Heart failure patient water inflow monitoring system and method based on intelligent water content evaluation

Abstract

The invention relates to a heart failure patient water inflow monitoring system and method based on intelligent water content evaluation, and the system can monitor and evaluate the inflow of a heart failure patient every time through a weight sensor, a three-dimensional camera, a clinical terminal and a nurse station background on a clinical dining carrier. Comprising the food type, the weight and the water content ratio of the current food to be eaten, and the total weight and the total water content of the food eaten this time, and generating a daily in-out report in combination with the sunrise of the heart failure patient. In this way, the daily incoming and outgoing amount of the heart failure patient can be accurately evaluated, specific parameters of the eaten food can be counted in detail and the total water content of the eaten food can be evaluated in detail without tracking and recording of nurses, the subsequent diet balance adjusting strategy is greatly influenced, visual eating data are automatically provided for background doctors, and the efficiency is improved. Doctors can conveniently adjust nursing or rehabilitation and diagnosis and treatment schemes according to the in-out amount of the patient.

Description

Technical Field

[0001] The present disclosure relates to the field of intelligent medical monitoring technology, and in particular to a water intake monitoring system and method for heart failure patients based on intelligent water content assessment, and an electronic device. Background Art

[0002] Recording the amount of fluid in and out of the body in patients with heart failure has the following clinical implications:

[0003] First, the heart pumping function of patients with heart failure is impaired, resulting in poor circulation of fluid in the body and easy accumulation of excess fluid. Scientific management of daily intake and output has become a necessary means to maintain fluid balance and reduce the burden on the heart. By recording the intake and output, fluid retention can be detected and corrected in time, thereby reducing the patient's edema and other discomfort symptoms.

[0004] Secondly, recording the amount of urine in and out helps evaluate the effectiveness of diuretics. Diuretics are often used in the treatment of heart failure to increase urine excretion and reduce the amount of fluid in the body. By recording the daily urine volume, medical staff can determine whether the diuretic dosage is appropriate, adjust the treatment plan in time, and ensure the effectiveness and safety of diuretics.

[0005] In addition, the intake and output records can also be used as an important indicator to evaluate the condition of patients with heart failure. The intake and output of patients with heart failure should maintain a certain negative balance, that is, the output is greater than the intake, in order to reduce the accumulation of fluid in the body. If the intake and output cannot achieve a negative balance, or there are abnormal changes in the intake and output, it may indicate that the patient's condition has worsened and timely treatment measures are needed.

[0006] At the same time, intake and output records can also help guide patients' diet and lifestyle. Heart failure patients need to limit their intake of water and high-salt foods to avoid increasing the burden on the heart. By recording intake and output, patients can better understand their water intake and eating habits, thereby cooperating with treatment and improving their quality of life.

[0007] However, in actual monitoring, the intake and output of heart failure patients are mainly recorded by patients or nurses, and only the number of food consumed can be roughly counted, but the specific parameters of the food consumed cannot be counted in detail and the water content of the food consumed cannot be evaluated. Furthermore, the total water content of the food consumed cannot be evaluated in detail. This has a great impact on the subsequent adjustment of the dietary balance strategy and cannot provide doctors with intuitive eating data.

[0008] Therefore, it is necessary to provide a solution that can record in detail the food information and intake and output of each meal of heart failure patients, so as to facilitate physicians to adjust nursing or rehabilitation and treatment plans according to the patient's intake and output. Summary of the invention

[0009] To solve the above problems, the present application proposes a monitoring system, method, and electronic device for the water intake of heart failure patients based on intelligent water content assessment.

[0010] On the one hand, the present application proposes a monitoring system for the water intake of heart failure patients based on intelligent water content assessment, including a clinical dining carrier for carrying food to be eaten. Among them, the following are integrated on the clinical dining carrier:

[0011] A weight sensor for monitoring the weight signal of the food to be eaten and sending it to the clinical terminal;

[0012] A three-dimensional camera for collecting three-dimensional image data of the food to be eaten and sending it to the clinical terminal;

[0013] A clinical terminal for processing the weight signal and calculating the weight corresponding to the food to be eaten; and, identifying the type of the food to be eaten according to the three-dimensional image data, matching the water content ratio corresponding to the type, and estimating the total water content of this meal:

[0014]

[0015] Among them, i represents the food type of the current food to be eaten, W i represents the weight of the current food to be eaten, W total(i) represents the total weight of the food eaten this time, k (i) represents the water content ratio corresponding to the food type, and C represents the total water content of this meal;

[0016] The nurse station background is used to read the intake of heart failure patients each time from the clinical terminal in real time, including the food type, weight, water content ratio, total weight of the food eaten this time, and total water content of the current food to be eaten, and generate a daily intake and output report in combination with the daily output of heart failure patients;

[0017] The weight sensor and the three-dimensional camera are communicatively connected to the clinical terminal;

[0018] The clinical terminal is communicatively connected to the nurse station background.

[0019] As an optional implementation scheme of the present application, optionally, the identifying the type of the food to be eaten according to the three-dimensional image data includes:

[0020] Randomly extracting the food frame image from the three-dimensional image data and preprocessing it;

[0021] Inputting the food frame image into the food type recognition model pre-deployed on the clinical terminal, and the food type recognition model recognizes and outputs the corresponding type and the water content ratio corresponding to the type.

[0022] As an alternative implementation of the present application, optionally, the method for generating the food type recognition model includes:

[0023] Pre-collect food images of several types and perform preprocessing;

[0024] Use a convolutional neural network to extract image features from food images of various types, and label the corresponding type labels and water content ratio labels;

[0025] Statistically analyze the image features of various types, construct a feature set, and divide it into a training set and a validation set according to a preset ratio;

[0026] Input the training set into a preset CNN model to train and generate the initial food type recognition model;

[0027] Use the delivery validation set to verify the recognition performance of the food type recognition model:

[0028] If the verification passes, then deploy and apply the food type recognition model to the clinical terminal;

[0029] Otherwise, retrain.

[0030] As an alternative implementation of the present application, optionally, the clinical terminal is further configured to:

[0031] Generate a food three-dimensional model corresponding to the to-be-eaten food according to the three-dimensional image data;

[0032] Perform mesh division on the food three-dimensional model to obtain several food blocks;

[0033] Extract the three-dimensional parameters of the food blocks: length L, width W, and height H through an image recognition algorithm, and estimate the volume V of the to-be-eaten food:

[0034]

[0035] where Vi is the volume of the i-th food block:

[0036]

[0037] As an alternative implementation of the present application, optionally, the nurse station background is further configured to:

[0038] Pre-set upper and lower limit values for the daily intake of patients with heart failure;

[0039] Statistically calculate the daily intake based on each intake amount, and determine whether the daily intake of the heart failure patient is between the upper and lower limit values:

[0040] If in a certain state, generate the corresponding daily intake and output report;

[0041] Otherwise, generate a corresponding warning signal and send it to the clinical terminal.

[0042] On the other hand, this application proposes a method for monitoring the water intake of heart failure patients based on intelligent assessment of water content, including the following steps:

[0043] Start eating, and place the food to be eaten on the clinical dining carrier;

[0044] The weight sensor monitors the weight signal of the food to be eaten and sends it to the clinical terminal;

[0045] The three-dimensional camera collects the three-dimensional image data of the food to be eaten and sends it to the clinical terminal;

[0046] The clinical terminal processes the weight signal and calculates the weight of the corresponding food to be eaten; and, based on the three-dimensional image data, identifies the type of the corresponding food to be eaten, matches the water content ratio corresponding to the type, and estimates the total water content of this meal:

[0047]

[0048] Among them, i represents the food type of the current food to be eaten, W i represents the weight of the current food to be eaten, W total ( i) represents the total weight of the food eaten this time, k (i) represents the water content ratio corresponding to the food type, and C represents the total water content of this meal;

[0049] The nurse station background reads the intake of each time of the heart failure patient from the clinical terminal in real time, including the food type, weight, water content ratio, total weight of the food eaten this time, and total water content of the current food to be eaten, and generates a daily intake and output report in combination with the daily output of the heart failure patient; and, based on the intake of each time, calculates the daily intake, and determines whether the daily intake of the heart failure patient is between the preset upper limit and lower limit:

[0050] If in a certain state, generate the corresponding daily intake and output report;

[0051] Otherwise, generate a corresponding warning signal and send it to the clinical terminal;

[0052] The clinical terminal responds to the warning signal.

[0053] On the other hand, this application also proposes an electronic device, including:

[0054] A processor;

[0055] A memory for storing processor-executable instructions;

[0056] Wherein, when the processor is configured to execute the executable instructions, the intelligent evaluation method for monitoring the water intake of heart failure patients based on water content is implemented.

[0057] Technical effects of the present invention:

[0058] Based on the implementation of this application, the present invention includes, on a clinical dining carrier: a weight sensor for monitoring the weight signal of the food to be eaten and sending it to a clinical terminal; a three-dimensional camera for collecting three-dimensional image data of the food to be eaten and sending it to the clinical terminal; the clinical terminal for processing the weight signal and calculating the weight corresponding to the food to be eaten; and, identifying the type of the food to be eaten according to the three-dimensional image data, matching the water content ratio corresponding to the type of the food to be eaten, and estimating the total water content of this meal; a nurse station background for reading the amount of each intake of heart failure patients from the clinical terminal in real time, including the food type, weight, water content ratio, total weight of the food eaten this time and total water content of the current food to be eaten, and generating a daily intake and output report in combination with the daily output of the heart failure patient. In this way, the daily intake and output of heart failure patients can be accurately evaluated. Without the need for nurses to track and record, the specific parameters of the food eaten can be detailedly counted and the total water content of the eaten food can be detailedly evaluated, which has a great impact on subsequent adjustment of the diet balance strategy. Automatically provides intuitive eating data for the physicians in the background, facilitating the physicians to adjust the nursing, rehabilitation, and diagnosis and treatment plans according to the intake and output of the patients.

[0059] With reference to the following detailed description of the exemplary embodiments with reference to the accompanying drawings, other features and aspects of the present disclosure will become apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The drawings included in the specification and constituting a part of the specification, together with the specification, illustrate the exemplary embodiments, features, and aspects of the present disclosure and are used to explain the principles of the present disclosure.

[0061] Figure 1 Shown as an application structure schematic diagram of the present invention;

[0062] Figure 2 Shown as a schematic diagram of the composition of the application system of the present invention;

[0063] Figure 3 Shown as a schematic diagram of model training of the present invention;

[0064] Figure 4 Shown as an application schematic diagram of an electronic device of the present invention DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] Various exemplary embodiments, features, and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. Like reference numerals in the drawings denote functionally identical or similar elements. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0066] As used herein, the term "exemplary" means "serving as an example, embodiment, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as superior or better than other embodiments.

[0067] In addition, for a better description of the present disclosure, numerous specific details are given in the following detailed description. Those skilled in the art should understand that the present disclosure can be implemented without some specific details. In some instances, well-known means, elements, and circuits have not been described in detail so as to highlight the gist of the present disclosure.

[0068] Embodiment 1

[0069] As Figure 1 shown, on the one hand, the present application provides a monitoring system for the water intake of heart failure patients based on intelligent assessment of water content, including a clinical dining carrier 1, such as a hospital bed, on which there is a movable bracket, and on the movable bracket there is a dinner plate 3 (which can be installed on the vertical rod of the bracket through a sleeve and the height can be adjusted up and down by a positioning pin), and the dinner plate 3 is used to carry the food to be eaten. Among them, the clinical dining carrier 1 is integrated with:

[0070] A weight sensor 2, which is arranged inside the dinner plate 3 and covered with a protective layer, such as a plate made of PC material, on which food can be placed, and is used to monitor the weight signal of the food to be eaten and send it to the clinical terminal;

[0071] A three-dimensional camera 4, which is arranged above the dinner plate 3 and fixed on the bracket, and can collect the food video image data stream and transmit it to the clinical terminal 5 (PC or other terminal servers, which can provide software services such as image processing), and is used to collect the three-dimensional image data of the food to be eaten and send it to the clinical terminal;

[0072] A clinical terminal 5, which is used to process the weight signal and calculate the weight corresponding to the food to be eaten; and, according to the three-dimensional image data, identify the type of the food to be eaten, match the water content ratio corresponding to the type of the food to be eaten, and estimate the total water content of this meal:

[0073]

[0074] Among them, i represents the food type of the current food to be eaten, W i represents the weight of the current food to be eaten, W total(i)Represents the total weight of the food for this meal, k (i) Represents the water content percentage of the corresponding food type, and C represents the total water content of this meal;

[0075] The nurse station background, the server deployed at the nurse station, can communicate with the clinical terminals in each patient's ward, perform data monitoring and collection. The administrator can configure and record in advance each patient and their corresponding terminal numbers id, etc., for easy management. It is used to read the intake of heart failure patients each time from the clinical terminal in real time, including the food type, weight, water content percentage, total weight of the food for this meal, and total water content of the currently to-be-consumed food, and generate a daily intake and output report in combination with the daily output of heart failure patients;

[0076] The weight sensor and the 3D camera are communicatively connected to the clinical terminal;

[0077] The clinical terminal is communicatively connected to the nurse station background.

[0078] Next, the application of this system will be described in detail with reference to the attached Figure 2 as shown below.

[0079] I. Hardware Selection and Deployment

[0080] 1. Weight Sensor

[0081] Model: HX711 High-precision Weighing Sensor Module (range 0 - 5 kg, resolution ±1 g)

[0082] Installation: Embedded at the bottom of the dinner plate, covered with a PC protection board (thickness 2 mm), fixed by bolts, and the sensor cable is led out through the holes in the protection board.

[0083] Communication: Connected to the clinical terminal through an RS485 to USB module, baud rate 9600 bps, data format: ASCII code weight value (unit g).

[0084] Calibration example: Send the TARE command to zero when the plate is empty, load a 500 g standard weight for calibration, error compensation formula: W actual = 0.998 × W original + 2.1 g.

[0085] 2. 3D Camera

[0086] Model: Intel RealSense D415 (RGB-D camera, resolution 1280×720@30fps, depth accuracy ±1 mm)

[0087] Bracket: Aluminum alloy adjustable bracket (height 30 - 50 cm, pitch angle ±15°), fixed 30 cm directly above the dinner plate to ensure full coverage of the dinner plate area.

[0088] Data stream: Connect to the clinical terminal via USB 3.0, output RGB images + depth point cloud data, with a frame rate synchronized at 30 fps.

[0089] 3. Clinical terminal

[0090] Hardware: Intel NUC 11 (i5-1135G7 / 16GB RAM / 512GB SSD), pre-installed with Windows 10 IoT system.

[0091] Interfaces: Built-in Wi-Fi 6 and Bluetooth 5.2, with extended USB / RS485 interfaces to connect sensors and cameras.

[0092] Terminal-sensor communication

[0093] Protocols: Modbus RTU (weight sensor) + UVC protocol (camera).

[0094] Packet example:

[0095] / / Weight sensor data

[0096] {"device_id":"SENSOR_001","weight_g":250,"timestamp":1630000000}

[0097] / / Camera data stream

[0098] RGB frame: H.264 encoded, depth frame: 16-bit grayscale image (unit: mm).

[0099] Terminal-nurse station communication

[0100] Protocols: MQTT (real-time data) + HTTP RESTful API (batch reporting).

[0101] JSON reporting example:

[0102]

[0103] Clinical terminal software module

[0104] Sensor data reception: A Python script reads weight sensor data through the pyserial library, and a filtering algorithm (sliding average window of 10 frames) eliminates noise.

[0105] Image preprocessing: OpenCV 4.5 intercepts the ROI area of the dinner plate (positioned based on depth point cloud coordinates), and the cropped image is scaled to 224×224 pixels.

[0106] Food type recognition model:

[0107] As Figure 3 shown, as an optional implementation of the present application, optionally, the method for generating the food type recognition model includes:

[0108] Pre-collect food images of several types and preprocess them;

[0109] Use a convolutional neural network to extract image features from food images of various types, and label the corresponding type labels and water content ratio labels;

[0110] Statistically analyze the image features of various types, construct a feature set, and divide it into a training set and a validation set according to a preset ratio;

[0111] Input the training set into a preset CNN model to train and generate the initial food type recognition model;

[0112] Use the delivery validation set to verify the recognition performance of the food type recognition model:

[0113] If the verification passes, deploy and apply the food type recognition model to the clinical terminal;

[0114] Otherwise, retrain.

[0115] Training steps of the food type recognition model:

[0116] I. Data collection and preprocessing

[0117] Public dataset: Select 20 types of high-frequency foods related to the hospital scenario (such as apples, rice, porridge, bread) from the Food-101 (101 types of foods, 1000 images for each type).

[0118] Self-collected data: Collect 500 images of 20 types of foods in the hospital cafeteria through Intel RealSense D415, covering multiple angles, lighting conditions, and plate placement forms.

[0119] Total dataset: 20 types × 500 images = 10,000 images (examples: apples, white porridge, steamed fish, etc. In the figure, sweet potatoes are used as an example, and in this embodiment, single-food intake is used as an example).

[0120] Preprocessing process

[0121] ROI extraction: Locate the plate area based on the depth point cloud data (depth threshold 30 cm ± 5 cm), and crop the effective area.

[0122] #Pseudocode: Depth threshold segmentation

[0123] depth_mask = (depth_image > 250mm) & (depth_image < 350mm)

[0124] roi = cv2.bitwise_and(rgb_image, depth_mask)

[0125] Size normalization: Scale to 224×224 pixels (to fit the input of ResNet).

[0126] Data augmentation:

[0127] Geometric transformation: Random horizontal flipping (probability 50%), rotation (±15°).

[0128] Color adjustment: Brightness (±20%), contrast (±15%).

[0129] Background replacement: Replace non-dinner plate areas with a solid color background (such as white) to reduce interference.

[0130] Case:

[0131] Original apple image (1200×800) → Crop ROI (600×600) → Scale to 224×224 → Generate 4 variants after enhancement (rotation, brightness adjustment).

[0132] II. Feature Extraction and Annotation

[0133] Convolutional Neural Network (CNN) architecture

[0134] Base model: ResNet-18 (Pretrained weights: ImageNet).

[0135] Feature extraction layer: Keep the first 17 convolutional layers and remove the top fully connected layer.

[0136] Output layer:

[0137] Classification branch: 20-dimensional softmax output (corresponding to 20 food categories).

[0138] Regression branch: 1-dimensional fully connected layer predicts the water content ratio (for end-to-end prediction if needed).

[0139] Label annotation

[0140] Type label: Manually annotate the food category of each image (such as "Apple-1", "White Porridge-2").

[0141] Water content label: Based on the preset values in the medical nutrition database (lookup table method):

[0142] Food type Water content ratio (%) Apple 85 White congee 90 Steamed fish 70

[0143] Feature statistics example: The fruit images output feature vectors (512 dimensions) after passing through the 17th layer of ResNet. After PCA dimensionality reduction, the visualization shows clustering (intra-class feature distance < 0.3, inter-class > 0.7).

[0144] III. Dataset Division and Model Training

[0145] Dataset Division

[0146] Ratio: 70% (7,000 images) for the training set and 30% (3,000 images) for the validation set. Stratified sampling is used to ensure class balance.

[0147] File Structure:

[0148]

[0149] Training Parameters

[0150] Framework: PyTorch 1.12 + OpenVINO 2022.3.

[0151] Hyperparameters

[0152] Optimizer: Adam (lr = 1e-4, weight_decay = 1e-5).

[0153] Loss function: Cross-entropy (for classification) + L1 Loss (for water content regression, weight 0.3).

[0154] Batch Size: 32 (the measured memory occupancy on NUC 11 ≤ 4GB).

[0155] Training epochs: 50 epochs, with early stopping strategy (terminate when val_loss does not decrease for 5 consecutive times).

[0156] Example of the training process:

[0157] Epoch 10: train_loss = 0.15, val_acc = 89.3%.

[0158] Epoch 20: train_loss = 0.08, val_acc = 92.1%.

[0159] IV. Model Validation and Deployment

[0160] Validation Metrics

[0161] Classification accuracy: The overall accuracy of the validation set ≥ 90% (hard pass standard).

[0162] Confusion matrix (example):

[0163] True / Predicted Apple White congee Steamed fish Apple 145 3 2 White congee 2 142 6 Steamed fish 1 4 145

[0164] Regression error (if enabled): Moisture content prediction MAE ≤ 3%.

[0165] Deployment optimization

[0166] Model compression: FP32→INT8 quantization (accuracy loss ≤ 1%), the size is compressed from 89MB to 23MB.

[0167] Reasoning acceleration: OpenVINO asynchronous reasoning engine, single frame time ≤ 50ms (meets 30fps real-time performance).

[0168] Clinical terminal integration:

[0169] # Pseudocode: Model loading and inference

[0170] model=OpenVINO.load("food_classifier_int8.xml")

[0171] img = preprocess(image)

[0172] cls_result,water_ratio=model.infer(img).

[0173] Examples:

[0174] Input: preprocessed porridge image → Output: type "porridge" (confidence 96%), water content 90%.

[0175] 5. Failure handling and retraining

[0176] Verification failed scenario

[0177] Condition: Validation set accuracy < 90% or MAE > 3%.

[0178] Root Cause Analysis:

[0179] Data bias: Insufficient sample size for a certain category (e.g., only 200 samples for “steamed fish”).

[0180] Overfitting: train_acc = 99% but val_acc = 85%.

[0181] Solution:

[0182] Data enhancement: Targetedly increase the number of enhanced samples of underfitting categories.

[0183] Adjust the model: add a Dropout layer (rate=0.5) or increase the L2 regularization coefficient.

[0184] Retraining Example:

[0185] Expand "Steamed Fish" to 500 images (add 300 new images). After retraining, the val_acc is increased to 93.5%.

[0186] In this solution, the measured throughput on Intel NUC 11 reaches 20 FPS, supporting real-time recognition of 20 types of food, with the water content calculation error ≤ 2%, meeting the accurate water control requirements for heart failure patients.

[0187] 4. Nurse Station Back-end Server

[0188] Model: Dell PowerEdge R350 (Xeon E-2314 / 32GB RAM / 2TB HDD), deployed with Ubuntu Server 22.04 LTS.

[0189] Database: MySQL 8.0, storing patient ID, terminal number, and eating records (fields: timestamp, food type, weight, water content).

[0190] Nurse Station Back-end Functions

[0191] Data Dashboard: The Web interface (Vue.js + ECharts) displays patients' eating data in real-time, supporting filtering by time (such as 24-hour trend chart).

[0192] Warning Logic: If the water content of a single meal > 500 ml, trigger a pop-up warning and send a text message to the nurse's mobile phone (through the Twilio API).

[0193] Report Generation: Automatically generate a PDF report daily, calculate the balance of intake and output (formula: Δ = intake - output), and highlight values exceeding the threshold in red.

[0194] Application Example

[0195] Scenario 1: Eating an apple + congee

[0196] Weight Sensor: Apple 150g + Congee 300g → Total weight 450g.

[0197] Image Recognition: Apple (confidence 98%), Congee (confidence 96%).

[0198] Water Content Calculation: 150 × 0.85 + 300 × 0.9 = 397.5 ml.

[0199] Nurse Station Display: The cumulative intake for the day is 397.5 ml (if the output is 300 ml, the balance value is +97.5 ml).

[0200] Therefore, based on the above solution, the water content of food can be quickly evaluated and statistically analyzed by the back-end.

[0201] As an alternative implementation of the present application, optionally, identifying the type of the food to be eaten according to the three-dimensional image data includes:

[0202] Randomly extract the food frame images from the three-dimensional image data and preprocess them;

[0203] Input the food frame images into the food type recognition model pre-deployed on the clinical terminal, and the food type recognition model identifies and outputs the corresponding type and the water content ratio of the corresponding type.

[0204] Obtain food pictures by frame extraction and perform AI recognition.

[0205] Random frame extraction

[0206] Input data: RGB-D video stream (resolution 1280×720@30fps, depth accuracy ±1mm), extract 1 frame every 30 seconds (covering the complete meal-taking action cycle).

[0207] #Pseudocode: Random sampling based on timestamp (example)

[0208] import random

[0209] total_frames = 900 #30 seconds × 30fps

[0210] sample_rate = 0.033 #1 frame / 30 seconds

[0211] sampled_indices = random.sample(range(total_frames), int(total_frames * sample_rate)):ml-citation{ref="2,3"data="citationList"}。

[0212] Preprocessing steps

[0213] Depth threshold segmentation:

[0214] Filter non-dining plate areas through depth data (depth range 250 - 350mm), and extract the food ROI (example: original image 1200×800 → ROI area 600×600).

[0215] Geometric normalization:

[0216] Scale to 224×224 pixels, adapt to the input size of ResNet, and at the same time perform central cropping (eliminating edge interference).

[0217] Data augmentation:

[0218] Random horizontal flip (probability 50%);

[0219] Brightness adjustment (±20%), contrast adjustment (±15%).

[0220] Food type recognition model

[0221] Base model: ResNet-18 (transfer learning) + classification head (20 food categories).

[0222] Just identify the food type through the model and output the corresponding label. See the above model training and application principles for details.

[0223] It is possible to perform type recognition on a number of randomly selected images, and finally output the result with the most types to avoid recognition errors and recognition errors caused by the appearance of food.

[0224] Here, the terminal can also construct a three-dimensional model of the food based on the three-dimensional image to statistically calculate the corresponding food size.

[0225] As an alternative implementation of this application, optionally, the clinical terminal is further configured to:

[0226] Generate a three-dimensional model of the food to be eaten corresponding to the three-dimensional image data;

[0227] Perform mesh division on the three-dimensional model of the food to obtain a number of food blocks;

[0228] Extract the three-dimensional parameters of the food blocks through an image recognition algorithm: length L, width W, and height H, and estimate the volume V of the food to be eaten:

[0229]

[0230] where Vi is the volume of the i-th food block:

[0231]

[0232] Here, because the food is in an irregular shape, it is divided into blocks, and the volume is calculated separately and then summed to obtain the total volume.

[0233] Specifically:

[0234] I. Three-dimensional model generation process

[0235] Three-dimensional data acquisition

[0236] Use an RGB-D camera (such as Intel RealSense L515) to collect the three-dimensional point cloud data of the food to be eaten, with a depth accuracy of ±1mm and a resolution of 1280×720@30fps.

[0237] Example data: The apple point cloud contains approximately 50,000 coordinate points, covering the surface uneven details.

[0238] Three-dimensional reconstruction algorithm

[0239] Surface reconstruction: Based on the Poisson Surface Reconstruction algorithm, convert the discrete point cloud into a continuous triangular mesh model.

[0240] # Pseudocode: Core steps of Poisson reconstruction

[0241] points = load_point_cloud("apple.ply")

[0242] mesh = poisson_reconstruction(points, depth = 10)

[0243] mesh.export("apple_mesh.obj").

[0244] Model optimization: Apply the Laplace smoothing algorithm to eliminate noise and reduce surface jaggedness (smoothing iteration times = 3).

[0245] II. Mesh division and parameter extraction

[0246] Mesh division method

[0247] Uniform division: Divide the three-dimensional food model into a spatial grid of 10mm×10mm×10mm cubes, generating approximately 1,000 blocks (taking an apple with a diameter of 80mm as an example).

[0248] Dynamic adjustment: For irregular shapes (such as vegetable leaves), use octree adaptive subdivision (minimum mesh size 5mm).

[0249] Three-dimensional parameter calculation

[0250] Bounding box extraction: Calculate the axis-aligned bounding box (AABB) for each block to obtain the extreme differences in length (L), width (W), and height (H).

[0251] # Pseudocode: Calculation of block bounding box

[0252] for chunk in mesh_chunks:

[0253] min_coord = np.min(chunk.vertices, axis = 0)

[0254] max_coord = np.max(chunk.vertices, axis = 0)

[0255] L = max_coord - min_coord # X - axis length

[0256] W = max_coord:ml - citation{ref = "1" data = "citationList"} - min_coord:ml - citation{ref = "1" data = "citationList"} # Y - axis width

[0257] H = max_coord:ml - citation{ref = "2" data = "citationList"} - min_coord:ml - citation{ref = "2" data = "citationList"} # Z - axis height.

[0258] Estimation of the volume of the block. Assuming that the shape of the food is approximately an ellipsoid, the volume calculation formula is:

[0259]

[0260] Calculation of the total volume

[0261] Accuracy test (taking standard geometric bodies as examples)

[0262]

[0263]

[0264] As an alternative implementation of the present application, optionally, the back - end of the nurse station is further configured to:

[0265] Pre - set the upper limit value and the lower limit value of the daily intake for patients with heart failure;

[0266] According to the daily intake statistics obtained from each intake, determine whether the daily intake of the heart - failure patient is between the upper limit value and the lower limit value:

[0267] If it is, generate the corresponding daily intake and output report;

[0268] Otherwise, generate the corresponding warning signal and send it to the clinical terminal.

[0269] The specific steps are as follows:

[0270] 1. Setting the upper and lower limits of patient intake

[0271] Data table structure:

[0272]

[0273] Configuration process:

[0274] After the doctor issues a prescription, the nurse inputs or synchronizes the data of the electronic medical order system through the background interface (example: the upper limit of the daily intake of patient A is 1500 ml, and the lower limit is 800 ml).

[0275] Input verification: The lower limit value must be ≤ the upper limit value, and the difference between the upper and lower limits must be ≥ 200 ml (to prevent misoperation).

[0276] 2. Intake statistics logic

[0277] Data source:

[0278] Each feeding data reported by the clinical terminal (structure example):

[0279]

[0280] The total daily intake can be the amount entered each time when feeding.

[0281] 3. Threshold judgment and warning mechanism

[0282] Daily scheduled task: Trigger a check at 23:59 every day. If the total intake is not within the [lower limit, upper limit] range, generate a warning signal.

[0283] #Pseudo code: Threshold judgment

[0284]

[0285] Warning signal type:

[0286]

[0287] 4. Report and warning distribution

[0288] Report generation, content template:

[0289]

[0290] 5. Warning signal distribution

[0291] Transport protocol: Push to the clinical terminal in real time through the MQTT protocol (including warning level and handling suggestions).

[0292] Terminal display example:

[0293]

[0294] Early warning handling process: The nurse station receives a red warning → A pop-up reminder appears on the clinical terminal and an alarm sound is played → After the nurse confirms, the warning is manually closed.

[0295] Historical data traceability, supporting the export of PDF reports by patient and date range (example file name: P202311001_2023-11-05_Intake_Report.pdf).

[0296] Therefore, it is possible to visually obtain the daily intake and output of heart failure patients in the background, facilitating nurses or doctors to adjust the diagnosis, treatment, and rehabilitation plans based on the patient's intake and output. For the output, such as urine collection, it can be implemented in combination with the existing hospital monitoring plan, and this plan does not involve its description. Statistical monitoring data of the output can be added to the report.

[0297] Obviously, those skilled in the art should understand that to implement all or part of the processes in the above embodiments, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above control embodiments. Those skilled in the art can understand that to implement all or part of the processes in the above embodiments, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above control embodiments. Among them, the storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include a combination of the above types of memories.

[0298] Embodiment 2

[0299] Based on the implementation principle of Embodiment 1, on the other hand, this application proposes a method for monitoring the water intake of heart failure patients based on intelligent evaluation of water content, including the following steps:

[0300] Start dining, and place the food to be eaten on the clinical dining carrier;

[0301] The weight sensor monitors the weight signal of the food to be eaten and sends it to the clinical terminal;

[0302] The three-dimensional camera collects the three-dimensional image data of the food to be eaten and sends it to the clinical terminal;

[0303] The clinical terminal processes the weight signal and calculates the weight corresponding to the food to be eaten; and, identifies the type of the food to be eaten according to the three-dimensional image data, matches the water content ratio corresponding to the type of the food to be eaten, and estimates the total water content of this meal:

[0304]

[0305] wherein, i represents the food type of the currently to-be-eaten food, W i represents the weight of the currently to-be-eaten food, W total(i) represents the total weight of the food for this meal, k (i) represents the water content ratio corresponding to the food type, and C represents the total water content of this meal;

[0306] The nurse station background reads the intake of the heart failure patient by the clinical terminal in real time, including the food type, weight, water content ratio, total weight of the food for this meal, and total water content of the currently to-be-eaten food, and generates a daily intake and output report in combination with the daily output of the heart failure patient; and, statistically obtains the daily intake according to each intake, and determines whether the daily intake of the heart failure patient is between the preset upper limit value and lower limit value:

[0307] If it is, then generate the corresponding daily intake and output report;

[0308] Otherwise, generate a corresponding warning signal and send it to the clinical terminal;

[0309] The clinical terminal responds to the warning signal.

[0310] The above steps should be understood and implemented in combination with the corresponding steps in Embodiment 1, and will not be elaborated here.

[0311] Each module or each step of the above-mentioned present invention can be implemented by a general-purpose computing system. They can be concentrated on a single computing system or distributed on a network composed of multiple computing systems. Optionally, they can be implemented by program codes executable by the computing system. Thus, they can be stored in the storage system and executed by the computing system, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module to implement. In this way, the present invention is not limited to any specific combination of hardware and software.

[0312] Embodiment 3

[0313] As Figure 4 shown, further, on the other hand, the present application also proposes an electronic device, including:

[0314] A processor;

[0315] A memory for storing processor-executable instructions;

[0316] Wherein, when the processor is configured to execute the executable instructions, it implements a method for monitoring the water intake of heart failure patients based on intelligent evaluation of water content described in Embodiment 2.

[0317] An electronic device according to an embodiment of the present disclosure includes a processor and a memory for storing processor-executable instructions. Wherein, when the processor is configured to execute the executable instructions, it implements a method for monitoring the water intake of heart failure patients based on intelligent evaluation of water content described in Embodiment 2 above.

[0318] Here, it should be noted that the number of processors can be one or more. At the same time, in the electronic device according to the embodiment of the present disclosure, an input system and an output system may also be included. Wherein, the processor, the memory, the input system and the output system can be connected through a bus or in other ways, which is not specifically limited here.

[0319] The memory, as a computer-readable storage medium, can be used to store software programs, computer-executable programs and various modules, such as: the program or module corresponding to a method for monitoring the water intake of heart failure patients based on intelligent evaluation of water content according to an embodiment of the present disclosure. The processor executes various functional applications and data processing of the electronic device by running the software programs or modules stored in the memory.

[0320] The input system can be used to receive input numbers or signals. Wherein, the signal can be a key signal related to the user settings and function control of the device / terminal / server. The output system may include a display device such as a display screen.

[0321] The embodiments of the present disclosure have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical applications or technical improvements in the market, or to enable other ordinary skill in the art in the technical field to understand the disclosed embodiments.

Claims

1. A water intake monitoring system for heart failure patients based on intelligent assessment of water content, characterized in that It includes a clinical dining carrier for carrying the food to be eaten. Among them, the following are integrated on the clinical dining carrier: A weight sensor for monitoring the weight signal of the food to be eaten and sending it to the clinical terminal; A three-dimensional camera for collecting three-dimensional image data of the food to be eaten and sending it to the clinical terminal; A clinical terminal for processing the weight signal and calculating the weight corresponding to the food to be eaten; and, identifying the type of the food to be eaten according to the three-dimensional image data, matching the water content ratio corresponding to the type of the food to be eaten, and estimating the total water content of this meal: where, i represents the food type of the food to be consumed currently, W i represents the weight of the food to be consumed currently, W total(i) represents the total weight of the food consumed in this meal, k (i) represents the water content percentage of the corresponding food type, and C represents the total water content consumed in this meal; The nurse station background is used to read the intake of heart failure patients from the clinical terminal in real time, including the food type, weight, water content ratio, total food weight and total water content of the current food to be eaten, and generate a daily intake and output report in combination with the daily output of heart failure patients; The weight sensor and the three-dimensional camera are communicatively connected to the clinical terminal; The clinical terminal is communicatively connected to the nurse station background.

2. The water intake monitoring system for heart failure patients based on intelligent assessment of water content according to claim 1, wherein The identifying the type of the food to be eaten according to the three-dimensional image data includes: Randomly extracting the food frame images from the three-dimensional image data and preprocessing them; Inputting the food frame images into the food type recognition model pre-deployed on the clinical terminal, and the food type recognition model recognizes and outputs the corresponding type and the water content ratio of the corresponding type.

3. The water intake monitoring system for heart failure patients based on intelligent evaluation of water content according to claim 2, characterized in that, The method for generating the food type recognition model includes: Pre-collecting food images of several types and preprocessing them; Using a convolutional neural network to extract the image features in the food images of various types, and labeling the corresponding type labels and water content ratio labels; Counting the image features of various types, constructing a feature set and dividing it into a training set and a validation set according to a preset ratio; Inputting the training set into a preset CNN model to train and generate the initial food type recognition model; Using the delivery validation set to verify the recognition performance of the food type recognition model: If the verification is passed, then deploy and apply the food type recognition model to the clinical terminal; Otherwise, retrain.

4. The water intake monitoring system for heart failure patients based on intelligent assessment of water content according to claim 1, wherein The clinical terminal is also used for: Generating a three-dimensional food model corresponding to the food to be eaten according to the three-dimensional image data; Performing mesh division on the three-dimensional food model to obtain several food blocks; Extracting the three-dimensional parameters of the food blocks through an image recognition algorithm: length L, width W and height H, and estimating the volume V of the food to be eaten: where Vi is the volume of the i-th food block:

5. The water intake monitoring system for heart failure patients based on intelligent assessment of water content according to claim 1, characterized in that, The nurse station background is also used for: Presetting the upper limit value and the lower limit value of the daily intake for heart failure patients in advance; Counting the daily intake according to each intake and judging whether the daily intake of heart failure patients is between the upper limit value and the lower limit value: If it is, then generate the corresponding daily intake and output report; Otherwise, generate a corresponding warning signal and send it to the clinical terminal.

6. A method for monitoring the water intake of heart failure patients based on intelligent assessment of water content, characterized in that, It includes the following steps: Start dining and place the food to be eaten on the clinical dining carrier; The weight sensor monitors the weight signal of the food to be eaten and sends it to the clinical terminal; Collect three-dimensional image data of the food to be eaten by a three-dimensional camera and send it to the clinical terminal; The clinical terminal processes the weight signal and calculates the weight corresponding to the food to be eaten; And, identify the type of the food to be eaten according to the three-dimensional image data, match the water content ratio corresponding to the type of the food to be eaten, and estimate the total water content of this meal: where i represents the food type of the food to be consumed currently, W i represents the weight of the food to be consumed currently, W total(i) represents the total weight of the food consumed in this meal, k (i) represents the water content percentage of the corresponding food type, and C represents the total water content of this meal; The background of the nurse station reads the intake of the heart failure patient each time from the clinical terminal in real time, including the food type, weight, water content ratio, total food weight and total water content of the current food to be eaten, and generates a daily intake and output report in combination with the daily output of the heart failure patient; and, count the daily intake according to each intake, and judge whether the daily intake of the heart failure patient is between the preset upper limit value and the lower limit value: If it is, generate the corresponding daily intake and output report; Otherwise, generate a corresponding warning signal and send it to the clinical terminal; The clinical terminal responds to the warning signal.

7. An electronic device, characterized in that, Includes: A processor; A memory for storing instructions executable by the processor; Wherein, when the processor is configured to execute the executable instructions, it implements a method for monitoring the water intake of heart failure patients based on intelligent evaluation of water content as claimed in claim 6.

Images