Ultrasound-guided enteral nutrition nursing decision support method and device for critically ill patients

Abstract

The present application provides an ultrasound-guided enteral nutrition nursing decision support method and device for critically ill patients, which relates to the field of intelligent nursing. It first performs a nutritional assessment on critically ill patients to obtain nutritional assessment data including medical history data and laboratory test data, and at the same time collects the abdominal ultrasound image of the critically ill patient, and performs image enhancement processing on it to obtain an enhanced abdominal ultrasound image. Subsequently, the enhanced abdominal ultrasound image and the nutritional assessment data are input together into a nutritional nursing decision support engine based on a large model to obtain nutritional nursing decision support suggestions. In this way, by real-time monitoring and analysis of the patient's health changes and intestinal function status, it is conducive to providing personalized, accurate and timely nutritional nursing suggestions.

Description

Technical Field

[0001] The present application relates to the field of intelligent nursing, and more specifically, to an ultrasound-guided enteral nutrition nursing decision support method and device for critically ill patients. Background Art

[0002] In the field of critical care medicine, it is crucial to provide scientific and effective enteral nutrition support for critically ill patients. Reasonable enteral nutrition can not only provide the energy and nutrients needed by the body and maintain the intestinal mucosal barrier function, but also reduce the occurrence of complications such as infection, which is of great significance to improving the prognosis of patients.

[0003] However, there are many urgent problems to be solved in the decision-making of enteral nutrition care for critically ill patients. On the one hand, traditional nutritional care usually relies on standardized or universal nutritional guidelines. Although these guidelines provide basic guidelines, they fail to fully consider the individual differences of each patient and make it difficult to achieve precise care. On the other hand, traditional methods are mainly based on medical history data and laboratory test data. Although these indicators can reflect the overall nutritional status of patients, they lack direct observation of specific intestinal functions such as intestinal peristalsis and intestinal wall thickness, resulting in incomplete nutritional care decisions and inability to accurately reflect the functional status of patients' organs.

[0004] Therefore, an optimized enteral nutrition nursing decision support program for critically ill patients is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The present application provides an ultrasound-guided enteral nutrition nursing decision support method and device for critically ill patients.

[0006] According to one aspect of the present application, an ultrasound-guided enteral nutrition nursing decision support device for critically ill patients is provided, comprising: a nutrition assessment data acquisition module, used for performing nutrition assessment on a critically ill patient object to obtain nutrition assessment data, wherein the nutrition assessment data includes medical history data and laboratory test data; an abdominal ultrasound image data acquisition module, used for obtaining an abdominal ultrasound image of the critically ill patient object; an abdominal ultrasound image enhancement module, used for performing image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image; a nutrition nursing decision support suggestion generation module, used for inputting the enhanced abdominal ultrasound image and the nutrition assessment data into a nutrition nursing decision support engine based on a large model to obtain a nutrition nursing decision support suggestion;

[0007] Wherein, the abdominal ultrasound image enhancement module includes:

[0008] An abdominal ultrasound image shallow and deep feature extraction unit, used to extract image shallow features and deep features from the abdominal ultrasound image to obtain an abdominal ultrasound shallow image feature map and an abdominal ultrasound deep image feature map;

[0009] An abdominal ultrasound image deep feature selection unit, used for performing local activity-aware feature selection on the abdominal ultrasound deep image feature map to obtain a sparse abdominal ultrasound deep image feature map;

[0010] An abdominal ultrasound image depth-shallowness feature joint encoding unit is used to perform depth-shallowness semantic-guided joint perception on the sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map to obtain an abdominal ultrasound depth-shallowness joint perception encoding feature map;

[0011] The enhanced abdominal ultrasound image generation unit is used to obtain the enhanced abdominal ultrasound image based on the abdominal ultrasound depth-shallowness joint perception coding feature map.

[0012] Furthermore, the abdominal ultrasound image shallow and deep feature extraction unit is used to: use a feature extractor based on a hollow pyramid model to extract image shallow features and deep features from the abdominal ultrasound image to obtain the abdominal ultrasound shallow image feature map and the abdominal ultrasound deep image feature map.

[0013] Furthermore, the abdominal ultrasound image deep feature selection unit comprises:

[0014] An abdominal ultrasound deep layer feature decoupling and flattening subunit, used for performing feature decoupling and feature flattening on the abdominal ultrasound deep layer image feature map to obtain a set of abdominal ultrasound deep layer local feature vectors;

[0015] An abdominal ultrasound deep layer local feature descending order arrangement subunit, used to arrange the set of abdominal ultrasound deep layer local feature vectors in descending order to obtain a descending sequence of abdominal ultrasound deep layer local feature vectors;

[0016] The sparse abdominal ultrasound deep layer image feature map generating subunit is used to perform feature selection and feature shape reshaping on the descending sequence of the abdominal ultrasound deep layer local feature vectors to obtain the sparse abdominal ultrasound deep layer image feature map.

[0017] Furthermore, the abdominal ultrasound deep layer local feature descending order arrangement subunit is used for:

[0018] Performing importance measurement on each abdominal ultrasound deep layer local feature vector in the set of abdominal ultrasound deep layer local feature vectors to obtain a set of abdominal ultrasound deep layer local feature importance score values;

[0019] Based on the set of abdominal ultrasound deep layer local feature importance score values, the set of abdominal ultrasound deep layer local feature vectors is arranged in descending order to obtain a descending sequence of the abdominal ultrasound deep layer local feature vectors.

[0020] Furthermore, the sparse abdominal ultrasound deep layer image feature map generating subunit is used to:

[0021] Based on the neighborhood features of each abdominal ultrasound deep layer local feature vector in the descending sequence of the abdominal ultrasound deep layer local feature vectors, calculating the feature neighborhood activity of each abdominal ultrasound deep layer local feature vector to obtain a sequence of abdominal ultrasound deep layer local feature neighborhood activity;

[0022] Based on the sequence of neighborhood activity of the deep abdominal ultrasound local feature, feature selection is performed on the descending sequence of the deep abdominal ultrasound local feature vectors to obtain a descending sequence of the selected deep abdominal ultrasound local feature vectors;

[0023] The descending sequence of the selected abdominal ultrasound deep layer local feature vectors is reshaped to obtain the sparse abdominal ultrasound deep layer image feature map.

[0024] Furthermore, the abdominal ultrasound image depth feature joint encoding unit comprises:

[0025] a sparse abdominal ultrasound deep layer image feature map upsampling subunit, configured to upsample the sparse abdominal ultrasound deep layer image feature map to obtain an upsampled sparse abdominal ultrasound deep layer image feature map, wherein the upsampled sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map have the same size;

[0026] An upsampled sparse abdominal ultrasound deep layer feature mapping modulation subunit is used to map and modulate the upsampled sparse abdominal ultrasound deep layer feature map based on the semantic information field between the upsampled sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map to obtain a sparse abdominal ultrasound deep layer image field modulation feature map;

[0027] The abdominal ultrasound depth-shallowness joint perception coding feature generation subunit is used to perform multiple attention feature perception on the abdominal ultrasound shallow image feature map and the sparse abdominal ultrasound deep image field modulation feature map to obtain the abdominal ultrasound depth-shallowness joint perception coding feature map.

[0028] Furthermore, the upsampling and sparse abdominal ultrasound deep feature mapping modulation subunit is used to:

[0029] The upsampled and sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map are feature-connected and then input into the semantic information field predictor of gated convolution to obtain the abdominal ultrasound feature semantic information field as the semantic information field;

[0030] The upsampled sparse abdominal ultrasound deep image feature map is mapped to the abdominal ultrasound feature semantic information field to obtain the sparse abdominal ultrasound deep image field modulation feature map.

[0031] Furthermore, the enhanced abdominal ultrasound image generation unit is used to: input the abdominal ultrasound depth-shallowness joint perceptual coding feature map into an image enhancement module based on a generative adversarial network to obtain the enhanced abdominal ultrasound image.

[0032] According to another aspect of the present application, an ultrasound-guided enteral nutrition nursing decision support method for critically ill patients is provided, comprising: performing a nutrition assessment on a critically ill patient to obtain nutrition assessment data, wherein the nutrition assessment data includes medical history data and laboratory test data; obtaining an abdominal ultrasound image of the critically ill patient; performing image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image; inputting the enhanced abdominal ultrasound image and the nutrition assessment data into a nutrition nursing decision support engine based on a large model to obtain nutrition nursing decision support recommendations;

[0033] The step of performing image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image includes:

[0034] Extracting image superficial features and deep features from the abdominal ultrasound image to obtain an abdominal ultrasound superficial image feature map and an abdominal ultrasound deep image feature map;

[0035] Performing local activity-aware feature selection on the abdominal ultrasound deep image feature map to obtain a sparse abdominal ultrasound deep image feature map;

[0036] Performing deep-shallow semantic-guided joint perception on the sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map to obtain an abdominal ultrasound deep-shallow joint perception coding feature map;

[0037] Based on the abdominal ultrasound depth and shallowness joint perception coding feature map, the enhanced abdominal ultrasound image is obtained.

[0038] Furthermore, shallow features and deep features of the image are extracted from the abdominal ultrasound image to obtain an abdominal ultrasound shallow image feature map and an abdominal ultrasound deep image feature map, including: using a feature extractor based on a hollow pyramid model to extract shallow features and deep features of the image from the abdominal ultrasound image to obtain the abdominal ultrasound shallow image feature map and the abdominal ultrasound deep image feature map.

[0039] Compared with the prior art, the ultrasound-guided enteral nutrition care decision support method and device for critically ill patients provided by the present application first implements a nutrition assessment on the critically ill patient to obtain nutrition assessment data including medical history data and laboratory test data, and at the same time collects the abdominal ultrasound image of the critically ill patient, and performs image enhancement processing on it to obtain an enhanced abdominal ultrasound image, and then inputs the enhanced abdominal ultrasound image and the nutrition assessment data together into a nutrition care decision support engine based on a large model to obtain nutrition care decision support recommendations. In this way, by real-time monitoring and analysis of the patient's health changes and intestinal function status, it is conducive to providing personalized, accurate and timely nutrition care recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0041] Figure 1 It is a block diagram of an ultrasound-guided enteral nutrition care decision support device for critically ill patients according to an embodiment of the present application.

[0042] Figure 2 The present invention is a block diagram of an abdominal ultrasound image enhancement module in an ultrasound-guided enteral nutrition care decision support device for critically ill patients according to an embodiment of the present application.

[0043] Figure 3 The present invention is a schematic diagram of data flow of an abdominal ultrasound image enhancement module in an ultrasound-guided enteral nutrition care decision support device for critically ill patients according to an embodiment of the present application.

[0044] Figure 4 The present invention is a flowchart of an ultrasound-guided enteral nutrition care decision support method for critically ill patients according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0046] In the field of critical care medicine, the importance of implementing scientific and effective enteral nutrition support for critically ill patients is self-evident. Reasonable enteral nutrition can not only supply the body with necessary energy and nutrients and maintain the intestinal mucosal barrier function, but also reduce the incidence of complications such as infection, which is of great significance for improving the prognosis of patients.

[0047] However, there are currently a series of urgent problems to be solved in the decision-making of enteral nutrition care for critically ill patients. On the one hand, traditional nutritional care often relies on standardized or universal nutritional guidelines. Although these guidelines provide basic guiding principles, they fail to fully consider the individual differences of each patient, making precise care difficult to achieve. On the other hand, traditional methods are mainly based on medical history data and laboratory test data. Although these indicators can reflect the patient's overall nutritional status, they lack direct observation of specific intestinal functions such as intestinal motility and intestinal wall thickness. This makes nutritional care decisions not comprehensive enough and unable to accurately reflect the patient's organ function status.

[0048] It should be understood that ultrasound is a sound wave with a frequency higher than the upper limit of human hearing (20,000 Hz). In the medical field, ultrasound examination mainly uses the reflection principle of ultrasound. The images obtained by ultrasound can observe local characteristics such as intestinal peristalsis and intestinal wall thickness, which are very important for evaluating intestinal function and the feasibility of enteral nutrition.

[0049] Based on this, this application proposes an ultrasound-guided enteral nutrition care decision support device for critically ill patients, which can monitor and analyze the patient's health changes and intestinal function status in real time, thereby providing personalized, accurate and timely nutrition care recommendations. Specifically, Figure 1 FIG. 1 is a block diagram of an ultrasound-guided enteral nutrition care decision support device for critically ill patients according to an embodiment of the present application. Figure 1 As shown, in the ultrasound-guided enteral nutrition care decision support device 100 for critically ill patients, it includes: a nutrition assessment data acquisition module 110, used to perform nutrition assessment on critically ill patients to obtain nutrition assessment data, and the nutrition assessment data includes medical history data and laboratory examination data; an abdominal ultrasound image data acquisition module 120, used to obtain an abdominal ultrasound image of the critically ill patient; an abdominal ultrasound image enhancement module 130, used to perform image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image; and a nutrition care decision support recommendation generation module 140, used to input the enhanced abdominal ultrasound image and the nutrition assessment data into a nutrition care decision support engine based on a large model to obtain nutrition care decision support recommendations.

[0050] In an embodiment of the present application, the nutrition assessment data acquisition module 110 is used to perform nutrition assessment on a critically ill patient to obtain nutrition assessment data, and the nutrition assessment data includes medical history data and laboratory test data. It should be understood that the medical history data specifically includes basic disease information, drug use history information, allergy history information, etc.; laboratory test data includes blood biochemical index information, blood routine index information, nutritional metabolism index, etc. In detail, the basic disease information includes chronic diseases (such as hypertension, diabetes, coronary heart disease, etc.) and acute diseases (such as infectious diseases, trauma, surgery, etc.) that the patient has suffered in the past. These diseases may affect the patient's metabolic function, organ function, and nutritional needs and absorption capacity. For example, diabetic patients may have glucose metabolism disorders and need to pay special attention to carbohydrate intake and blood sugar control; patients with chronic kidney disease may have restrictions on protein and electrolyte intake. Drug use history information is information such as the drugs used by the patient, their dosage, and medication time. Certain drugs may affect appetite, absorption or metabolism of nutrients. Allergy history refers to the patient's allergies to food, drugs, etc. By understanding the allergy history, it is possible to avoid the use of nutritional preparations that may cause allergic reactions in patients during nutritional support, thereby ensuring the safety of patients. Blood biochemical indicators specifically include protein-related indicators such as serum albumin, prealbumin, transferrin, blood sugar indicators, blood lipid indicators, and electrolyte indicators. A decrease in serum albumin levels may indicate protein malnutrition, while prealbumin and transferrin have a short half-life and are more sensitive to changes in nutritional status, which can more timely reflect the patient's recent protein intake and metabolism. Blood sugar indicators help to assess the patient's sugar metabolism and provide a basis for formulating carbohydrate supply plans. For example, for patients with poor blood sugar control, it is necessary to adjust the sugar content and use of nutritional preparations to avoid excessive blood sugar fluctuations. Blood lipid indicators can reflect the patient's lipid metabolism and are instructive for the supply of fat nutrients. Electrolyte indicators such as sodium, potassium, calcium, magnesium, phosphorus, etc. are necessary to maintain normal physiological functions of the human body. Their abnormalities may affect the functions of the heart, neuromuscular and other systems, and the balance of these electrolytes needs to be ensured. Hemoglobin and hematocrit in routine blood tests can reflect whether the patient has anemia, which can affect tissue oxygen supply. Nutritional support may require the supplementation of iron, folic acid, vitamin B12 and other nutrients that promote hematopoiesis. Nutritional metabolic indicators such as serum vitamin levels (such as vitamin D, B vitamins, etc.), trace elements (such as zinc, copper, selenium, etc.), and nitrogen balance can help understand whether the patient has corresponding nutrient deficiencies or metabolic disorders, so as to provide targeted supplements. In general, nutritional assessment data provides a basic basis for the final nutritional care decision support recommendations. Disease information and drug use history in medical history data can allow the model to understand the patient's basic condition and predict possible nutritional problems or special needs for nutritional support.For example, if a patient has a gastrointestinal disease, a special enteral nutrition method may be required, or the composition of the nutritional preparation may be adjusted to accommodate the impaired gastrointestinal function. Laboratory test data can more accurately quantify the patient's current nutritional status, and based on these data, it can be determined whether the patient is in a state of malnutrition or overnutrition, thereby determining the intensity of nutritional support and the direction of adjustment.

[0051] In an embodiment of the present application, the abdominal ultrasound image data acquisition module 120 is used to obtain the abdominal ultrasound image of the critically ill patient object. It should be understood that the abdominal ultrasound image of the critically ill patient object contains intestinal morphology and structural information such as intestinal wall thickness and intestinal mucosal integrity, as well as intestinal motility information, mesenteric vascular information, etc. In detail, if the abdominal ultrasound image shows that there are serious organic lesions in the intestine, such as intestinal necrosis and extensive intestinal adhesions that cause intestinal obstruction that cannot be relieved by conservative treatment, the intestine cannot digest and absorb nutrients normally at this time, and the decision-making recommendation may tend to choose parenteral nutrition to ensure that the patient can obtain sufficient nutritional support and avoid increasing the burden on the intestine. If the intestinal structure is basically normal, but there is a slight weakening of motility or mild edema of the intestinal wall, enteral nutrition can be considered, but the changes in intestinal function need to be closely monitored. In this case, it may be recommended to use a slow infusion method or use drugs that promote gastrointestinal motility to ensure the smooth implementation of enteral nutrition. When abdominal ultrasound indicates that the digestion and absorption function of the intestine is impaired, such as thinning of the intestinal wall and significantly weakened peristalsis, nutritional preparations that are easy to digest and absorb can be selected, such as elemental preparations containing short peptides or amino acids, to reduce the content of fat and dietary fiber to reduce the burden on the intestine. If the intestinal mucosa is found to be damaged, such as disordered intestinal wall structure, ingredients that help repair the intestinal mucosa, such as glutamine, can be added to the nutritional preparation. In other words, based on the abdominal ultrasound images of critically ill patients, the most appropriate nutritional formula and route of administration can be tailored for each patient.

[0052] In an embodiment of the present application, the abdominal ultrasound image enhancement module 130 is used to perform image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image. It should be understood that the original abdominal ultrasound image may have insufficient contrast and may be interfered by various factors to generate noise. Therefore, through image enhancement, the contrast between the target area (such as the intestine) and the background area can be increased, the interference of noise can be reduced, and the key structure can be made clearer to highlight the relevant features of intestinal function. However, traditional image enhancement methods are mostly based on fixed rules and algorithms, which are difficult to adapt to individual differences and complex situations of ultrasound images, and have limitations in improving image quality and highlighting key features to meet the needs of accurate decision-making in enteral nutrition care.

[0053] Correspondingly, in the abdominal ultrasound image enhancement module, the technical concept of the present application is to use machine vision-based image recognition processing and enhancement technology to extract deep and shallow features of the abdominal ultrasound image, and then perform feature sparse processing of local activity on the abdominal ultrasound deep image features, so as to achieve enhanced abdominal ultrasound images based on the deep and shallow semantic joint representation between the sparse abdominal ultrasound deep image features and the abdominal ultrasound shallow image features. By extracting and processing the deep and shallow features of ultrasound images, the present application can adapt to the individual differences and complex situations of ultrasound images of different patients, thereby significantly improving image quality, highlighting key features, and better meeting the needs of accurate decision-making in enteral nutrition care.

[0054] Specifically, Figure 2 The present invention is a block diagram of an abdominal ultrasound image enhancement module in an ultrasound-guided enteral nutrition care decision support device for critically ill patients according to an embodiment of the present application. Figure 3 Schematic diagram of data flow of the abdominal ultrasound image enhancement module in the ultrasound-guided enteral nutrition nursing decision support device for critically ill patients according to an embodiment of the present application. Figure 2 and Figure 3 As shown, the abdominal ultrasound image enhancement module 130 includes: an abdominal ultrasound image shallow and deep feature extraction unit 131, which is used to extract image shallow features and deep features from the abdominal ultrasound image to obtain an abdominal ultrasound shallow image feature map and an abdominal ultrasound deep image feature map; an abdominal ultrasound image deep feature selection unit 132, which is used to perform local activity perception feature selection on the abdominal ultrasound deep image feature map to obtain a sparse abdominal ultrasound deep image feature map; an abdominal ultrasound image deep and shallow feature joint encoding unit 133, which is used to perform deep and shallow semantic-guided joint perception on the sparse abdominal ultrasound deep image feature map and the abdominal ultrasound shallow image feature map to obtain an abdominal ultrasound deep and shallow joint perception coding feature map; an enhanced abdominal ultrasound image generation unit 134, which is used to obtain the enhanced abdominal ultrasound image based on the abdominal ultrasound deep and shallow joint perception coding feature map.

[0055] In an embodiment of the present application, the shallow and deep features extraction unit 131 of the abdominal ultrasound image is used to extract shallow features and deep features of the image from the abdominal ultrasound image to obtain a shallow abdominal ultrasound image feature map and a deep abdominal ultrasound image feature map. Specifically, in an embodiment of the present application, the shallow and deep features extraction unit 131 of the abdominal ultrasound image is used to: use a feature extractor based on a hollow pyramid model to extract shallow features and deep features of the image from the abdominal ultrasound image to obtain the shallow abdominal ultrasound image feature map and the deep abdominal ultrasound image feature map. Accordingly, considering that the abdominal ultrasound image contains tissue structures and pathological features of different sizes, for example, the detailed features such as the diameter of the intestine and the thickness of the intestinal wall are small in size, while the size of the entire liver, spleen and other organs is large, it is very important to preliminarily identify and locate the structure in the image. Based on this, the present application extracts shallow features and deep features of the image from the abdominal ultrasound image by using a feature extractor based on a hollow pyramid model to obtain the shallow abdominal ultrasound image feature map and the deep abdominal ultrasound image feature map. It should be understandable that the dilated pyramid model can obtain features of different receptive field sizes at the same level by setting different dilation rates at different convolutional layers. This allows the model to simultaneously capture information from tiny details to larger structures in abdominal ultrasound images. For example, a convolution operation with a smaller dilation rate can focus on details such as the subtle texture of the intestinal wall, while a convolution operation with a larger dilation rate can grasp the overall morphology and positional relationship of the organ, thereby better representing the richness and accuracy of image features.

[0056] In an embodiment of the present application, the abdominal ultrasound image deep feature selection unit 132 is used to perform local activity-aware feature selection on the abdominal ultrasound deep image feature map to obtain a sparse abdominal ultrasound deep image feature map. Specifically, in an embodiment of the present application, the abdominal ultrasound image deep feature selection unit 132 includes: an abdominal ultrasound deep feature decoupling and flattening subunit, used to perform feature decoupling and feature flattening on the abdominal ultrasound deep image feature map to obtain a set of abdominal ultrasound deep local feature vectors; an abdominal ultrasound deep local feature descending subunit, used to perform descending order on the set of abdominal ultrasound deep local feature vectors to obtain a descending sequence of abdominal ultrasound deep local feature vectors; a sparse abdominal ultrasound deep image feature map generation subunit, used to perform feature selection and feature shape reshaping on the descending sequence of the abdominal ultrasound deep local feature vectors to obtain the sparse abdominal ultrasound deep image feature map.

[0057] It should be understood that the abdominal ultrasound deep image feature map contains a lot of information, many of which may be redundant or have little support for subsequent nutritional care decisions. For example, in the deep feature map, there may be some features that are related to background noise or are not directly related to the description of the key state of organs and tissues. These redundant features will increase the amount of data and the computational burden, and may also interfere with the extraction and analysis of key information. Based on this, the present application obtains a sparse abdominal ultrasound deep image feature map by performing local activity-aware feature selection on the abdominal ultrasound deep image feature map. That is, this method uses the feature neighborhood activity to dynamically evaluate the importance and interaction of features in their surrounding areas (i.e., semantic neighborhoods), thereby highlighting key features and reducing the dimensionality of features, so that subsequent models based on these features (such as models for image enhancement or nutritional care decisions) can be more efficient.

[0058] Specifically, it is first necessary to perform feature decoupling and feature flattening on the abdominal ultrasound deep layer image feature map to obtain a set of abdominal ultrasound deep layer local feature vectors. The above process can be expressed as:

[0059] ;

[0060] ;

[0061] in, is the abdominal ultrasound deep layer image characteristic diagram, is a feature decoupling operation, , , and They are the first, second, and third local feature matrices of the deep abdominal ultrasound image. and The local feature matrix of deep abdominal ultrasound images, is the feature flattening operation, , , and They are the first, second, and third local feature vectors in the set of deep abdominal ultrasound. and The local feature vectors of deep abdominal ultrasound.

[0062] It should be understood that the features contained in the abdominal ultrasound deep image feature map are relatively complex, and are the result of multiple information intertwined. For example, multiple features such as the position, morphology, intestinal wall thickness, and the relationship between the surrounding tissues of the intestine are mixed in this abdominal ultrasound deep image feature map. These complex features are directly used for subsequent analysis, and it is difficult for the model to accurately extract key information. In order to split these complex mixed features into relatively independent local features, so as to facilitate a more detailed analysis of the role of each local feature in the subsequent analysis, the present application needs to perform feature decoupling on the abdominal ultrasound deep image feature map to obtain a set of local feature matrices of the abdominal ultrasound deep image, and feature flatten the set of local feature matrices of the abdominal ultrasound deep image to obtain a set of local feature vectors of the abdominal ultrasound deep layer, so as to ensure that the elements in each local feature vector of the abdominal ultrasound deep layer represent different but related feature dimensions. Moreover, the generation of local independent features of the abdominal ultrasound deep layer is guaranteed by feature decoupling and feature flattening, which can provide materials for subsequent feature selection.

[0063] Specifically, in the embodiment of the present application, the abdominal ultrasound deep layer local feature descending order arrangement subunit is used to: perform importance measurement on each abdominal ultrasound deep layer local feature vector in the set of abdominal ultrasound deep layer local feature vectors to obtain a set of abdominal ultrasound deep layer local feature importance score values. This process can be expressed as:

[0064] ;

[0065] ;

[0066] in, is the first local feature vector in the set of deep abdominal ultrasound The deep local feature vectors of abdominal ultrasound, and They are The corresponding weight matrix and bias vector, is matrix multiplication, is the weight score vector, is a collection of importance scores of deep local features of abdominal ultrasound. , , and They are the first, second, and third in the set of importance scores of deep local features of abdominal ultrasound. and Importance score of deep local features of abdominal ultrasound;

[0067] Based on the set of importance score values ​​of the deep abdominal ultrasound local features, the set of the deep abdominal ultrasound local feature vectors is arranged in descending order to obtain a descending sequence of the deep abdominal ultrasound local feature vectors. This process can be expressed as:

[0068] ;

[0069] in, Based on The set of abdominal ultrasound deep layer local feature vectors is arranged in descending order, , , and They are the first, second, and third in the descending sequence of the deep local feature vectors of abdominal ultrasound. and The local feature vectors of deep abdominal ultrasound.

[0070] It should be understood that after obtaining the set of deep local feature vectors of abdominal ultrasound, each vector contains many feature dimensions, but not all features are equally important for the final nutritional care decision support. For example, some features may only reflect some accidental factors in the image acquisition process, and are of no substantial help in judging the actual intestinal condition of the patient and formulating a nutritional care plan. Through importance measurement, the value of deep local feature vectors of abdominal ultrasound in the overall task (judging the functional state of the intestine to assist in making nutritional care decisions) can be clarified, and key features can be distinguished from secondary features, which is conducive to optimizing the use of feature data, enabling the model to focus on core information and improve processing efficiency.

[0071] Accordingly, considering that in the set of deep local feature vectors of abdominal ultrasound, the importance of each feature vector to the enteral nutrition nursing decision support task for critically ill patients varies. By calculating the set of deep local feature importance score values ​​of abdominal ultrasound, the set of deep local feature vectors of abdominal ultrasound is arranged in descending order, so that the most important feature vectors can be intuitively placed at the front of the sequence, so as to quickly identify the features that are most critical to the task objectives. For example, when judging the digestive and absorptive function of the intestine, feature vectors related to intestinal peristalsis and intestinal wall thickness may have higher importance scores, and they can be easily distinguished from other relatively minor feature vectors after sorting. It is worth noting that descending order itself does not change the content of the deep local features of abdominal ultrasound, but provides a clear view of the relative importance of features. This step can not only simplify the feature selection process, but also enable subsequent computing resources to be focused on the most important features.

[0072] Specifically, in the embodiment of the present application, the sparse abdominal ultrasound deep layer image feature map generating subunit is used to: based on the neighborhood features of each abdominal ultrasound deep layer local feature vector in the descending sequence of the abdominal ultrasound deep layer local feature vector, calculate the feature neighborhood activity of each abdominal ultrasound deep layer local feature vector to obtain a sequence of abdominal ultrasound deep layer local feature neighborhood activity. The process can be expressed as:

[0073] ;

[0074] in, yes Middle The eigenvalues ​​at the positions, yes The number of eigenvalues ​​in , and They are and The corresponding weighted average eigenvalue of deep abdominal ultrasound features, yes The corresponding local feature neighborhood activity of deep abdominal ultrasound;

[0075] Based on the sequence of neighborhood activity of the deep abdominal ultrasound local feature, feature selection is performed on the descending sequence of the deep abdominal ultrasound local feature vectors to obtain a descending sequence of the selected deep abdominal ultrasound local feature vectors. The process can be expressed as:

[0076] ;

[0077] ;

[0078] in, For Perform feature selection, is the preset threshold, , , and The first, second, and third local feature vectors of the deep layer of posterior abdominal ultrasound are selected in descending order. and A local feature vector of deep layer of posterior abdominal ultrasound was selected;

[0079] The descending sequence of the selected abdominal ultrasound deep layer local feature vectors is reshaped to obtain the sparse abdominal ultrasound deep layer image feature map. This process can be expressed as:

[0080] ;

[0081] in, , , and The first, second, and third local feature vectors of the deep layer of posterior abdominal ultrasound are selected in descending order. and Select the deep local feature vector of abdominal ultrasound. For the reshape operation, It is the thinned abdominal ultrasound deep image feature map.

[0082] It should be understood that when analyzing the deep local features of abdominal ultrasound, the importance of a single feature vector is critical, but the relationship between features cannot be ignored. For example, when judging intestinal lesions, a local feature vector represents the texture feature of a certain area of ​​the intestinal wall, and its surrounding feature vectors may reflect the relationship between the area and the adjacent tissues. The synergy between these features may be ignored based only on the importance of a single feature vector. The present application calculates the feature neighborhood activity of each deep local feature vector of abdominal ultrasound based on the neighborhood features of each deep local feature vector of abdominal ultrasound in the descending sequence of the deep local feature vector of abdominal ultrasound, and can take into account the relationship between each deep local feature of abdominal ultrasound and its neighboring features, and comprehensively analyze the interaction between features to more accurately capture the information contained in the abdominal ultrasound image. That is, the calculated neighborhood activity of the deep local feature of abdominal ultrasound can reveal hidden information that is difficult to detect from the importance of a single feature, so as to more comprehensively reflect the actual situation of the intestine and lay the foundation for accurate nutritional care decisions.

[0083] Then, based on the sequence of neighborhood activity of the deep local features of the abdominal ultrasound, feature selection is performed on the descending sequence of the deep local feature vectors of the abdominal ultrasound to obtain a descending sequence of the selected deep local feature vectors of the abdominal ultrasound. That is, by using the sequence of neighborhood activity of the deep local features of the abdominal ultrasound, the most valuable deep local feature vectors of the abdominal ultrasound can be further screened out. At this stage, feature selection no longer depends solely on the importance score of individual features, but comprehensively considers the relationship between the deep local features of the abdominal ultrasound and their neighboring features. The deep local features of the abdominal ultrasound selected in this way can not only well represent the information in the original abdominal ultrasound image, but also better capture the key content in the image. The selected feature sequence will be more compact and more expressive, which is conducive to improving model performance and interpretability. In the specific implementation process, the feature selection strategy can be flexibly adjusted according to the specific application, such as setting a fixed activity threshold, dynamically selecting a certain percentage of the highest activity features, or adopting more complex combination rules.

[0084] Finally, the descending sequence of the selected deep abdominal ultrasound local feature vectors is reshaped to obtain the sparse deep abdominal ultrasound image feature map. That is, the selected deep abdominal ultrasound local feature vectors are reorganized back into the spatial structure of the original feature map to form a sparse deep abdominal ultrasound image feature map. The feature reshaping process involves reorganizing the one-dimensional deep abdominal ultrasound local feature vectors into a multi-dimensional feature map format, which ensures consistency with the original input format and helps maintain the consistency and compatibility of the model architecture.

[0085] In an embodiment of the present application, the abdominal ultrasound image depth and shallow feature joint encoding unit 133 is used to perform depth and shallow semantic-guided joint perception on the sparse abdominal ultrasound deep image feature map and the abdominal ultrasound shallow image feature map to obtain an abdominal ultrasound depth and shallow joint perception encoding feature map. Specifically, in an embodiment of the present application, the abdominal ultrasound image depth feature joint encoding unit 133 includes: a sparse abdominal ultrasound deep image feature map upsampling subunit, which is used to upsample the sparse abdominal ultrasound deep image feature map to obtain an upsampled sparse abdominal ultrasound deep image feature map, wherein the upsampled sparse abdominal ultrasound deep image feature map and the abdominal ultrasound shallow image feature map have the same size; an upsampled sparse abdominal ultrasound deep feature mapping modulation subunit, which is used to map and modulate the upsampled sparse abdominal ultrasound deep image feature map based on the semantic information field between the upsampled sparse abdominal ultrasound deep image feature map and the abdominal ultrasound shallow image feature map to obtain a sparse abdominal ultrasound deep image field modulation feature map; an abdominal ultrasound depth joint perception coding feature generation subunit, which is used to perform multiple attention feature perception on the abdominal ultrasound shallow image feature map and the sparse abdominal ultrasound deep image field modulation feature map to obtain the abdominal ultrasound depth joint perception coding feature map.

[0086] It should be understood that the feature map of the shallow abdominal ultrasound image is rich in detail information, such as the edges and textures of organs and other low-level semantics; while the sparse deep abdominal ultrasound image feature map contains more abstract, high-level semantics, such as the overall morphology of organs, the relationship between tissues, etc. In order to enable these two different levels of semantics to interact effectively in the process of joint feature perception and fully present the image semantics, the present application fuses the sparse deep abdominal ultrasound image feature map and the shallow abdominal ultrasound image feature map through a joint perception mechanism guided by deep and shallow semantics to obtain an abdominal ultrasound deep and shallow joint perception coding feature map. In particular, this mechanism can build semantic connections between different features, guide feature fusion based on semantic information, and ensure that the fusion of shallow and deep features can better reflect the real anatomical structure and physiological state, thereby improving the fusion effect.

[0087] In detail, the sparse abdominal ultrasound deep layer image feature map is first upsampled to obtain an upsampled sparse abdominal ultrasound deep layer image feature map, wherein the upsampled sparse abdominal ultrasound deep layer image feature map has the same size as the abdominal ultrasound shallow layer image feature map. The above process can be expressed as:

[0088] ;

[0089] in, is the sparse abdominal ultrasound deep layer image feature map, is the upsampling operation, It is the upsampled and sparse abdominal ultrasound deep image feature map.

[0090] It should be understood that upsampling the sparse abdominal ultrasound deep image feature map can not only increase the spatial resolution of this feature map, but more importantly, restore some of the detail information lost in the deep network. Here, by using methods such as transposed convolution (deconvolution) or interpolation, fine-grained abdominal ultrasound deep structure features can be reconstructed. Specifically, the edge information of the object can be restored to a certain extent through transposed convolution, and these edges are critical for accurately defining the boundaries of organs; the interpolation method can reasonably fill in the missing details based on the surrounding feature information, so that important visual clues in the image can be retained. These restored detail information can more comprehensively reflect the true morphology and physiological state of the structure in the abdominal ultrasound image, thereby improving the accuracy of the entire image feature extraction and analysis, and providing a more reliable basis for the decision support recommendations for enteral nutrition care for critically ill patients generated based on this.

[0091] Specifically, in an embodiment of the present application, the upsampled sparse abdominal ultrasound deep layer feature map modulation subunit is used to: feature connect the upsampled sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map, and then input them into the gated convolution semantic information field predictor to obtain the abdominal ultrasound feature semantic information field as the semantic information field. The process can be expressed as:

[0092] ;

[0093] in, is the upsampled sparse abdominal ultrasound deep image feature map, is the characteristic graph of the superficial layer of abdominal ultrasound image, is the feature connection operation, is a convolutional code with a convolution kernel of 3×3. is a convolutional code with a convolution kernel of 1×1. It is the semantic information field of abdominal ultrasound features;

[0094] The upsampled sparse abdominal ultrasound deep image feature map is mapped to the abdominal ultrasound feature semantic information field to obtain the sparse abdominal ultrasound deep image field modulation feature map. This process can be expressed as:

[0095] ;

[0096] in, is the semantic information field of abdominal ultrasound features, is the upsampled sparse abdominal ultrasound deep image feature map, It is the point multiplication by position. It is the field modulation feature map of the sparse abdominal ultrasound deep image.

[0097] It should be understood that the upsampled sparse abdominal ultrasound deep image feature map contains high-level, abstract semantics and partially restored detail information, while the abdominal ultrasound shallow image feature map is rich in detail information such as low-level semantics such as organ edges and textures. By performing feature connection operations on these two feature maps, features at different levels can interact directly, fully combining the advantages of both, which helps to capture multi-scale information in abdominal ultrasound images. Then, in order to construct a representation that can reflect the global and local semantic structure of abdominal ultrasound images, the feature map after feature connection needs to be input into the semantic information field predictor of the gated convolution to obtain the semantic information field of abdominal ultrasound features. As the core component of the semantic information field predictor, the gated convolution can adaptively adjust the importance of feature channels by introducing additional learning parameters, which helps to highlight the feature channels that are more critical to the expression of abdominal ultrasound features and suppress irrelevant or interfering channel information, thereby improving the quality and effectiveness of the semantic information field of abdominal ultrasound features.

[0098] Then, the upsampled sparse abdominal ultrasound deep image feature map is mapped to the abdominal ultrasound feature semantic information field to obtain the sparse abdominal ultrasound deep image field modulation feature map. It should be understood that although the upsampled sparse abdominal ultrasound deep image feature map contains certain high-level semantic information, its semantic expression can be further strengthened by establishing a connection with the abdominal ultrasound feature semantic information field. That is, the abdominal ultrasound feature semantic information field reflects the representation of the global and local semantic structure of the abdominal ultrasound image. By mapping the upsampled sparse abdominal ultrasound deep image feature map to this semantic field, each feature element in the feature map can be adjusted according to the semantic context, so that the obtained sparse abdominal ultrasound deep image field modulation feature map can more accurately reflect semantic information such as the functional state of the intestine.

[0099] Finally, multiple attention feature perception is performed on the abdominal ultrasound shallow layer image feature map and the sparse abdominal ultrasound deep layer image field modulation feature map to obtain the abdominal ultrasound deep and shallow joint perception coding feature map. The above process can be expressed as:

[0100] ;

[0101] ;

[0102] ;

[0103] in, is the field modulation feature map of the sparse abdominal ultrasound deep layer image. is a deep point-by-point convolution operation, is the batch normalization operation, It is the sparse abdominal ultrasound deep image field modulation enhancement feature map. is the activation function, is the global average pooling operation, It is a shallow point-by-point convolution operation. It is the enhanced characteristic map of the superficial layer of abdominal ultrasound image. Click on the location to add. It is the abdominal ultrasound depth and shallowness joint perception coding feature map.

[0104] It should be understood that the feature map of the superficial abdominal ultrasound image is rich in low-level visual information such as organ edges and textures, while the sparse deep abdominal ultrasound image field modulation feature map has a high-level semantic understanding after semantic modulation. In order to build a multi-level feature representation system so that the model can not only capture the detailed information in the image, but also understand the image content from the overall semantic level, it is necessary to perform multiple attention feature perception on the superficial abdominal ultrasound image feature map and the sparse deep abdominal ultrasound image field modulation feature map in this application. Multiple attention feature perception gives the model the ability to automatically learn and dynamically adjust, so that it can quickly adjust the focus of shallow and deep features according to specific task requirements, and then obtain the most relevant features for the current task. That is, the final generated abdominal ultrasound depth joint perception coding feature map realizes the organic fusion of low-level visual information and high-level semantic understanding, and can well represent the key information in the original intestinal ultrasound image, thereby providing strong support for the realization of accurate enteral nutrition care decisions.

[0105] In an embodiment of the present application, the enhanced abdominal ultrasound image generation unit 134 is used to obtain the enhanced abdominal ultrasound image based on the abdominal ultrasound depth-shallow joint perception coding feature map. Specifically, in an embodiment of the present application, the enhanced abdominal ultrasound image generation unit 134 is used to: input the abdominal ultrasound depth-shallow joint perception coding feature map into an image enhancement module based on a generative adversarial network to obtain the enhanced abdominal ultrasound image. That is, the abdominal ultrasound depth-shallow joint perception coding feature map obtained by joint perception using the sparse abdominal ultrasound deep image feature map and the abdominal ultrasound shallow image feature map is generated and processed to achieve the enhanced abdominal ultrasound image. Specifically, the generative adversarial network consists of a generator and a discriminator, which are mutually confrontational and cooperative. The task of the generator is to generate forged data based on the input data to make it as close to the real data as possible; the discriminator is responsible for judging whether the input data is real data or forged data generated by the generator. During the training process, the generator is continuously optimized to deceive the discriminator, and the discriminator is continuously optimized to better identify forged data, and finally a dynamic balance is achieved. The joint perception coding feature map of abdominal ultrasound depth and shallowness is used as the input data of the generator in the trained adversarial generative network. After receiving the joint perception coding feature map of abdominal ultrasound depth and shallowness, the generator performs complex feature mapping and conversion operations through a series of convolutional layers, deconvolutional layers (or transposed convolutional layers), normalization layers and activation functions, and finally outputs an enhanced abdominal ultrasound image with similar size and structure to the original abdominal ultrasound image. This generated image has clearer details, more accurate contrast and richer texture information, and can better reflect key information such as intestinal peristalsis and intestinal wall thickness. The clear and high-quality enhanced abdominal ultrasound image can provide more reliable support for subsequent nutritional care decision support recommendations.

[0106] In particular, since the sparse abdominal ultrasound deep image feature map and the abdominal ultrasound shallow image feature map respectively represent the shallow image semantic features and deep selective image semantic features of the abdominal ultrasound image, when performing deep and shallow semantic-guided joint perception, the semantic information field representation in the local image semantic space will have a micro-guided transfer-macro joint perception deviation relative to the global image semantic space domain, which will cause a dynamic deviation in the mapping of features to the generation target probability when the abdominal ultrasound deep and shallow joint perception encoding feature map is input into the image enhancement module based on the generative adversarial network, thereby reducing the image semantic quality of the enhanced abdominal ultrasound image.

[0107] Preferably, inputting the abdominal ultrasound depth-shallowness joint perceptual coding feature map into an image enhancement module based on a generative adversarial network to obtain the enhanced abdominal ultrasound image comprises:

[0108] Determine the feature mean of the abdominal ultrasound depth joint perception coding feature map and characteristic variance , and the feature mean Divide by the feature mean With the characteristic variance The sum of the two is used to obtain the probability value of the joint perception coding of the depth of abdominal ultrasound. The process can be expressed as:

[0109] ;

[0110] in, is the feature mean of the abdominal ultrasound depth-shallowness joint perception coding feature map, is the feature variance of the abdominal ultrasound depth joint perception coding feature map, represents the combined perception coding probability value of the abdominal ultrasound depth;

[0111] The feature mean Multiplying by the abdominal ultrasound depth joint perception coding probability value to obtain the abdominal ultrasound depth joint perception coding statistical field value, the process can be expressed as:

[0112] ;

[0113] in, is the feature mean of the abdominal ultrasound depth-shallowness joint perception coding feature map, represents the probability value of the combined perception coding of the abdominal ultrasound depth, Indicates the statistical field value of the abdominal ultrasound depth-shallowness joint perception coding;

[0114] Subtract one from the abdominal ultrasound depth joint perception coding statistical field value and divide it by the abdominal ultrasound depth joint perception coding statistical field value to obtain the abdominal ultrasound depth joint perception coding partial probability value. This process can be expressed as:

[0115] ;

[0116] in, represents the abdominal ultrasound depth and shallowness joint perception coding statistical field value, Indicates the partial probability value of the abdominal ultrasound depth-shallowness joint perception coding;

[0117] The power function of the abdominal ultrasound depth joint perception coding feature map is calculated with the abdominal ultrasound depth joint perception coding partial probability value as the exponent, and multiplied by the abdominal ultrasound depth joint perception coding partial probability value to obtain the abdominal ultrasound depth joint perception coding microscopic representation feature map. The process can be expressed as:

[0118] ;

[0119] in, represents the partial probability value of the abdominal ultrasound depth joint perception coding, It means point multiplication by position. Indicates the calculation of the abdominal ultrasound depth joint perception coding feature map is the exponential power function, A microscopic representation feature map representing the abdominal ultrasound depth-shallowness joint perception coding;

[0120] After the abdominal ultrasound depth joint perception coding feature map is multiplied by the abdominal ultrasound depth joint perception coding partial probability value, an exponential function with a natural constant as the base is calculated to obtain the abdominal ultrasound depth joint perception coding macro mapping feature map. The process can be expressed as:

[0121] ;

[0122] in, Indicates the calculation of the abdominal ultrasound depth joint perception coding feature map is the exponential power function, It means point multiplication by position. represents the exponential function value with the natural constant e as the base, A macroscopic mapping feature map representing the abdominal ultrasound depth-shallowness joint perception coding;

[0123] After calculating the logarithm of the microscopic representation feature map of the joint perceptual coding of the abdominal ultrasound depth with the base 2, the logarithm is weighted and summed with the macroscopic mapping feature map of the joint perceptual coding of the abdominal ultrasound depth to obtain the optimized joint perceptual coding feature map of the abdominal ultrasound depth. The process can be expressed as:

[0124] ;

[0125] in, represents the value of the natural logarithm function with base 2, represents the abdominal ultrasound depth-shallowness joint perception coding macroscopic mapping feature map, represents the abdominal ultrasound depth joint perception coding microscopic representation feature map, It means point multiplication by position. Indicates the location point plus, and represents the weighted hyperparameter, Representing the optimized abdominal ultrasound depth-shallowness joint perception coding feature map;

[0126] The optimized abdominal ultrasound depth-shallowness joint perceptual coding feature map is input into the image enhancement module based on the adversarial generation network to obtain the enhanced abdominal ultrasound image.

[0127] That is, the partial low-order derivative of the statistical distribution field corresponding to the abdominal ultrasound depth and shallowness joint perceptual coding feature map is used as the non-overlapping macro-feature representation behavior patch of the abdominal ultrasound depth and shallowness joint perceptual coding feature map, so as to strengthen the dynamic sensitivity of the long-series micro-complex information distribution of the abdominal ultrasound depth and shallowness joint perceptual coding feature map to the generation probability macro-representation behavior, thereby promoting the iterative dynamic consistency of the generation target between the generation target and the extracted features in the feature space-generation probability mapping, so as to improve the image semantic quality of the enhanced abdominal ultrasound image obtained by inputting the abdominal ultrasound depth and shallowness joint perceptual coding feature map into the image enhancement module based on the adversarial generation network.

[0128] In summary, the abdominal ultrasound image enhancement module 130 is clearly explained, which uses machine vision-based image recognition processing and enhancement technology to extract deep and shallow features of the abdominal ultrasound image, and then performs feature sparse processing of local activity on the abdominal ultrasound deep image features, so as to achieve an enhanced abdominal ultrasound image based on the deep and shallow semantic joint representation between the sparse abdominal ultrasound deep image features and the abdominal ultrasound shallow image features. In this way, by extracting and processing the deep and shallow features of the ultrasound image, it is possible to adapt to the individual differences and complex situations of ultrasound images of different patients, thereby significantly improving the image quality, highlighting key features, and better meeting the needs of accurate decision-making in enteral nutrition care.

[0129] In an embodiment of the present application, the nutritional care decision support suggestion generating module 140 is used to input the enhanced abdominal ultrasound image and the nutritional assessment data into a nutritional care decision support engine based on a large model to obtain nutritional care decision support suggestions. It should be understood that the enhanced abdominal ultrasound image provides real-time and intuitive morphological and functional information of the patient's abdominal organs. For example, it can display the peristalsis of the intestine, the thickness and integrity of the intestinal wall. This information directly reflects the current physiological state of the intestine, which is crucial to judging the feasibility and digestion and absorption capacity of enteral nutrition. Nutritional assessment data covers medical history data and laboratory test data. Medical history can reflect the patient's long-term health status, such as chronic disease history, which can affect nutritional needs and metabolism; laboratory test data can present nutritional indicators, biochemical parameters, etc. in the patient's blood, and evaluate nutritional status from an overall level. Therefore, in order to provide a comprehensive basis for nutritional care decisions and to meet complex decision-making needs, the present application inputs the enhanced abdominal ultrasound image and the nutritional assessment data into a nutritional care decision support engine based on a large model to obtain nutritional care decision support suggestions. In particular, after receiving the integrated input data, the large model (such as the model with the Transformer architecture) performs deep feature learning on the data through its internal multi-layer neural network structure. In this process, the model will explore the potential relationship and interaction between the features of the enhanced abdominal ultrasound image and the features of the nutritional assessment data. For example, the model may learn the association between intestinal motility features and digestive system diseases in the patient's medical history, and how this association affects nutritional care decisions, thereby predicting the most appropriate nutritional care plan for the critically ill patient. For example, when the nutritional assessment data shows that the patient has diabetes, and the enhanced abdominal ultrasound image shows that the intestinal function is basically normal, the nutritional care decision support recommendation may be to choose an enteral nutrition preparation with low content of slow-release carbohydrates and rich dietary fiber, which helps to stabilize blood sugar levels and promote intestinal motility with the help of dietary fiber; when the ultrasound image shows mild edema of the intestinal wall and slightly weakened motility, combined with the low serum protein in the nutritional assessment, the nutritional care decision support recommendation may be to choose a short peptide or amino acid enteral nutrition preparation, which can be absorbed without a complex digestion process, reducing the burden on the intestine and supplementing protein.

[0130] The following is a detailed description of a specific implementation process of "inputting the enhanced abdominal ultrasound image and the nutritional assessment data into a nutritional care decision support engine based on a large model to obtain nutritional care decision support suggestions":

[0131] First, the enhanced abdominal ultrasound images are preprocessed, including adjusting the image size to a uniform specification, such as 256×256 pixels, and normalizing the pixel values ​​to a range of 0 to 1 to meet the input format requirements of the large model. For nutritional assessment data, different types of data are processed separately, among which medical history data are carefully structured, and information such as chronic diseases (such as diabetes, hypertension, etc.), drug use history (involving drug name, dosage, frequency of use, etc.) and allergy history are encoded according to pre-set categories and coding rules, while various indicators in laboratory test data (such as serum albumin, blood sugar, etc.) are normalized according to the corresponding normal reference range. The processed data will be combined with the preprocessed ultrasound image to form a complete input sample.

[0132] Large models generally use the Transformer architecture, and the processing begins with an embedding layer that converts image data and nutritional assessment data into low-dimensional vector representations. For image data, a combination of a convolutional neural network (CNN) and the Transformer's multi-head attention mechanism can be used to extract local features using CNN, and then use the Transformer's multi-head attention mechanism to achieve global encoding to ensure the integrity and relevance of image information. For the text portion of the nutritional assessment data, natural language processing word embedding technology is used, such as the embedding layer of pre-trained models such as Word2Vec or BERT, to fully consider the semantic and grammatical structure between words and ensure accurate conversion of information.

[0133] Inside the model, the encoder part plays a key role. It contains multiple Transformer blocks, and the multi-head attention mechanism and feedforward neural network in these blocks deeply process the input data. The multi-head attention mechanism can simultaneously focus on the association between different parts of the data, such as the association analysis of the intestinal motility characteristics presented by ultrasound images and the medical history information in the nutritional assessment data, and deeply explore their impact on nutrient absorption and metabolism. The feedforward neural network further performs nonlinear transformation on the extracted features, thereby enhancing the model's expression ability and information processing capabilities.

[0134] In the decoder part, the large model generates corresponding nutritional care decision support recommendations based on the learned features and pre-set decision rules. These recommendations may involve different types of tasks, which may be classification tasks to determine the type of nutritional preparations, or regression tasks accurate to specific dosages. This depends on the model's learning of a large amount of training data and the established mapping relationship between input features and decision results.

[0135] Ultimately, the big model will output nutritional care decision support recommendations, which can be presented in text form, detailing information such as which enteral nutrition preparation is recommended, the specific dosage, the frequency of infusion, and the indicators that need to be monitored. It may also be output in the form of structured data, such as JSON format, containing information such as preparation type code, dosage value, infusion frequency, etc., to facilitate the integration and application of clinical systems. Clinical professionals will review these recommendations. Once they find that the recommendations are inconsistent with the actual clinical situation, such as the recommended nutritional preparation conflicts with the patient's allergy history, they will modify the recommendations and feedback the situation to the model development team to promote the optimization and improvement of the big model, thereby improving its accuracy and effectiveness in enteral nutrition care decisions for critically ill patients, and promoting enteral nutrition care for critically ill patients to develop in a more scientific, personalized and precise direction, and provide better support for improving the prognosis of critically ill patients.

[0136] In summary, the ultrasound-guided enteral nutrition care decision support device 100 for critically ill patients based on the embodiment of the present application is explained, which first performs a nutrition assessment on the critically ill patient to obtain nutrition assessment data including medical history data and laboratory test data, and at the same time collects the abdominal ultrasound image of the critically ill patient, and performs image enhancement processing on it to obtain an enhanced abdominal ultrasound image, and then inputs the enhanced abdominal ultrasound image and the nutrition assessment data together into a nutrition care decision support engine based on a large model to obtain nutrition care decision support suggestions. In this way, by real-time monitoring and analysis of the patient's health changes and intestinal function status, it is conducive to providing personalized, accurate and timely nutrition care suggestions.

[0137] Figure 4 FIG. 1 is a flow chart of an ultrasound-guided enteral nutrition nursing decision support method for critically ill patients according to an embodiment of the present application. Figure 4 As shown, in the ultrasound-guided enteral nutrition care decision support method for critically ill patients, the method includes: S110, performing a nutritional assessment on a critically ill patient to obtain nutritional assessment data, wherein the nutritional assessment data includes medical history data and laboratory test data; S120, acquiring an abdominal ultrasound image of the critically ill patient; S130, performing image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image; S140, inputting the enhanced abdominal ultrasound image and the nutritional assessment data into a nutrition care decision support engine based on a large model to obtain nutrition care decision support recommendations.

[0138] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned ultrasound-guided enteral nutrition nursing decision support method for critically ill patients have been described in the above reference. Figures 1 to 3 The ultrasound-guided enteral nutrition care decision support device for critically ill patients has been described in detail, and therefore, its repeated description will be omitted.

[0139] In summary, the ultrasound-guided enteral nutrition care decision support method for critically ill patients based on the embodiment of the present application is explained, which first implements a nutritional assessment on the critically ill patient to obtain nutritional assessment data including medical history data and laboratory test data, and at the same time collects the abdominal ultrasound image of the critically ill patient, and performs image enhancement processing on it to obtain an enhanced abdominal ultrasound image, and then inputs the enhanced abdominal ultrasound image and the nutritional assessment data together into a nutrition care decision support engine based on a large model to obtain nutritional care decision support recommendations. In this way, by real-time monitoring and analysis of the patient's health changes and intestinal function status, it is conducive to providing personalized, accurate and timely nutritional care recommendations.

Claims

1. An ultrasound-guided enteral nutrition nursing decision support device for critically ill patients, characterized in that: include: A nutrition assessment data acquisition module, used to perform nutrition assessment on a critically ill patient to obtain nutrition assessment data, wherein the nutrition assessment data includes medical history data and laboratory test data; An abdominal ultrasound image data acquisition module, used to acquire an abdominal ultrasound image of the critically ill patient; an abdominal ultrasound image enhancement module, used for performing image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image; a nutritional care decision support suggestion generation module, used for inputting the enhanced abdominal ultrasound image and the nutritional assessment data into a nutritional care decision support engine based on a large model to obtain a nutritional care decision support suggestion; Wherein, the abdominal ultrasound image enhancement module includes: An abdominal ultrasound image shallow and deep feature extraction unit, used for extracting image shallow features and deep features from the abdominal ultrasound image using a feature extractor based on a hollow pyramid model to obtain an abdominal ultrasound shallow image feature map and an abdominal ultrasound deep image feature map; An abdominal ultrasound image deep feature selection unit, used for performing local activity-aware feature selection on the abdominal ultrasound deep image feature map to obtain a sparse abdominal ultrasound deep image feature map; An abdominal ultrasound image depth-shallowness feature joint encoding unit is used to perform depth-shallowness semantic-guided joint perception on the sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map to obtain an abdominal ultrasound depth-shallowness joint perception encoding feature map; An enhanced abdominal ultrasound image generation unit, used for inputting the abdominal ultrasound depth-shallowness joint perceptual coding feature map into an image enhancement module based on a generative adversarial network to obtain the enhanced abdominal ultrasound image; Wherein, the abdominal ultrasound image deep feature selection unit comprises: An abdominal ultrasound deep layer feature decoupling and flattening subunit, used for performing feature decoupling and feature flattening on the abdominal ultrasound deep layer image feature map to obtain a set of abdominal ultrasound deep layer local feature vectors; The abdominal ultrasound deep layer local feature descending order arrangement subunit is used to: perform importance measurement on each abdominal ultrasound deep layer local feature vector in the set of abdominal ultrasound deep layer local feature vectors to obtain a set of abdominal ultrasound deep layer local feature importance score values; based on the set of abdominal ultrasound deep layer local feature importance score values, perform descending order arrangement on the set of abdominal ultrasound deep layer local feature vectors to obtain a descending sequence of abdominal ultrasound deep layer local feature vectors; The sparse abdominal ultrasound deep layer image feature map generating subunit is used to: calculate the feature neighborhood activity of each abdominal ultrasound deep layer local feature vector in the descending sequence of the abdominal ultrasound deep layer local feature vectors to obtain a sequence of the abdominal ultrasound deep layer local feature neighborhood activity, and the process is expressed as: in, It is The first local feature vector of the deep layer of abdominal ultrasound The eigenvalues ​​at the positions, It is The number of eigenvalues ​​in the deep local eigenvector of abdominal ultrasound, and They are The deep local feature vector of abdominal ultrasound and the The weighted average eigenvalue of the deep layer features of abdominal ultrasound corresponding to the deep layer local eigenvectors of abdominal ultrasound, It is The activity of the neighborhood of the local feature of the deep layer of abdominal ultrasound corresponds to the local feature vector of the deep layer of abdominal ultrasound; based on the sequence of the neighborhood activity of the local feature vector of the deep layer of abdominal ultrasound, feature selection is performed on the descending sequence of the local feature vector of the deep layer of abdominal ultrasound to obtain the descending sequence of the local feature vector of the deep layer of abdominal ultrasound after selection; the descending sequence of the local feature vector of the deep layer of abdominal ultrasound after selection is subjected to feature shape reshaping to obtain the sparse abdominal ultrasound deep layer image feature map; Wherein, the abdominal ultrasound image depth feature joint encoding unit comprises: a sparse abdominal ultrasound deep layer image feature map upsampling subunit, configured to upsample the sparse abdominal ultrasound deep layer image feature map to obtain an upsampled sparse abdominal ultrasound deep layer image feature map, wherein the upsampled sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map have the same size; An upsampled sparse abdominal ultrasound deep layer feature mapping modulation subunit is used to map and modulate the upsampled sparse abdominal ultrasound deep layer feature map based on the semantic information field between the upsampled sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map to obtain a sparse abdominal ultrasound deep layer image field modulation feature map; The abdominal ultrasound depth-shallowness joint perception coding feature generation subunit is used to perform multiple attention feature perception on the abdominal ultrasound shallow image feature map and the sparse abdominal ultrasound deep image field modulation feature map to obtain the abdominal ultrasound depth-shallowness joint perception coding feature map.

2. The ultrasound-guided enteral nutrition nursing decision support device for critically ill patients according to claim 1, characterized in that: The upsampling and sparse abdominal ultrasound deep feature mapping modulation subunit is used to: The upsampled and sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map are feature-connected and then input into the semantic information field predictor of gated convolution to obtain the abdominal ultrasound feature semantic information field as the semantic information field; The upsampled sparse abdominal ultrasound deep image feature map is mapped to the abdominal ultrasound feature semantic information field to obtain the sparse abdominal ultrasound deep image field modulation feature map.

3. An ultrasound-guided enteral nutrition nursing decision support method for critically ill patients, using the ultrasound-guided enteral nutrition nursing decision support device for critically ill patients according to claim 1, characterized in that: include: Performing a nutritional assessment on a critically ill patient to obtain nutritional assessment data, wherein the nutritional assessment data includes medical history data and laboratory test data; Acquiring an abdominal ultrasound image of the critically ill patient; Performing image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image; inputting the enhanced abdominal ultrasound image and the nutritional assessment data into a nutritional care decision support engine based on a large model to obtain nutritional care decision support recommendations; The step of performing image enhancement on the abdominal ultrasound image to obtain an enhanced abdominal ultrasound image includes: Extracting image superficial features and deep features from the abdominal ultrasound image to obtain an abdominal ultrasound superficial image feature map and an abdominal ultrasound deep image feature map; Performing local activity-aware feature selection on the abdominal ultrasound deep image feature map to obtain a sparse abdominal ultrasound deep image feature map; Performing deep-shallow semantic-guided joint perception on the sparse abdominal ultrasound deep layer image feature map and the abdominal ultrasound shallow layer image feature map to obtain an abdominal ultrasound deep-shallow joint perception coding feature map; Based on the abdominal ultrasound depth and shallowness joint perception coding feature map, the enhanced abdominal ultrasound image is obtained.

Images