to collect statistics on the patient's information to obtain a personal data set, and the personal data set includes: age, gender, history of nasal allergies, family allergy history, allergen environment, sneezing, nasal congestion, itchy nose, runny nose, itchy eyes, red and swollen eyes, itchy throat, cough and chest tightness.

[0021] In this embodiment, the ultrasound acquisition unit is used to obtain an ultrasound Doppler image of the test subject's nose. The ultrasound diagnostic instrument is set based on preset two-dimensional ultrasound scanning parameters and ultrasound Doppler parameters. A two-dimensional ultrasound scan of the nasal mucosa and superficial blood vessels on both sides of the test subject's nose is performed using an ultrasound linear array probe. The ultrasound Doppler function is permanently activated to obtain an ultrasound Doppler image of the test subject's nose. The preset two-dimensional ultrasonic scanning parameters include: ultrasonic frequency and scanning depth, the ultrasonic frequency is set to 12 MHz, and the scanning depth is set to 0.5 cm; The ultrasound Doppler parameters include Doppler gain and velocity range, and the velocity range is set to 2 cm / s.

[0022] In this embodiment, the infrared acquisition unit is used to acquire a near-infrared image of the subject's nose using a near-infrared imager, and perform a resting adjustment judgment on the subject. If the resting adjustment judgment result indicates that the subject is suitable for acquisition, the subject is illuminated by a light source with preset near-infrared acquisition parameters, and the near-infrared imager lens is vertically aligned with the nasal root of the subject to acquire a near-infrared image of the subject's nose. The resting adaptation operation includes: having the test subject sit still for T time, measuring the test subject's pulse and body temperature, and determining that the test subject's pulse and body temperature are suitable for collection when the test subject's pulse is less than the resting pulse threshold and the test subject's body temperature is less than the resting body temperature threshold; The preset near-infrared acquisition parameters include: the wavelength of the near-infrared light source and the near-infrared imager are set to 850 nm, and the distance between the near-infrared imager lens and the tester's nose is set to 30 cm.

[0023] In this embodiment, the personal information feature extraction unit is used to perform quantization coding processing on the tester's personal data set to obtain a personal data feature vector. The logic of the quantization coding processing includes: The age data in the personal data set is encoded by assigning 0 to the age group of 0-12 years old, 1 to the age group of 13-35 years old, 2 to the age group of 36-50 years old, and 3 to the age group of 50 years old and above to obtain the first quantitative code; The gender data in the personal data set is encoded by assigning 0 to male and 1 to female to obtain the second quantitative code; The nasal allergy history, family allergy history, and allergen environmental data in the personal data set are coded in sequence according to the value of 0 if they exist and 1 if they do not exist, to obtain the third quantitative code; The data of sneezing, nasal congestion, nasal itching, runny nose, itchy eyes, red and swollen eyes, itchy throat, cough, and chest tightness in the personal data set are coded in sequence according to the value of 0 for no symptoms, 1 for mild symptoms, and 2 for obvious symptoms to obtain the fourth quantitative code; The first quantization code, the second quantization code, the third quantization code and the fourth quantization code are sequentially connected in series to obtain a personal data feature vector.

[0024] In this embodiment, the ultrasonic Doppler image feature extraction unit is used to extract features from the nasal ultrasonic Doppler image through the first convolutional neural network path to obtain an ultrasonic Doppler output feature map. The processing logic of the first convolutional neural network path for extracting features from the nasal ultrasonic Doppler image includes: The nasal ultrasound Doppler image is input and deep denoised using a DRCNN module consisting of five dilated convolutional layers to obtain an ultrasound denoised feature map. The size of the five dilated convolutional layers in the DRCNN module is set to 3×3, and the dilation rates are set to 1, 2, 4, 2, and 1, respectively. The ultrasonic denoising feature map is extracted through the ResNeXt-50 network. The pixel values ​​of the ultrasonic denoising feature map are subjected to sliding convolution calculation through the convolution layer to obtain the weighted sum of the pixels corresponding to each window. The weighted sum data of the pixels corresponding to each window are nonlinearly mapped through the ReLU activation function to obtain the ultrasonic local feature map. The ultrasonic local feature map is subjected to maximum pooling calculation through a 2*2 pooling window in the pooling layer to obtain the downsampled feature map. The ultrasonic denoising feature map and the downsampled feature map are subjected to residual connection calculation through the residual block to obtain the ultrasonic Doppler output feature map.

[0025] In this embodiment, the near-infrared image feature extraction unit is used to extract features from the near-infrared image of the nose through the second convolutional neural network path to obtain a near-infrared output feature map. The processing logic of the second convolutional neural network path for extracting features from the near-infrared image of the nose includes: The nose near-infrared image is denoised by Gaussian filtering to obtain a denoised near-infrared image. The denoised near-infrared image is histogram equalized by the CLAHE algorithm to obtain an enhanced near-infrared image. The enhanced near-infrared image is preliminarily processed by a standard convolution layer of size 3x3, and a preliminary infrared feature map is obtained by batch normalization operation and ReLU function activation operation. The depth-wise separable convolution is used to convolve each channel in the preliminary infrared feature map to obtain a deep infrared feature map. The deep infrared feature map is subjected to global average pooling and Sigmoid function activation by the SE module, and is combined and calculated by a 1x1 point convolution kernel to obtain an abstract infrared feature map. The channel attention weight and spatial attention weight of the abstract infrared feature map are calculated by the CBAM attention module. The abstract infrared feature map is element-wise multiplied based on the channel attention weight and spatial attention weight to obtain the near-infrared output feature map.

[0026] In this embodiment, the feature fusion unit is used to perform three-phase feature fusion on the personal data feature vector, the ultrasound Doppler output feature map, and the near-infrared output feature map to obtain an output fusion feature. The processing logic for performing the three-phase feature fusion includes: Based on the channel number of the ultrasound Doppler output feature map and the near-infrared output feature map, the personal data feature vector is linearly transformed and expanded to obtain a two-dimensional tensor of personal data. A bidirectional feature pyramid network (BiFPN) is established, which assigns the two-dimensional tensor of personal data to the top level, the ultrasound Doppler output feature map to the middle level, and the near-infrared output feature map to the bottom level. According to the order from the bottom layer to the highest layer in BiFPN, each layer is upsampled in turn by bilinear interpolation to obtain the upsampled features corresponding to each layer. According to the order from the highest layer to the bottom layer in BiFPN, each layer is downsampled in turn by maximum pooling to obtain the downsampled features corresponding to each layer. The upsampled features and upsampled features of each layer are vector-concatenated to obtain the intermediate fusion features corresponding to each layer. The first convolution pooling module in the AOA attention is used to perform an attention calculation on the intermediate fusion features corresponding to each level to obtain the first fusion attention weight corresponding to each level. The first fusion attention weight corresponding to each level and the intermediate fusion features are multiplied element by element to obtain the first attention features corresponding to each level. The second convolution pooling module in the AOA attention is used to perform a secondary attention calculation on the first attention features corresponding to each level to obtain the second fusion attention weight corresponding to each level. The second fusion attention weight corresponding to each level and the first attention features are multiplied element by element to obtain the final fusion features corresponding to each level. The final fusion features corresponding to all levels are weighted summed and processed by maximum pooling to obtain the output fusion features.

[0027] In this embodiment, the classification prediction unit is used to predict and classify the output fusion features to obtain nasal allergy prevention results. The processing logic for performing the prediction and classification includes: The output fusion feature is input into the fully connected layer module of the two-layer structure, the output fusion feature is forward propagated through the first fully connected layer and the second fully connected layer module, the output fusion feature is multiplied by the first weight matrix in the first fully connected layer to obtain the first product, the first product is added by the first bias term, and a nonlinear transformation is performed through the ReLU activation function to obtain the transition output score; The number of neuron nodes in the first fully connected layer is N, and the number of neuron nodes in the second fully connected layer is 2; The output result of the first node is multiplied by the second weight matrix in the second fully connected layer to obtain a second product, the second product is added by the second bias term, and a nonlinear transformation is performed by the Softmax activation function to obtain the final output score result; Compare the two probability values ​​in the final output score result to obtain the larger probability value, and obtain the nasal allergy prevention judgment result based on the neuron node label corresponding to the larger probability value; The nasal allergy prevention judgment results include low prevention urgency and high prevention urgency.

[0028] The final output score result includes two probability values, which are output by two neuron nodes of the second fully connected layer.

[0029] In this embodiment, the terminal interaction unit is used to provide a display interface for the user and receive user operation instructions, allowing the user to adjust and set parameters of the data acquisition module, data analysis module and allergy prediction module; The data management unit is used to establish personal files of all testers through the database, and to store and update the output data of the data acquisition module, data analysis module and allergy prediction module.

[0030] Among them, in addition to collecting personal data, the present application also collects near-infrared images and ultrasonic Doppler images of the tester's nose for analysis, which can obtain information such as the condition of the nasal mucosa and the blood circulation condition of the nasal cavity, which is conducive to adding objective information to the patient's personal data for nasal allergy prevention analysis. When acquiring near-infrared images, a resting adjustment operation is added, which is conducive to reducing the error caused by the tester's body temperature. At the same time, the personal data is encoded to generate feature vectors, and allocated to the highest level of the bidirectional pyramid network during feature fusion, which is conducive to obtaining an abstract expression of personal data and fully maintaining the key position of personal data in the fusion process. Feature extraction of nasal ultrasonic Doppler images and nasal near-infrared images is performed through the first convolutional neural network path and the second convolutional neural network path respectively, which is conducive to targeted noise suppression and image feature extraction of the two image modal characteristics, and is conducive to obtaining abstract features that improve complementary information. Cross-level information interaction and feature fusion are achieved through the BiFPN architecture, and the AOA attention mechanism is introduced. Through two attention calculations, the key information in the fused features is accurately focused, which further improves the fusion effect, helps to improve the accuracy of subsequent classification predictions, and achieves early prevention of nasal allergies, thereby reducing the risk of nasal allergies.

[0031] Those skilled in the art will appreciate that embodiments of the present invention may provide methods, systems, or computer program products. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium may be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0032] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A nasal allergy prevention and management system, characterized in that: include: Data collection module, data analysis module, allergy prediction module and terminal management module; The data acquisition module includes a personal information acquisition unit, an ultrasound acquisition unit, and an infrared acquisition unit, which are used to collect statistics of the patient's information to obtain a personal data set, obtain an ultrasound Doppler image of the tester's nose, and collect a near-infrared image of the tester's nose through a near-infrared imager; The data analysis module includes a personal information feature extraction unit, an ultrasonic Doppler image feature extraction unit, and a near-infrared image feature extraction unit, which are used to perform quantization encoding processing on the tester's personal data set to obtain a personal data feature vector, perform feature extraction on the nasal ultrasonic Doppler image through a first convolutional neural network path to obtain an ultrasonic Doppler output feature map, and perform feature extraction on the nasal near-infrared image through a second convolutional neural network path to obtain a near-infrared output feature map; The allergy prediction module includes a feature fusion unit and a classification prediction unit, which are used to perform three-phase feature fusion on the personal data feature vector, the ultrasound Doppler output feature map, and the near-infrared output feature map to obtain output fusion features, and then predict and classify the output fusion features to obtain nasal allergy prevention results; The terminal management module includes a data management unit and a terminal interaction unit, which is used to provide a display interface for users and receive user operation instructions, establish personal files of all testers and store data.

2. The nasal allergy prevention and management system according to claim 1, characterized in that: The personal information collection unit is used to collect statistics on the patient's information to obtain a personal data set, which includes: age, gender, history of nasal allergies, family allergy history, allergen environment, sneezing, nasal congestion, itchy nose, runny nose, itchy eyes, red and swollen eyes, itchy throat, cough and chest tightness.

3. The nasal allergy prevention and management system according to claim 1, characterized in that: The ultrasonic acquisition unit is used to obtain an ultrasonic Doppler image of the test subject's nose. The ultrasonic diagnostic instrument is set based on preset two-dimensional ultrasonic scanning parameters and ultrasonic Doppler parameters. The ultrasonic linear array probe is used to perform a two-dimensional ultrasonic scan of the nasal mucosa and superficial blood vessels on both sides of the test subject's nose. The ultrasonic Doppler function is fixedly activated to obtain an ultrasonic Doppler image of the test subject's nose. The preset two-dimensional ultrasonic scanning parameters include: ultrasonic frequency and scanning depth, the ultrasonic frequency is set to 12 MHz, and the scanning depth is set to 0.5 cm; The ultrasound Doppler parameters include Doppler gain and velocity range, and the velocity range is set to 2 cm / s.

4. The nasal allergy prevention and management system according to claim 1, characterized in that: The infrared acquisition unit is used to acquire a near-infrared image of the test subject's nose using a near-infrared imager, perform a resting adjustment judgment on the test subject, and when the resting adjustment judgment result is suitable for acquisition, illuminate the test subject with a light source with preset near-infrared acquisition parameters, and vertically align the near-infrared imager lens with the test subject's nasal root to acquire a near-infrared image of the test subject's nose; The resting adaptation operation includes: having the test subject sit still for T time, measuring the test subject's pulse and body temperature, and determining that the test subject's pulse and body temperature are suitable for collection when the test subject's pulse is less than the resting pulse threshold and the test subject's body temperature is less than the resting body temperature threshold; The preset near-infrared acquisition parameters include: the wavelength of the near-infrared light source and the near-infrared imager are set to 850 nm, and the distance between the near-infrared imager lens and the tester's nose is set to 30 cm.

5. The nasal allergy prevention and management system according to claim 1, characterized in that: The personal information feature extraction unit is used to perform quantization coding processing on the tester's personal data set to obtain a personal data feature vector. The logic of the quantization coding processing includes: The age data in the personal data set is encoded by assigning 0 to the age group of 0-12 years old, 1 to the age group of 13-35 years old, 2 to the age group of 36-50 years old, and 3 to the age group of 50 years old and above to obtain the first quantitative code; The gender data in the personal data set is encoded by assigning 0 to male and 1 to female to obtain the second quantitative code; The nasal allergy history, family allergy history, and allergen environmental data in the personal data set are coded in sequence according to the value of 0 if they exist and 1 if they do not exist, to obtain the third quantitative code; The data of sneezing, nasal congestion, nasal itching, runny nose, itchy eyes, red and swollen eyes, itchy throat, cough, and chest tightness in the personal data set are coded in sequence according to the value of 0 for no symptoms, 1 for mild symptoms, and 2 for obvious symptoms to obtain the fourth quantitative code; The first quantization code, the second quantization code, the third quantization code and the fourth quantization code are sequentially connected in series to obtain a personal data feature vector.

6. The nasal allergy prevention and management system according to claim 1, characterized in that: The ultrasonic Doppler image feature extraction unit is used to extract features from the nasal ultrasonic Doppler image through the first convolutional neural network path to obtain an ultrasonic Doppler output feature map. The processing logic of the first convolutional neural network path for extracting features from the nasal ultrasonic Doppler image includes: The nasal ultrasound Doppler image is input and deep denoised using a DRCNN module consisting of five dilated convolutional layers to obtain an ultrasound denoised feature map. The size of the five dilated convolutional layers in the DRCNN module is set to 3×3, and the dilation rates are set to 1, 2, 4, 2, and 1, respectively. The ultrasonic denoising feature map is extracted through the ResNeXt-50 network. The pixel values ​​of the ultrasonic denoising feature map are subjected to sliding convolution calculation through the convolution layer to obtain the weighted sum of the pixels corresponding to each window. The weighted sum data of the pixels corresponding to each window are nonlinearly mapped through the ReLU activation function to obtain the ultrasonic local feature map. The ultrasonic local feature map is subjected to maximum pooling calculation through a 2*2 pooling window in the pooling layer to obtain the downsampled feature map. The ultrasonic denoising feature map and the downsampled feature map are subjected to residual connection calculation through the residual block to obtain the ultrasonic Doppler output feature map.

7. The nasal allergy prevention and management system according to claim 1, characterized in that: The near-infrared image feature extraction unit is used to extract features from the near-infrared image of the nose through the second convolutional neural network path to obtain a near-infrared output feature map. The processing logic of the second convolutional neural network path for extracting features from the near-infrared image of the nose includes: The nose near-infrared image is denoised by Gaussian filtering to obtain a denoised near-infrared image. The denoised near-infrared image is histogram equalized by the CLAHE algorithm to obtain an enhanced near-infrared image. The enhanced near-infrared image is preliminarily processed by a standard convolution layer of size 3x3, and a preliminary infrared feature map is obtained by batch normalization operation and ReLU function activation operation. The depth-wise separable convolution is used to convolve each channel in the preliminary infrared feature map to obtain a deep infrared feature map. The deep infrared feature map is subjected to global average pooling and Sigmoid function activation by the SE module, and is combined and calculated by a 1x1 point convolution kernel to obtain an abstract infrared feature map. The channel attention weight and spatial attention weight of the abstract infrared feature map are calculated by the CBAM attention module. The abstract infrared feature map is element-wise multiplied based on the channel attention weight and spatial attention weight to obtain the near-infrared output feature map.

8. The nasal allergy prevention and management system according to claim 1, characterized in that: The feature fusion unit is used to perform three-phase feature fusion on the personal data feature vector, the ultrasonic Doppler output feature map, and the near-infrared output feature map to obtain an output fusion feature. The processing logic for performing the three-phase feature fusion includes: Based on the channel number of the ultrasound Doppler output feature map and the near-infrared output feature map, the personal data feature vector is linearly transformed and expanded to obtain a two-dimensional tensor of personal data. A bidirectional feature pyramid network (BiFPN) is established, which assigns the two-dimensional tensor of personal data to the top level, the ultrasound Doppler output feature map to the middle level, and the near-infrared output feature map to the bottom level. According to the order from the bottom layer to the highest layer in BiFPN, each layer is upsampled in turn by bilinear interpolation to obtain the upsampled features corresponding to each layer. According to the order from the highest layer to the bottom layer in BiFPN, each layer is downsampled in turn by maximum pooling to obtain the downsampled features corresponding to each layer. The upsampled features and upsampled features of each layer are vector-concatenated to obtain the intermediate fusion features corresponding to each layer. The first convolution pooling module in the AOA attention is used to perform an attention calculation on the intermediate fusion features corresponding to each level to obtain the first fusion attention weight corresponding to each level. The first fusion attention weight corresponding to each level and the intermediate fusion features are multiplied element by element to obtain the first attention features corresponding to each level. The second convolution pooling module in the AOA attention is used to perform a secondary attention calculation on the first attention features corresponding to each level to obtain the second fusion attention weight corresponding to each level. The second fusion attention weight corresponding to each level and the first attention features are multiplied element by element to obtain the final fusion features corresponding to each level. The final fusion features corresponding to all levels are weighted summed and processed by maximum pooling to obtain the output fusion features.

9. The nasal allergy prevention and management system according to claim 1, characterized in that: The classification prediction unit is used to predict and classify the output fusion features to obtain nasal allergy prevention results. The processing logic for the prediction and classification includes: The output fusion feature is input into the fully connected layer module of the two-layer structure, the output fusion feature is forward propagated through the first fully connected layer and the second fully connected layer module, the output fusion feature is multiplied by the first weight matrix in the first fully connected layer to obtain the first product, the first product is added by the first bias term, and a nonlinear transformation is performed through the ReLU activation function to obtain the transition output score; The number of neuron nodes in the first fully connected layer is N, and the number of neuron nodes in the second fully connected layer is 2; The output result of the first node is multiplied by the second weight matrix in the second fully connected layer to obtain a second product, the second product is added by the second bias term, and a nonlinear transformation is performed by the Softmax activation function to obtain the final output score result; Compare the two probability values ​​in the final output score result to obtain the larger probability value, and obtain the nasal allergy prevention judgment result based on the neuron node label corresponding to the larger probability value; The nasal allergy prevention judgment results include low prevention urgency and high prevention urgency.

10. The nasal allergy prevention and management system according to claim 1, characterized in that: The terminal interaction unit is used to provide a display interface for the user and receive user operation instructions, allowing the user to adjust and set the parameters of the data acquisition module, data analysis module and allergy prediction module; The data management unit is used to establish personal files of all testers through the database, and to store and update the output data of the data acquisition module, data analysis module and allergy prediction module.

Images