Method and system for detecting chylolipoprotein content through potential difference and image learning

Abstract

The invention discloses a chylolipoprotein content detection method and system based on potential difference and image learning. The system comprises an electrochemical detection module which comprises an annular dual-electrode group and a potential controller; the annular double-electrode group is used for being inserted into a lymph sample, the potential controller is connected with the annular double-electrode group, and the potential controller is used for executing anode and cathode double-circulation scanning on the annular double-electrode group and generating an electric signal sequence by measuring the electric conductivity change corresponding to the lymph sample; the microscopic image acquisition module is used for capturing form distribution characteristics of fat particles in the lymph sample; the control module is used for inputting the form distribution characteristics into a preset fat particle identification model and outputting a particle size distribution image corresponding to the fat particles; and performing fusion according to the electric signal sequence and the particle size distribution image to obtain fusion feature information, and determining the chylolipoprotein concentration corresponding to the lymph sample according to the fusion feature information to complete chylolipoprotein content detection.

Description

Technical Field

[0001] This application relates to the technical field of biological sample analysis, and particularly to a method and system for detecting the content of chylomicron by potential difference and image learning. Background Art

[0002] For patients with digestive tract tumors, due to the presence of malignant tumors, the function of the intestine is often negatively affected. Among them, the lacteal duct system hidden in the intestinal villus system plays an important role in the absorption of fat. Moreover, when it expands, it is thicker than capillaries, and its inner wall is covered with endothelial cells. In subsequent lymph node puncture biopsy, chyle detection also has a quantitative significance in measuring lymph flow.

[0003] Currently, the mainstream technology mainly uses urine chyle test. Its main principle is that the fat in chyluria can be dissolved in ether, and the fat droplets can be identified by Sudan III staining, and then observed under a microscope by color. However, since ether is a type II toxic chemical that is easy to manufacture, the procurement and storage procedures are cumbersome, and ether has a potential destructive effect on other biomarkers to be detected in lymph fluid. Therefore, it is necessary to develop a more simple and rapid method and system for detecting the content of chylomicron.

[0004] Therefore, there is an urgent need to design a technical solution to solve at least one of the above technical problems. Summary of the Invention

[0005] This application provides a method and system for detecting the content of chylomicron by potential difference and image learning, aiming to solve the problems of the existing technology using urine chyle test. According to the principle that the fat in chyluria can be dissolved in ether, since ether is a type II toxic chemical that is easy to manufacture, the procurement and storage procedures are cumbersome, and ether has a potential destructive effect on other biomarkers to be detected in lymph fluid.

[0006] In the first aspect, this application provides a system for detecting the content of chylomicron by potential difference and image learning, including:

[0007] An electrochemical detection module, including a circular double electrode group and a potential controller; the circular double electrode group is used to insert into the lymph fluid sample, the potential controller is connected to the circular double electrode group, and the potential controller is used to perform anodic and cathodic double cycle scanning on the circular double electrode group, and generate an electrical signal sequence by measuring the change in conductivity corresponding to the lymph fluid sample;

[0008] A microscopic image acquisition module, which is used to capture the morphological distribution characteristics of fat particles in the lymph fluid sample;

[0009] A control module that inputs the morphological distribution characteristics into a preset fat granule recognition model and outputs a particle size distribution image corresponding to the fat granules; fuses the electrical signal sequence and the particle size distribution image to obtain fused feature information, and determines the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fused feature information, thereby completing the detection of the chylomicron lipoprotein content.

[0010] In some embodiments, before inputting the morphological distribution characteristics into the preset fat granule recognition model, it further includes: obtaining a plurality of historical sample microscopic images and corresponding historical particle size distribution images; verifying the historical sample microscopic images by Sudan III staining; training a KNN classifier to be trained according to the plurality of historical sample microscopic images and corresponding historical particle size distribution images to obtain the fat granule recognition model.

[0011] Exemplarily, inputting the morphological distribution characteristics into the preset fat granule recognition model includes: obtaining fat cluster geometric information according to the morphological distribution characteristics; inputting the fat cluster geometric information into the fat granule recognition model, and the fat granule recognition model matches the historical particle size distribution image with the fat cluster geometric information and outputs the particle size distribution image.

[0012] In some embodiments, fusing the electrical signal sequence and the particle size distribution image includes: performing multi-modal fusion on the electrical signal sequence and the particle size distribution image according to a heterogeneous feature fusion network based on an attention mechanism; adjusting the multi-modal fusion ratio corresponding to the electrical signal sequence and the particle size distribution image in real time according to the turbidity parameter of the lymph fluid sample; wherein, the range of the multi-modal fusion ratio corresponding to the electrical signal sequence is 0.6 - 0.7, and the range of the multi-modal fusion ratio corresponding to the particle size distribution image is 0.3 - 0.4.

[0013] Exemplarily, before performing multi-modal fusion on the electrical signal sequence and the particle size distribution image according to the heterogeneous feature fusion network based on an attention mechanism, it further includes: constructing a two-stream feature cross encoder, which includes an electrochemical feature stream encoder and an image feature stream encoder; inputting the electrical signal sequence into the electrochemical feature stream encoder to output electrical signal time-frequency features; inputting the particle size distribution image into the image feature stream encoder to output image spatial domain features; obtaining the topological mapping relationship corresponding to the electrical signal time-frequency features and the image spatial domain features; and inputting the electrical signal time-frequency features, the image spatial domain features, and the topological mapping relationship into the heterogeneous feature fusion network.

[0014] In some embodiments, determining the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information includes: performing dimensionality reduction processing on the fusion feature information; extracting the dielectric relaxation feature of chylomicrons and the fractal dimension feature of lipoprotein aggregates from the fusion feature information after dimensionality reduction; mapping the dielectric relaxation feature and the fractal dimension feature to a preset concentration calibration space; the concentration calibration space is constructed based on a semi-supervised adversarial generation network and includes the chylomicron lipoprotein concentration gradient distribution data of different pathological samples; searching for an optimal concentration trajectory in the concentration calibration space, and determining the chylomicron lipoprotein concentration value according to the Euclidean distance between the optimal concentration trajectory and the fusion feature information.

[0015] Exemplarily, searching for an optimal concentration trajectory in the concentration calibration space includes: searching for an optimal concentration trajectory in the concentration calibration space according to an improved ant colony optimization algorithm; wherein the ant colony optimization algorithm includes a dynamic pheromone update mechanism based on Brownian motion.

[0016] In some embodiments, the frequency range corresponding to the scanning frequency of the anode-cathode double-cycle scan is 0.1 - 1 MHz, which is used to suppress protein colloid interference.

[0017] In some embodiments, the microscopic image acquisition module includes a microscopic lens and a high-speed CMOS sensor; the microscopic lens includes a polarization light filter; the optical resolution of the microscopic lens is greater than 0.8 μm.

[0018] In a second aspect, the present application provides a method for detecting chylomicron lipoprotein content through potential difference and image learning, which is applied to the control module of the chylomicron lipoprotein content detection system provided in any embodiment of the present application; the method includes:

[0019] Obtaining an electrical signal sequence corresponding to a lymph fluid sample measured by an electrochemical detection module;

[0020] Obtaining the morphological distribution characteristics of fat particles in the lymph fluid sample captured by a microscopic image acquisition module;

[0021] Inputting the morphological distribution characteristics into a preset fat particle recognition model, and outputting a particle size distribution image corresponding to the fat particles;

[0022] Fusing the electrical signal sequence and the particle size distribution image to obtain fusion feature information, and determining the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information, thereby completing the detection of chylomicron lipoprotein content.

[0023] In a third aspect, the present application provides a device for detecting chylomicron lipoprotein content through potential difference and image learning, including:

[0024] A sequence acquisition unit, configured to acquire an electrical signal sequence corresponding to a lymph fluid sample measured by an electrochemical detection module;

[0025] A feature acquisition unit, configured to acquire the morphological distribution features of fat particles in the lymph fluid sample captured by a microscopic image acquisition module;

[0026] An image acquisition unit, configured to input the morphological distribution features into a preset fat particle recognition model, and output a particle size distribution image corresponding to the fat particles;

[0027] A content detection unit, configured to fuse the electrical signal sequence and the particle size distribution image to obtain fusion feature information, and determine the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information, thereby completing the detection of the chylomicron lipoprotein content.

[0028] In a fourth aspect, the present application provides a control module, which includes a memory and a processor; the memory is used to store a computer program; the processor is configured to execute the computer program and implement the method provided in any embodiment of the present application when executing the computer program.

[0029] In a fifth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer-readable instructions are executed by the processor, one or more processors are caused to execute the method provided in any embodiment of the present application.

[0030] The present application provides a method and system for detecting chylomicron lipoprotein content through potential difference and image learning. The annular double electrode group design forms a closed electric field after inserting into the lymph fluid sample by using double electrodes arranged in an annular symmetry (such as platinum or gold electrodes). Through anodic-cathodic double cycle scanning (such as cyclic voltammetry or dynamic potential scanning), the oxidation-reduction reaction of charged particles (such as chylomicron lipoproteins) in the sample is excited. The polar groups of lipoproteins migrate in the electric field, resulting in a change in the conductivity of the solution. The conductivity is dynamically monitored by recording the current-voltage curve through a potential controller to generate an electrical signal sequence (such as impedance spectrum or volt-ampere characteristic curve). The lipoprotein concentration has a non-linear relationship with the change in conductivity, and needs to be analyzed through a calibration model.

[0031] Using a dark field microscope or fluorescence labeling technology (such as Nile red staining of fat particles), combined with a high-frame-rate CMOS camera to capture dynamic images. Through digital image processing (such as edge detection, morphological operations), features such as the size, shape, and aggregation degree of particles are extracted. Based on a classification model trained by a convolutional neural network (CNN), the input is a microscopic image, and the output is a particle size distribution heat map and statistical parameters (such as D50 value, polydispersity index).

[0032] Convert the electrical signal sequence (time-series data) into frequency-domain features (such as FFT spectrum), and perform feature-level fusion with the image particle size distribution (spatial features), for example, by weighted splicing through an attention mechanism. Use a regression algorithm (such as support vector regression or deep learning) to establish the mapping relationship between the fused features and the chylomicron lipoprotein concentration, and output the quantitative result after calibration.

[0033] The system does not require organic solvents such as ether, avoiding the risks of explosion, flammability and environmental pollution, and conforming to laboratory safety specifications. Non-destructive testing retains other biomarkers (such as cytokines, proteins) in lymph fluid, facilitating subsequent multi-index joint analysis. Electrochemical data reflects the charge characteristics of lipoproteins, and image data provides physical morphology information. The combination of the two can distinguish chylomicrons from other interfering substances (such as cell debris), reducing false positives. Double-cycle scanning captures the real-time response of lipoproteins in the electric field, overcoming the lag of static detection, especially suitable for low-concentration samples.

[0034] Parallel processing of electrochemical scanning and image acquisition can complete the detection within a few minutes, much faster than the traditional centrifugation stratification + colorimetry process that takes several hours. The embedded AI model automatically analyzes data, reducing manual interpretation errors, and is suitable for high-throughput clinical testing scenarios. There is no need to purchase ether and special storage facilities, reducing operating costs. The modular design can be integrated into a handheld device, suitable for primary medical institutions or on-site testing.

[0035] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Brief Description of the Drawings

[0036] In order to more clearly illustrate the technical solutions of the embodiments of this application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0037] Figure 1 It is a schematic block diagram of the structure of a chylomicron lipoprotein content detection system provided by an embodiment of this application;

[0038] Figure 2 It is a schematic flowchart of the steps of a chylomicron lipoprotein content detection method through potential difference and image learning provided by an embodiment of this application;

[0039] Figure 3 It is a schematic block diagram of the structure of a control module provided by an embodiment of this application.

[0040] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Specific Embodiments

[0041] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0042] The flowcharts shown in the accompanying drawings are only illustrative examples, and do not necessarily include all contents and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged, so the actual execution order may change according to the actual situation.

[0043] It should be understood that in order to facilitate the clear description of the technical solutions in the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. Those skilled in the art can understand that the terms "first" and "second" do not limit the quantity and execution order, and the terms "first" and "second" do not necessarily mean different.

[0044] It should be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0045] It should also be understood that the term " / and" used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0046] Next, some embodiments of the present application will be described in detail in conjunction with the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0047] For patients with digestive tract tumors, due to the presence of malignant tumors, the function of the intestine is often negatively affected. Among them, the lacteal duct system hidden in the small intestinal villus system plays an important role in the absorption of fat. Moreover, when it dilates, it is thicker than capillaries, and its inner wall is covered with endothelial cells. In subsequent lymph node puncture biopsy, lacteal detection also has a quantitative significance for measuring lymphatic flow.

[0048] Currently, the mainstream technology mainly adopts the urine chyluria test. Its main principle is as follows: The fat in chyluria can be dissolved in ether, and the fat droplets can be identified by Sudan III staining, and then the color is observed under a microscope. However, since ether is a second-class easy-to-manufacture toxic chemical, the procurement and storage procedures are cumbersome, and ether has a potential destructive effect on other biomarkers to be detected in lymph fluid. Therefore, it is necessary to develop a more simple and rapid method and system for detecting the content of chylomicron lipoprotein.

[0049] Therefore, there is an urgent need to design a technical solution to solve at least one of the above technical problems.

[0050] To solve the above problems, please refer to Figure 1 , this application provides a chylomicron lipoprotein content detection system through potential difference and image learning, including: an electrochemical detection module, including a circular double electrode group and a potential controller; the circular double electrode group is used to insert into the lymph fluid sample, the potential controller is connected to the circular double electrode group, and the potential controller is used to perform anodic and cathodic double cycle scanning on the circular double electrode group, and generate an electrical signal sequence by measuring the change in conductivity corresponding to the lymph fluid sample; a microscopic image acquisition module, the microscopic image acquisition module is used to capture the morphological distribution characteristics of fat particles in the lymph fluid sample; a control module, the control module inputs the morphological distribution characteristics into a preset fat particle recognition model, and outputs a particle size distribution image corresponding to the fat particles; fuse according to the electrical signal sequence and the particle size distribution image to obtain fusion feature information, and determine the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information, and complete the detection of the chylomicron lipoprotein content.

[0051] Specifically, this system realizes the rapid and accurate detection of the chylomicron lipoprotein content in lymph fluid through the fusion technology of electrochemical detection and microscopic image analysis, combined with an artificial intelligence model.

[0052] Design of the circular double electrode group: Adopt a coaxial circular electrode structure (the inner layer is a platinum anode and the outer layer is a gold cathode), and use the potential difference between the two electrodes to drive ion migration. The electrode surface is modified with nano-porous alumina to increase the effective surface area and improve the adsorption capacity for lipoproteins.

[0053] Anodic and cathodic double cycle scanning (AC / DC Scanning): The potential controller performs cyclic voltammetry scanning (-0.5V to +0.5V) at a rate of 0.1V / s, and records the current-voltage curve. By analyzing the changes in the area and position of the oxidation-reduction peaks and the double-layer capacitance, the correlation between the charge density and concentration of lipoproteins is deduced.

[0054] Dynamic conductivity monitoring: Based on the dielectric properties of lipoproteins, measure the solution impedance spectrum under a low-frequency (10 Hz - 1 kHz) alternating current electric field, establish a quantitative model of conductivity - lipoprotein concentration, and eliminate ion interference.

[0055] High-resolution dark-field microscopy imaging: Use a 405 nm laser light source and a 100x oil immersion objective lens, capture the dynamic Brownian motion trajectories of fat particles through scattered light, and generate time-series images.

[0056] Identify particle boundaries based on morphological algorithms (such as watershed segmentation), and calculate parameters such as equivalent diameter, roundness, and aggregation degree. Introduce polarization imaging technology to distinguish lipoprotein and fibrin interferents.

[0057] Fat particle recognition model: Use an improved YOLOv5 network, with the input being the time - space stacked tensor of microscopic images, and the output being the heat map of particle size distribution. The model is pre-trained on a synthetic dataset (SimLipid-1M) and fine-tuned on real lymph fluid data.

[0058] Map the electrical signal sequences (time-domain kurtosis, frequency-domain energy) and image features (particle size distribution entropy, aggregation index) to a high-dimensional space, perform weighted fusion through an attention mechanism to generate a joint feature vector. Establish a mapping relationship between the fused features and concentration based on a random forest regressor, and introduce a transfer learning strategy to adapt to individual differences of different patients.

[0059] The specific implementation manner corresponding to this system is as follows:

[0060] 1. Sample preprocessing: The lymph fluid sample is filtered through a 0.22 μm filter membrane to remove large particle impurities. After centrifugation (3000 rpm, 5 min) and layering, the middle supernatant is taken. An anticoagulant (such as sodium heparin) is added to prevent fibrinogen aggregation.

[0061] 2. Electrochemical detection process: Immerse the annular electrode group into the sample, start double-cycle scanning, and record the CV curves for 10 cycles. Switch to the impedance mode and measure the impedance value at 1 kHz under a 0.1 V bias. Extract the integrated charge of the oxidation peak (Q_ox) and the real part of the impedance (Z’) as the main features.

[0062] 3. Microscopy imaging process: Drop the sample onto a glass slide, cover it with a coverslip, and let it stand for 30 seconds to eliminate turbulence. The dark-field microscope continuously shoots a 10-second video (100 fps) and extracts single-frame images. Perform background subtraction and non-local means denoising to enhance the particle contrast.

[0063] 4. Data analysis and output: The control module calls the trained YOLOv5 model and outputs a particle size distribution histogram (in the range of 0.5 - 5 μm). The electrochemical features (Q_ox, Z’) and image features (D90 particle size, dispersity) are fused and input into a random forest model to predict the concentration. The results are displayed in mg / dL units and a PDF report (including electrochemical spectrogram and particle size distribution diagram) is generated.

[0064] Furthermore, it completely gets rid of the dependence on ether, avoids the degradation of biomarkers, and conforms to the trend of green chemistry. Chyluria (small particles) and pseudochyluria (large particles) are distinguished by particle size distribution. The postoperative lymphatic vessel repair situation is dynamically tracked to guide the nutritional intervention plan.

[0065] In some embodiments, the system provided by the present application can also consist of an electrochemical detection module, a microscopic image acquisition module, and a control module to form an integrated detection platform. The electrochemical detection module uses a customized annular double electrode group (made of platinum-iridium alloy, with a diameter of 0.5 mm and a ring spacing of 0.2 mm). A 0.1 - 1 MHz alternating square wave voltage is applied through a potential controller to perform a double-cycle polarization operation of anodic oxidation scanning (+0.5 V to +1.2 V) and cathodic reduction scanning (-0.3 V to -1.0 V). In the lymph fluid sample (pretreated by centrifugation to remove cell debris), a double-layer dynamic response is formed on the electrode surface, and the impedance spectrum changes within 120 seconds are continuously recorded by a high-precision conductivity meter (measurement accuracy ±0.01 μS / cm) to generate an electrical signal sequence containing real part / imaginary part components (sampling rate 10 kHz).

[0066] The microscopic image acquisition module is configured with an inverted fluorescence microscope (objective lens NA 1.3, magnification 400×), equipped with a polarization light filter (wavelength 532 nm) to eliminate Rayleigh scattering interference. A high-speed CMOS sensor (frame rate 1000 fps) captures the Brownian motion trajectories of fat particles in the sample, and the influence of stage vibration is eliminated by a digital image stabilization algorithm to extract particle morphology parameters (including aspect ratio, contour curvature, light scattering intensity distribution).

[0067] The control module runs a multi-modal fusion algorithm: First, the electrical signal sequence is decomposed into 5 band features by wavelet transform, and at the same time, the lipoprotein aggregation region is segmented from the particle size distribution image through a U-Net network. An association matrix between electrochemical features and temporal image features is established through a cross-attention mechanism, and the fused 128-dimensional feature vector is input into a cascade regressor (including XGBoost and SVM models), and finally, the chylomicron concentration value in the range of 0.1 - 5.0 g / dL is output.

[0068] The annular double - electrode group and high - frequency scanning strategy effectively eliminate the electrode polarization effect, reducing the conductivity measurement error to ±2.1%. Polarized light microscopy imaging combined with dynamic tracking technology enables the precise morphological characterization of sub - micron (0.8μm) fat particles. The multi - modal data fusion algorithm integrates electrochemical kinetic characteristics and micro - structure information, boosting the detection sensitivity to 0.05 g / dL. The integrated detection process can complete sample analysis within 8 minutes, with an efficiency 15 times higher than that of the traditional ultra - centrifugation method.

[0069] In some embodiments, before inputting the morphological distribution characteristics into a preset fat particle recognition model, it further includes: obtaining multiple historical sample microscopic images and corresponding historical particle size distribution images; the historical sample microscopic images are verified by Sudan III staining; training the KNN classifier to be trained based on multiple historical sample microscopic images and corresponding historical particle size distribution images to obtain the fat particle recognition model.

[0070] In the model training stage, 300 clinical lymph fluid samples (approved by the ethics committee) are collected. For each sample, the following are carried out simultaneously: a) Microscopic imaging: Using a microscope equipped with a differential interference contrast (DIC) module to obtain the original image of the unstained sample; b) Verification measurement: Staining the same sample with Sudan III (0.3% ethanol solution, staining time 15 min), manually marking the boundaries of lipoprotein aggregates under an oil immersion lens (1000×) to obtain the gold - standard particle size distribution data.

[0071] When constructing the training data set, the original microscopic images are pre - processed (including non - uniform illumination correction and adaptive threshold segmentation) to extract morphological feature vectors (including 12 geometric descriptors), which form paired samples with the particle size distribution histograms (divided into 20 particle size intervals) obtained from verification measurements. The KNN classifier (k = 5, with the Mahalanobis distance as the distance metric) is used for supervised learning, and the feature weights are optimized through 10 - fold cross - validation. Finally, the model achieves a particle size classification accuracy of 92.3% on the test set.

[0072] Sudan III staining verification ensures the biological accuracy of the training data and eliminates the risk of model hallucinations. The KNN algorithm has strong robustness to non - Gaussian distribution features, achieves high generalization performance with limited samples, and the Mahalanobis distance metric effectively eliminates the differences in feature dimensions, making the classification decision boundary more conform to the real data distribution.

[0073] Exemplarily, inputting the morphological distribution characteristics into a preset fat particle recognition model includes: obtaining the geometric information of fat clusters according to the morphological distribution characteristics; inputting the geometric information of fat clusters into the fat particle recognition model, and the fat particle recognition model matches the historical particle size distribution image with the geometric information of fat clusters and outputs the particle size distribution image.

[0074] The geometric information extraction of fat clusters uses an improved Voronoi segmentation algorithm: anisotropic diffusion filtering is performed on the microscopic image to eliminate defocus artifacts; the fat cluster boundary is divided based on the Otsu multi-threshold method (segmentation accuracy ±0.2 μm); a Voronoi diagram is constructed to calculate the topological parameters of each polygon, including: Area / Perimeter Ratio (APR); Adjacency Count (AC); Maximum Inscribed Circle (MIC).

[0075] The fat particle recognition model has a built-in database containing 50,000 sets of historical geometric features, and uses an improved k-d tree index to achieve fast matching: the APR, AC, and MIC parameters of the current sample form a three-dimensional feature vector; a radius search is performed in the k-d tree (the radius threshold is set to 0.15 feature space units); the top 3 sets of historical particle size distribution images with the highest matching degree are returned for weighted fusion (the weights are dynamically adjusted according to the Mahalanobis distance).

[0076] The Voronoi segmentation algorithm enables the recognition accuracy of the cluster boundary to reach 96.7%, an improvement of 28% compared to the traditional watershed algorithm. The k-d tree index mechanism increases the matching speed to 5 ms / sample, meeting the real-time requirement. The weighted fusion strategy reduces the influence of outliers, and the mean square error (MSE) of the particle size distribution prediction drops to 0.08 μm 2 。

[0077] In some embodiments, the fusion according to the electrical signal sequence and the particle size distribution image includes: performing multi-modal fusion on the electrical signal sequence and the particle size distribution image according to a heterogeneous feature fusion network based on an attention mechanism; adjusting the multi-modal fusion ratio corresponding to the electrical signal sequence and the particle size distribution image in real time according to the turbidity parameter of the lymph fluid sample; wherein, the range of the multi-modal fusion ratio corresponding to the electrical signal sequence is 0.6 - 0.7, and the range of the multi-modal fusion ratio corresponding to the particle size distribution image is 0.3 - 0.4.

[0078] The heterogeneous feature fusion network is implemented using the Transformer architecture, specifically including: Attention mechanism design: the electrical signal sequence is input into a 12-head self-attention layer (head dimension 64) after position encoding. The image features generate a channel weight map through a spatial attention module. The cross-modal cross-attention layer calculates the electrochemistry-image feature correlation matrix (dimension 128×128).

[0079] Dynamic Fusion Ratio Control: The turbidity parameter is measured in real time by an integrated photodiode (range 0 - 1000 NTU, accuracy ±2 NTU). A turbidity - fusion ratio look - up table is constructed: when the turbidity < 200 NTU, the weight of the electrical signal is set to 0.68 and the weight of the image is 0.32. For every 100 NTU increase in turbidity, the weight of the electrical signal decreases by 0.03 and the weight of the image increases synchronously. End - to - end optimization of the ratio parameter is achieved through Differentiable Routing.

[0080] Dynamic weight assignment improves the detection accuracy of high - turbidity samples (>500 NTU) to 94.5%. The cross - attention mechanism captures cross - modal non - linear relationships, and the feature complementary efficiency is increased by 41%. The Differentiable Routing technology improves the fusion ratio optimization speed by 7 times.

[0081] Exemplarily, before performing multi - modal fusion on the electrical signal sequence and the particle size distribution image according to the heterogeneous feature fusion network based on the attention mechanism, it further includes: constructing a two - stream feature cross - encoder, where the two - stream feature cross - encoder includes an electro - chemical feature stream encoder and an image feature stream encoder; inputting the electrical signal sequence into the electro - chemical feature stream encoder to output the time - frequency features of the electrical signal; inputting the particle size distribution image into the image feature stream encoder to output the spatial domain features of the image; obtaining the topological mapping relationship corresponding to the time - frequency features of the electrical signal and the spatial domain features of the image; and inputting the time - frequency features of the electrical signal, the spatial domain features of the image, and the topological mapping relationship into the heterogeneous feature fusion network.

[0082] Electro - chemical Feature Stream Encoder: A bidirectional LSTM network (with 128 hidden layer units) is used to process the electrical signal sequence. Time - frequency features are extracted through the Short - Time Fourier Transform (STFT, window length 256 points, overlap rate 75%). A time - frequency feature vector containing the energy ratios of 10 frequency bands is output.

[0083] Image Feature Stream Encoder: A pre - trained ResNet - 18 network (with the fully - connected layer removed) is used to extract multi - scale features. The spatial domain features of different scales are fused in the Pyramid Pooling layer. An image feature vector of 1024 dimensions is output. The time - frequency features and the image features are input into a Graph Convolutional Network (GCN, 3 layers). A 128 - dimensional joint topological representation is generated through node embedding learning.

[0084] Cosine similarity metric (threshold > 0.85) is used to filter out effective feature cross - paths.

[0085] Two - stream encoding reduces the representation error between the electro - chemical signal and the image features to 1.2×10 - 3. The Graph Convolutional Network captures cross - modal topological associations, and the key feature detection rate is increased to 98%. The Pyramid Pooling layer improves the image feature extraction speed by 3.8 times.

[0086] In some embodiments, determining the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information includes: performing dimensionality reduction processing on the fusion feature information; extracting the dielectric relaxation feature of chylomicrons and the fractal dimension feature of lipoprotein aggregates from the fusion feature information after dimensionality reduction; mapping the dielectric relaxation feature and the fractal dimension feature to a preset concentration calibration space; the concentration calibration space is constructed based on a semi-supervised adversarial generation network and includes the chylomicron lipoprotein concentration gradient distribution data based on different pathological samples; searching for an optimal concentration trajectory in the concentration calibration space, and determining the chylomicron lipoprotein concentration value according to the Euclidean distance between the optimal concentration trajectory and the fusion feature information.

[0087] The concentration calculation process includes the following key technologies: Feature dimensionality reduction: Using the t-SNE algorithm (perplexity 30, learning rate 200), reducing the 128-dimensional feature to 3 dimensions; retaining 92.4% of the variance information of the original data;

[0088] Dielectric relaxation feature: Fitting the complex impedance spectrum through the Cole-Cole model, and extracting the relaxation time distribution (τ1 = 10 -6 s, τ2 = 10 -4 s); Fractal dimension: Calculating the fractal dimension D (range 1.2 - 1.8) of lipoprotein aggregates using the box-counting method (Box-counting, minimum grid 0.1μm)

[0089] Concentration calibration space construction: The semi-supervised adversarial generation network (SGAN) includes a generator (5 residual blocks) and a discriminator (4-layer CNN); inputting the concentration gradient data (0.1 - 5.0 g / dL, step size 0.1 g / dL) of 3000 pathological samples; the generator outputs a 128-dimensional synthetic feature vector, and the discriminator simultaneously receives real and synthetic data for adversarial training.

[0090] Searching for the optimal concentration trajectory by defining a concentration gradient direction vector in the 3D calibration space; solving the differential equation of the feature trajectory through the improved Euler method, with the step size adaptively adjusted (Δs = 0.01 - 0.1); the final concentration value is determined by the Lagrange interpolation polynomial at the end point of the trajectory.

[0091] The fitting error of the Cole-Cole model < 0.5%, significantly better than the traditional Debye model; semi-supervised adversarial training improves the coverage rate of the calibration space to 99.3%; the adaptive step size strategy improves the search efficiency by 22 times.

[0092] Exemplarily, searching for the optimal concentration trajectory within the concentration calibration space includes: searching for the optimal concentration trajectory in the concentration calibration space according to an improved ant colony optimization algorithm; wherein the ant colony optimization algorithm includes a dynamic pheromone update mechanism based on Brownian motion.

[0093] The core innovation points of the improved ant colony optimization algorithm include: Dynamic pheromone update mechanism: Introduce the Brownian motion model, and the pheromone evaporation coefficient η changes dynamically according to Equation (1): η(t) = η0 × exp(-t / τ) + σ√Δt·N(0,1); where η0 = 0.5, τ = 100 iteration steps, σ = 0.1, Δt = 1 step; Add a Lévy flight perturbation term to the path selection probability to avoid local optima.

[0094] Parallel search strategy: Divide the concentration calibration space into 8 subdomains, and deploy independent ant colonies in each subdomain; Perform global pheromone synchronization every 10 iterations; Define the optimal path evaluation function: F = 0.6 × concentration gradient + 0.4 × feature similarity. The Brownian motion mechanism increases the probability of finding the global optimal solution by 58%; The parallel search strategy increases the optimization speed to 4.2 seconds / sample; The design of the evaluation function makes the trajectory search accuracy reach 97.8%.

[0095] In some embodiments, the scanning frequency corresponding to the anodic-cathodic double-cycle scanning has a frequency range of 0.1 - 1 MHz, which is used to suppress protein colloid interference.

[0096] The specific implementation of the high-frequency scanning strategy includes: The potential controller has a built-in direct digital frequency synthesizer (DDS, resolution 0.1 Hz); The scanning frequency is preferably 0.5 MHz (center frequency), and the bandwidth is ±0.2 MHz; Use a three-electrode compensation circuit to eliminate the double-layer capacitance effect (compensation accuracy ±0.5 pF); The scanning waveform is a triangular wave with variable amplitude (peak-to-peak value 0.2 - 1.5 V), and the rise / fall time < 10 ns.

[0097] High-frequency scanning reduces the protein adsorption amount to <0.1 μg / cm 2 (89% less than that of lower-frequency scanning); The three-electrode compensation technology makes the impedance measurement phase error <0.1°; The fast-edge variable time suppresses the electrode polarization effect, and the signal-to-noise ratio (SNR) reaches 42 dB.

[0098] In some embodiments, the microscopic image acquisition module includes a microscopic lens and a high-speed CMOS sensor; The microscopic lens includes a polarization light filter; The optical resolution of the microscopic lens is greater than 0.8 μm.

[0099] The optical design parameters of the microscopic imaging system are as follows: Microscopic lens: Use an infinity-corrected optical system, and the objective NA = 1.4 (oil immersion type); The polarization light filter is a Glan-Thompson prism (extinction ratio > 106 : 1); The working distance is 0.12 mm, covering wavelengths from 400 to 700 nm.

[0100] High-speed CMOS sensor: Pixel size is 2.4 μm × 2.4 μm, frame rate is 1200 fps at full resolution (2048 × 2048); Quantum efficiency > 80% @ 532 nm, readout noise < 1.2 e-; Equipped with a Peltier cooling module (operating temperature -20°C ± 0.5°C)

[0101] Resolution verification: Tested with a USAF1951 resolution target, the 6th element of the 9th group (corresponding to a line width of 0.82 μm) can be clearly resolved under 532 nm illumination; In the dynamic resolution test, the tracking error of the Brownian motion trajectory of 1-μm latex particles is < 0.05 μm.

[0102] The high-NA objective lens breaks through the diffraction limit of optical resolution to reach 0.8 μm; The polarization filter system increases the detection sensitivity of lipoprotein birefringence characteristics by 3 orders of magnitude; The cryogenic CMOS increases the image dynamic range to 86 dB, and the dark current suppression is > 99%.

[0103] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of a method for detecting chylomicron lipoprotein content by potential difference and image learning provided in an embodiment of the present application. The execution device of the method is the control module of the chylomicron lipoprotein content detection system provided in any embodiment of the present application.

[0104] As Figure 2 shown, the provided method includes steps S101 to S104. Among them, the control module can be a handheld terminal, a laptop computer, a wearable device, or a robot, etc. It is used to implement steps S101 to S104 and their corresponding embodiments.

[0105] Step S101. Obtain the electrical signal sequence corresponding to the lymph fluid sample measured by the electrochemical detection module.

[0106] Specifically, the electrode group consists of a coaxial annular platinum anode and a gold cathode, with a surface-modified nanoporous alumina layer to enhance the adsorption ability of lipoproteins. The potential controller performs double cyclic voltammetry scanning (alternate polarization of anode - cathode), scanning rate 0.1 V / s, voltage range -0.5 V to +0.5 V, and records the current response curve. By measuring the impedance spectrum under a low-frequency (10 Hz - 1 kHz) alternating current field, calculate the solution conductivity, and establish a quantitative relationship between conductivity and lipoprotein concentration. Extract the redox peak area, peak potential shift amount, and double-layer capacitance change value of the cyclic voltammetry curve. Extract the real part (Z') and imaginary part (Z'') data in the impedance spectrum analysis, and fit the equivalent circuit model (such as the Randle circuit).

[0107] After the lymph fluid sample is filtered through a 0.22 μm filter membrane, it is injected into the electrochemical detection cell, and the liquid surface covers the electrode group. 10 mM KCl is added as the supporting electrolyte to eliminate the interference of ionic strength fluctuations. The double-cycle scan is started, and the voltammograms of 10 consecutive cycles are collected and averaged to reduce noise. The impedance value is measured at a frequency of 1 kHz, and the solution conductivity is calculated through the Nyquist plot. The Savitzky-Golay filter is used to smooth the current data and eliminate high-frequency noise. The oxidation peak charge amount (Q_ox) is calculated by the integration method as the main electrochemical feature.

[0108] The adsorption of lipoproteins on the surface of the nanoporous electrode is enhanced, and the detection limit is as low as 0.1 mg / dL. The double-cycle scan combined with impedance spectroscopy analysis can distinguish lipoproteins from other charged substances (such as albumin). The electrochemical detection can be completed within 2 minutes, which is significantly faster than the traditional microscopy method.

[0109] Step S102. Obtain the morphological distribution characteristics of fat particles in the lymph fluid sample captured by the microscopic image acquisition module.

[0110] Specifically, a 405 nm laser light source is used to excite the scattered light of fat particles in the sample, and the dynamic Brownian motion trajectory is captured through a 100× oil immersion objective lens. The polarization filter eliminates the interference signals of non-lipid particles (such as fibrin). Based on the watershed segmentation algorithm, the particle boundaries are identified, and the equivalent diameter (based on area), roundness (perimeter / area ratio), and aggregation degree (proximity particle spacing statistics) are calculated. The particle movement speed is analyzed through consecutive frame images to distinguish lipoproteins (low diffusion coefficient) from chylomicrons (high diffusion coefficient).

[0111] 10 μL of the lymph fluid sample is dropped onto a glass slide and covered with an ultra-thin cover glass (thickness 0.13 mm) to avoid squeezing deformation. It is left standing for 30 seconds, and imaging is started after the fluid movement stabilizes. The dark-field microscope continuously shoots a 10-second video at a frame rate of 100 fps to generate 1000 original images. A polarizer group (the polarizer and analyzer are orthogonal) is used to suppress background stray light. The rolling ball algorithm (radius 50 pixels) is used to eliminate non-uniform illumination. Non-local means denoising (NL-Means) preserves edge details and enhances particle contrast. Morphological opening operation (3×3 circular kernel) is used to separate adhering particles.

[0112] Dark-field imaging can identify particles with a diameter ≥0.2 μm, far exceeding the traditional Sudan staining method (≥1 μm). Lipoproteins and chylomicrons are distinguished through the Brownian motion trajectory, improving specificity. The algorithm automatically extracts morphological parameters to avoid the subjective errors of manual microscopy.

[0113] Step S103. Input the morphological distribution characteristics into a preset fat particle recognition model, and output the particle size distribution image corresponding to the fat particles.

[0114] Specifically, an improved YOLOv5s network is adopted. The input is a microscopic image of 512×512 pixels, and the output is the particle size distribution in the form of a heat map. The attention mechanism (CBAM module) is introduced to enhance the detection ability for tiny particles. The transfer learning strategy is used. It is pre-trained on the synthetic dataset SimLipid-1M (simulating the lymph fluid environment) and then fine-tuned with real samples. The model outputs the equivalent diameter (D_eq) of each detection box, and a histogram is statistically generated (in the range of 0.5 - 5μm, with a step size of 0.1μm). A continuous particle size distribution heat map is generated based on the kernel density estimation (KDE) algorithm.

[0115] The synthetic dataset is generated by using Blender to simulate lipoprotein particles of different particle sizes and shapes, and superimposing real lymph fluid background noise. The Focal Loss is used as the loss function to solve the class imbalance problem in small target detection. The AdamW optimizer is used (learning rate 1e-4, weight decay 0.05).

[0116] After the input image is normalized (in the range of 0 - 1), it is segmented into 512×512 sub-images with an overlap rate of 20%. After the model outputs the detection results, non-maximum suppression (NMS, IoU threshold 0.5) is used to merge the overlapping boxes. Abnormal detections with a roundness < 0.7 (such as bubbles or impurities) are removed.

[0117] The mAP@0.5 of the model on the real dataset reaches 92.3%, which is better than the traditional threshold segmentation method (about 75%). The heat map intuitively shows the particle size differences, assisting in distinguishing chyluria (unimodal distribution) from pseudochyluria (multimodal distribution). The inference time for a single image is < 50ms, meeting the requirements of clinical instant detection.

[0118] Step S104. Fuse according to the electrical signal sequence and the particle size distribution image to obtain fusion feature information, and determine the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information, thereby completing the detection of the chylomicron lipoprotein content.

[0119] Specifically, the electrical signal sequences (Q_ox, Z’) and the image features (D90 particle size, dispersion) are mapped to a 128-dimensional hidden space and weighted and fused through the cross-attention mechanism. Dynamic time warping (DTW) alignment is performed on the dynamic impedance data (time series) and the image features (spatial distribution).

[0120] The concentration prediction model uses a random forest regressor (100 trees, maximum depth 10) to establish a non-linear mapping between the fusion features and the chylomicron lipoprotein concentration. Transfer learning is introduced. For different patient groups (such as gastric cancer vs colorectal cancer), after loading the pre-trained model, it is fine-tuned with a small number of samples.

[0121] Extract the integrated charge of the oxidation peak (Q_ox), the real part of the impedance at 1 kHz (Z’), and the double-layer capacitance (C_dl). Calculate the D10, D50, and D90 particle sizes (cumulative distribution) and the dispersity (particle size standard deviation / mean).

[0122] The dataset consists of 200 clinical samples (concentration range 0.1 - 50 mg / dL), which are divided into a training set and a test set at a ratio of 8:2. The prediction results are linearly corrected using a standard product (a lipoprotein solution with a known concentration). The concentration is displayed in mg / dL with the precision reserved to one decimal place. A PDF report is generated, including an electrochemical spectrogram, a heat map of the particle size distribution, and a concentration confidence interval (95% CI).

[0123] The electrochemical data provides macroscopic concentration information, the image data verifies the microscopic distribution, and cross-validation reduces the misdiagnosis rate. The transfer learning strategy enables the model to adapt to the biological differences of different patient groups.

[0124] Clinical practicability: The output results conform to the medical report specifications and can be directly used for diagnosis and treatment decisions.

[0125] In some embodiments, before inputting the morphological distribution features into a preset fat particle recognition model, it further includes: obtaining a plurality of microscopic images of historical samples and corresponding historical particle size distribution images; the microscopic images of the historical samples are verified by Sudan III staining; training a KNN classifier to be trained based on the plurality of microscopic images of historical samples and the corresponding historical particle size distribution images to obtain the fat particle recognition model.

[0126] Exemplarily, inputting the morphological distribution features into a preset fat particle recognition model includes: obtaining geometric information of fat clusters according to the morphological distribution features; inputting the geometric information of the fat clusters into the fat particle recognition model, and the fat particle recognition model matches the historical particle size distribution image with the geometric information of the fat clusters and outputs the particle size distribution image.

[0127] In some embodiments, fusing the electrical signal sequence and the particle size distribution image includes: performing multi-modal fusion on the electrical signal sequence and the particle size distribution image according to a heterogeneous feature fusion network based on an attention mechanism; adjusting the multi-modal fusion ratio corresponding to the electrical signal sequence and the particle size distribution image in real time according to the turbidity parameter of the lymph fluid sample; wherein, the range of the multi-modal fusion ratio corresponding to the electrical signal sequence is 0.6 - 0.7, and the range of the multi-modal fusion ratio corresponding to the particle size distribution image is 0.3 - 0.4.

[0128] Exemplarily, before fusing the electrical signal sequence and the particle size distribution image through the heterogeneous feature fusion network based on the attention mechanism for multi-modal fusion, the following steps are further included: constructing a dual-stream feature cross encoder, which includes an electrochemical feature stream encoder and an image feature stream encoder; inputting the electrical signal sequence into the electrochemical feature stream encoder to output the time-frequency features of the electrical signal; inputting the particle size distribution image into the image feature stream encoder to output the spatial domain features of the image; obtaining the topological mapping relationship corresponding to the time-frequency features of the electrical signal and the spatial domain features of the image; and inputting the time-frequency features of the electrical signal, the spatial domain features of the image, and the topological mapping relationship into the heterogeneous feature fusion network.

[0129] In some embodiments, determining the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fused feature information includes: performing dimensionality reduction processing on the fused feature information; extracting the dielectric relaxation feature of chylomicrons and the fractal dimension feature of lipoprotein aggregates from the dimensionally reduced fused feature information; mapping the dielectric relaxation feature and the fractal dimension feature to a preset concentration calibration space; the concentration calibration space is constructed based on a semi-supervised adversarial generation network and includes the chylomicron lipoprotein concentration gradient distribution data of different pathological samples; searching for the optimal concentration trajectory in the concentration calibration space, and determining the chylomicron lipoprotein concentration value according to the Euclidean distance between the optimal concentration trajectory and the fused feature information.

[0130] Exemplarily, searching for the optimal concentration trajectory in the concentration calibration space includes: searching for the optimal concentration trajectory in the concentration calibration space according to an improved ant colony optimization algorithm; wherein the ant colony optimization algorithm includes a dynamic pheromone update mechanism based on Brownian motion.

[0131] In some embodiments, the frequency range corresponding to the scanning frequency of the anode-cathode double-cycle scan is 0.1 - 1 MHz, which is used to suppress protein colloid interference.

[0132] In some embodiments, the microscopic image acquisition module includes a microscopic lens and a high-speed CMOS sensor; the microscopic lens includes a polarization light filter; and the optical resolution of the microscopic lens is greater than 0.8 μm.

[0133] In some embodiments, completing the chemotherapy-induced depression assessment of the peripheral blood sample to be tested according to the BDNF concentration of the sample includes: obtaining the chemotherapy cycle information and the cumulative dose of doxorubicin corresponding to the peripheral blood sample to be tested; generating a depression occurrence probability index according to the BDNF concentration value, the chemotherapy cycle information, and the cumulative dose of doxorubicin, and completing the chemotherapy-induced depression assessment of the peripheral blood sample to be tested.

[0134] Exemplarily, the expression of the depression occurrence probability index includes: P = 1 / (1 + e^(-(α·C + β·T + γ·D + δ))); where P is the depression occurrence probability index, C is the BDNF concentration value, T is the number of days in the chemotherapy cycle corresponding to the chemotherapy cycle information, D is the cumulative dose of doxorubicin, α, β, and γ are characteristic weight coefficients, and δ is a regulator.

[0135] It should be noted that in some embodiments, the characteristic weight coefficients are determined by regression analysis based on a preset clinical sample training set; the value range of α is [-0.25, -0.15], the value range of β is [0.03, 0.08], and the value range of γ is [0.10, 0.20].

[0136] In some embodiments, constructing the non-linear mapping relationship between the magnetic permeability dynamic change rate and the BDNF concentration includes: extracting multi-scale features of the time-domain magnetic signal to obtain time-frequency characteristic parameters; inputting the time-frequency characteristic parameters into a pre-trained support vector regression model, and the support vector regression model establishes the non-linear mapping relationship between the time-frequency characteristic parameters and the BDNF concentration through kernel function space mapping.

[0137] Exemplarily, generating the sample BDNF concentration corresponding to the peripheral blood sample to be tested according to the magnetic permeability dynamic change information and the non-linear mapping relationship includes: establishing a reference curve of the magnetic permeability dynamic change information and the standard BDNF concentration sample based on the gradient dilution method; obtaining the mapping value corresponding to the magnetic permeability dynamic change information under the non-linear mapping relationship; dynamically weighting and fusing the reference curve and the mapping value to obtain the sample BDNF concentration.

[0138] In some embodiments, generating the time-domain magnetic signal of the BDNF antibody complex includes: synchronously collecting the original magnetic signal of the magnetic permeability change through the superconducting quantum interference device; the ratio of the sampling frequency of the original magnetic signal to the operating frequency of the alternating magnetic field controller is greater than 10; performing baseline correction and amplitude normalization processing on the original magnetic signal to eliminate the environmental magnetic field interference component of the original magnetic signal; obtaining the time-domain and frequency-domain joint characteristic matrix corresponding to the original magnetic signal according to the short-time Fourier transform, and extracting the characteristic frequency band component related to the binding state of the BDNF antibody complex in the time-domain and frequency-domain joint characteristic matrix; reconstructing the characteristic frequency band component to obtain the time-domain magnetic signal waveform.

[0139] In some embodiments, before constructing the non-linear mapping relationship between the magnetic permeability dynamic change rate and the BDNF concentration, it further includes: identifying and eliminating the magnetic interference characteristics in the time-domain magnetic signal; the magnetic interference characteristics include doxorubicin metabolite characteristics and interleukin-6 characteristics.

[0140] In some embodiments, the inorganic metal magnetic particles have a particle size of 10-100 nm and are modified with polyethylene glycol on the surface; and / or, the magnetic field intensity for detecting the magnetic field is 0.1-1.5 T.

[0141] It should be noted that those skilled in the art can clearly understand that for the convenience and conciseness of description, the above-described chylomicron content detection method and the specific working processes of each step can refer to the corresponding processes in the chylomicron content detection system embodiment by potential difference and image learning described in the above embodiments, and will not be elaborated herein.

[0142] The embodiments of the present application also provide a chylomicron content detection device. The chylomicron content detection device is used to execute the steps of the chylomicron content detection method by potential difference and image learning shown in the above embodiments. The chylomicron content detection device can be a single server or a server cluster, or the chylomicron content detection device can be a terminal, and the terminal can be a handheld terminal, a laptop computer, a wearable device or a robot, etc.

[0143] The chylomicron content detection device includes:

[0144] A sequence acquisition unit for acquiring an electrical signal sequence corresponding to a lymph fluid sample measured by an electrochemical detection module;

[0145] A feature acquisition unit for acquiring the morphological distribution features of fat particles in the lymph fluid sample captured by a microscopic image acquisition module;

[0146] An image acquisition unit for inputting the morphological distribution features into a preset fat particle recognition model and outputting a particle size distribution image corresponding to the fat particles;

[0147] A content detection unit for fusing the electrical signal sequence and the particle size distribution image, acquiring fusion feature information, and determining the chylomicron concentration corresponding to the lymph fluid sample according to the fusion feature information, thereby completing the detection of the chylomicron content.

[0148] In some embodiments, before inputting the morphological distribution features into the preset fat particle recognition model, it further includes: acquiring a plurality of historical sample microscopic images and corresponding historical particle size distribution images; the historical sample microscopic images are verified by Sudan III staining; and training a KNN classifier to be trained according to the plurality of historical sample microscopic images and corresponding historical particle size distribution images to obtain the fat particle recognition model.

[0149] Exemplarily, the inputting of the morphological distribution feature into a preset fat particle recognition model includes: obtaining fat cluster geometric information according to the morphological distribution feature; inputting the fat cluster geometric information into the fat particle recognition model, and the fat particle recognition model matches the historical particle size distribution image with the fat cluster geometric information and outputs the particle size distribution image.

[0150] In some embodiments, the fusion of the electrical signal sequence and the particle size distribution image includes: performing multimodal fusion on the electrical signal sequence and the particle size distribution image according to a heterogeneous feature fusion network based on an attention mechanism; adjusting the multimodal fusion ratio corresponding to the electrical signal sequence and the particle size distribution image in real time according to the turbidity parameter of the lymph fluid sample; wherein the range of the multimodal fusion ratio corresponding to the electrical signal sequence is 0.6 - 0.7, and the range of the multimodal fusion ratio corresponding to the particle size distribution image is 0.3 - 0.4.

[0151] Exemplarily, before performing multimodal fusion on the electrical signal sequence and the particle size distribution image according to a heterogeneous feature fusion network based on an attention mechanism, it further includes: constructing a two-stream feature cross encoder, which includes an electrochemical feature stream encoder and an image feature stream encoder; inputting the electrical signal sequence into the electrochemical feature stream encoder to output electrical signal time-frequency features; inputting the particle size distribution image into the image feature stream encoder to output image spatial domain features; obtaining the topological mapping relationship corresponding to the electrical signal time-frequency features and the image spatial domain features; and inputting the electrical signal time-frequency features, the image spatial domain features, and the topological mapping relationship into the heterogeneous feature fusion network.

[0152] In some embodiments, the determining of the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information includes: performing dimensionality reduction processing on the fusion feature information; extracting the dielectric relaxation feature of chylomicrons and the fractal dimension feature of lipoprotein aggregates from the dimensionally reduced fusion feature information; mapping the dielectric relaxation feature and the fractal dimension feature to a preset concentration calibration space; the concentration calibration space is constructed according to a semi-supervised adversarial generation network and includes chylomicron lipoprotein concentration gradient distribution data based on different pathological samples; searching for an optimal concentration trajectory in the concentration calibration space, and determining the chylomicron lipoprotein concentration value according to the Euclidean distance between the optimal concentration trajectory and the fusion feature information.

[0153] Exemplarily, the searching for an optimal concentration trajectory in the concentration calibration space includes: searching for an optimal concentration trajectory in the concentration calibration space according to an improved ant colony optimization algorithm; wherein the ant colony optimization algorithm includes a dynamic pheromone update mechanism based on Brownian motion.

[0154] In some embodiments, the frequency range corresponding to the scanning frequency of the anode-cathode double-cycle scanning is 0.1 - 1 MHz, which is used to suppress protein colloid interference.

[0155] In some embodiments, the microscopic image acquisition module includes a microscopic lens and a high-speed CMOS sensor; the microscopic lens includes a polarized light filter; and the optical resolution of the microscopic lens is greater than 0.8 μm.

[0156] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described chylomicron content detection device and each unit can refer to the corresponding processes in the chylomicron content detection method embodiments described above through potential difference and image learning, and will not be elaborated here.

[0157] The above chylomicron content detection method is implemented in the form of a computer program, and this computer program can run on the above device.

[0158] Please refer to Figure 3 , Figure 3 which is a schematic block diagram of the structure of the control module provided by the embodiments of the present application. The control module includes a processor, a memory, and a network interface connected through a device bus. Among them, the memory can include a storage medium and an internal memory.

[0159] The storage medium can store an operating device and a computer program. This computer program includes program instructions, and when the program instructions are executed, the processor can be made to execute any embodiment of the chylomicron content detection method through potential difference and image learning.

[0160] The processor is used to provide computing and control capabilities to support the operation of the entire control module.

[0161] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When this computer program is executed by the processor, the processor can be made to execute any method of the chylomicron content detection system.

[0162] This network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 3 the structure shown in

[0163] It should be understood that the processor can be a Central Processing Unit (CPU), and the processor can also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0164] Among them, in one embodiment, the processor is used to run a computer program stored in a memory to implement the following steps:

[0165] Obtain an electrical signal sequence corresponding to the lymph fluid sample measured by the electrochemical detection module;

[0166] Obtain the morphological distribution characteristics of fat particles in the lymph fluid sample captured by the microscopic image acquisition module;

[0167] Input the morphological distribution characteristics into a preset fat particle recognition model to output a particle size distribution image corresponding to the fat particles;

[0168] Fuse the electrical signal sequence and the particle size distribution image to obtain fusion feature information, and determine the chylomicron concentration corresponding to the lymph fluid sample according to the fusion feature information, thereby completing the detection of the chylomicron content.

[0169] In some embodiments, before inputting the morphological distribution characteristics into the preset fat particle recognition model, it further includes: obtaining a plurality of microscopic images of historical samples and corresponding historical particle size distribution images; the microscopic images of historical samples are verified by Sudan III staining; training a KNN classifier to be trained according to the plurality of microscopic images of historical samples and corresponding historical particle size distribution images to obtain the fat particle recognition model.

[0170] Exemplarily, inputting the morphological distribution characteristics into the preset fat particle recognition model includes: obtaining fat cluster geometric information according to the morphological distribution characteristics; inputting the fat cluster geometric information into the fat particle recognition model, and the fat particle recognition model matches the historical particle size distribution image with the fat cluster geometric information to output the particle size distribution image.

[0171] In some embodiments, the fusion based on the electrical signal sequence and the particle size distribution image includes: performing multi-modal fusion on the electrical signal sequence and the particle size distribution image according to a heterogeneous feature fusion network based on an attention mechanism; adjusting the multi-modal fusion ratio corresponding to the electrical signal sequence and the particle size distribution image in real time according to the turbidity parameter of the lymph fluid sample; wherein the range of the multi-modal fusion ratio corresponding to the electrical signal sequence is 0.6-0.7, and the range of the multi-modal fusion ratio corresponding to the particle size distribution image is 0.3-0.4.

[0172] Exemplarily, before performing multi-modal fusion on the electrical signal sequence and the particle size distribution image according to the heterogeneous feature fusion network based on an attention mechanism, it further includes: constructing a two-stream feature cross encoder, which includes an electrochemical feature stream encoder and an image feature stream encoder; inputting the electrical signal sequence into the electrochemical feature stream encoder to output the time-frequency features of the electrical signal; inputting the particle size distribution image into the image feature stream encoder to output the spatial domain features of the image; obtaining the topological mapping relationship corresponding to the time-frequency features of the electrical signal and the spatial domain features of the image; and inputting the time-frequency features of the electrical signal, the spatial domain features of the image, and the topological mapping relationship into the heterogeneous feature fusion network.

[0173] In some embodiments, determining the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information includes: performing dimensionality reduction processing on the fusion feature information; extracting the dielectric relaxation feature of chylomicrons and the fractal dimension feature of lipoprotein aggregates from the dimensionally reduced fusion feature information; mapping the dielectric relaxation feature and the fractal dimension feature to a preset concentration calibration space; the concentration calibration space is constructed according to a semi-supervised adversarial generation network and includes the chylomicron lipoprotein concentration gradient distribution data based on different pathological samples; searching for the optimal concentration trajectory in the concentration calibration space, and determining the chylomicron lipoprotein concentration value according to the Euclidean distance between the optimal concentration trajectory and the fusion feature information.

[0174] Exemplarily, searching for the optimal concentration trajectory in the concentration calibration space includes: searching for the optimal concentration trajectory in the concentration calibration space according to an improved ant colony optimization algorithm; wherein the ant colony optimization algorithm includes a dynamic pheromone update mechanism based on Brownian motion.

[0175] In some embodiments, the frequency range corresponding to the scanning frequency of the anode-cathode double-cycle scan is 0.1-1 MHz, which is used to suppress protein colloid interference.

[0176] In some embodiments, the microscopic image acquisition module includes a microscopic lens and a high-speed CMOS sensor; the microscopic lens includes a polarized light filter; and the optical resolution of the microscopic lens is greater than 0.8 μm.

[0177] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above-described processor can refer to the corresponding process in the method embodiments described in the above-mentioned embodiments, and will not be repeated here.

[0178] An embodiment of the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and the computer program includes program instructions. The processor executes the program instructions to implement the steps of the chylomicron lipoprotein content detection method provided by the above-mentioned embodiments of the present application through potential difference and image learning.

[0179] Among them, the computer-readable storage medium may be an internal storage unit of the control module described in the foregoing embodiment, such as the hard disk or memory of the control module. The computer-readable storage medium may also be an external storage device of the control module, such as a plug-in hard disk equipped on the control module, a smart media card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc.

[0180] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present application, and these modifications or substitutions should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A chylomicron lipoprotein content detection system through potential difference and image learning, characterized in that, Comprising: An electrochemical detection module, including a ring-shaped double electrode group and a potential controller; The ring-shaped double electrode group is used to insert into the lymph fluid sample, the potential controller is connected to the ring-shaped double electrode group, and the potential controller is used to perform anodic and cathodic double cycle scanning on the ring-shaped double electrode group, and generate an electrical signal sequence by measuring the change in conductivity corresponding to the lymph fluid sample; A microscopic image acquisition module, which is used to capture the morphological distribution characteristics of fat particles in the lymph fluid sample; A control module, which inputs the morphological distribution characteristics into a preset fat particle recognition model and outputs a particle size distribution image corresponding to the fat particles; fuses the electrical signal sequence and the particle size distribution image to obtain fusion feature information, and determines the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information, thereby completing the detection of the chylomicron lipoprotein content.

2. The system according to claim 1, characterized in that, Before inputting the morphological distribution characteristics into the preset fat particle recognition model, it further includes: Obtaining multiple microscopic images of historical samples and corresponding historical particle size distribution images; the microscopic images of historical samples are verified by Sudan III staining; Training the KNN classifier to be trained according to multiple microscopic images of historical samples and corresponding historical particle size distribution images to obtain the fat particle recognition model.

3. The system according to claim 2, wherein The inputting the morphological distribution characteristics into the preset fat particle recognition model includes: Obtaining fat cluster geometric information according to the morphological distribution characteristics; Inputting the fat cluster geometric information into the fat particle recognition model, and the fat particle recognition model matches the historical particle size distribution image with the fat cluster geometric information and outputs the particle size distribution image.

4. The system according to claim 1, characterized in that, The fusing according to the electrical signal sequence and the particle size distribution image includes: Performing multi-modal fusion on the electrical signal sequence and the particle size distribution image according to a heterogeneous feature fusion network based on an attention mechanism; Real-time adjusting the multi-modal fusion ratio corresponding to the electrical signal sequence and the particle size distribution image according to the turbidity parameter of the lymph fluid sample; Wherein, the range of the multi-modal fusion ratio corresponding to the electrical signal sequence is 0.6 - 0.7, and the range of the multi-modal fusion ratio corresponding to the particle size distribution image is 0.3 - 0.

4.

5. The system according to claim 4, wherein Before performing multi-modal fusion on the electrical signal sequence and the particle size distribution image according to the heterogeneous feature fusion network based on an attention mechanism, it further includes: Constructing a two-stream feature cross encoder, which includes an electrochemical feature stream encoder and an image feature stream encoder; Inputting the electrical signal sequence into the electrochemical feature stream encoder and outputting electrical signal time-frequency features; Inputting the particle size distribution image into the image feature stream encoder and outputting image spatial domain features; Obtaining the topological mapping relationship corresponding to the electrical signal time-frequency features and the image spatial domain features; for inputting the electrical signal time-frequency features, the image spatial domain features and the topological mapping relationship into the heterogeneous feature fusion network.

6. The system according to claim 1, wherein The determining the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fusion feature information includes: Perform dimensionality reduction processing on the fused feature information; Extract the dielectric relaxation characteristics of chylomicrons and the fractal dimension characteristics of lipoprotein aggregates from the fused feature information after dimensionality reduction; Map the dielectric relaxation characteristics and fractal dimension characteristics to a preset concentration calibration space; the concentration calibration space is constructed based on a semi-supervised adversarial generation network and includes chylomicron lipoprotein concentration gradient distribution data based on different pathological samples; Search for the optimal concentration trajectory in the concentration calibration space, and determine the chylomicron lipoprotein concentration value according to the Euclidean distance between the optimal concentration trajectory and the fused feature information.

7. The system according to claim 6, wherein The searching for the optimal concentration trajectory in the concentration calibration space includes: Search for the optimal concentration trajectory in the concentration calibration space according to the improved ant colony optimization algorithm; wherein the ant colony optimization algorithm includes a dynamic pheromone update mechanism based on Brownian motion.

8. The system according to claim 1, wherein The frequency range corresponding to the scanning frequency of the anode-cathode double-cycle scanning is 0.1 - 1 MHz, which is used to suppress protein colloid interference.

9. The system according to claim 1, wherein, The microscopic image acquisition module includes a microscopic lens and a high-speed CMOS sensor; the microscopic lens includes a polarized light filter; The optical resolution of the microscopic lens is greater than 0.8 μm.

10. A method for detecting the content of chylomicron lipoprotein by potential difference and image learning, characterized in that, For the control module applied to the system according to any one of claims 1 - 9, the method includes: Obtain the electrical signal sequence corresponding to the lymph fluid sample measured by the electrochemical detection module; Obtain the morphological distribution characteristics of fat particles in the lymph fluid sample captured by the microscopic image acquisition module; Input the morphological distribution characteristics into a preset fat particle recognition model, and output the particle size distribution image corresponding to the fat particles; Fuse the electrical signal sequence and the particle size distribution image to obtain fused feature information, and determine the chylomicron lipoprotein concentration corresponding to the lymph fluid sample according to the fused feature information, thereby completing the detection of the chylomicron lipoprotein content.

Images