A deep learning-based urine image feature extraction and recognition method

Abstract

The present application relates to a deep learning-based urine image feature extraction and recognition method, comprising: acquiring the original image of the urine sample to be tested by an image acquisition device and transmitting it to a central processing system; preprocessing and feature extraction of the original urine image, determining the corresponding analysis mode, the analysis mode including the conventional physicochemical characteristic mode, the formed component recognition mode and the multi-feature comprehensive diagnosis mode; according to different modes, the corresponding deep learning model is called to perform feature recognition and classification, and finally the urine component analysis result or comprehensive diagnosis prompt information is generated. The present application realizes the automatic and high-precision analysis of the multi-dimensional characteristics of urine color, transparency and formed components by combining multi-mode cooperation with deep learning model, effectively improving the accuracy and diagnosis efficiency of urine detection.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for extracting and identifying features from urine images based on deep learning. Background Technology

[0002] Urinalysis is an important routine examination in clinical diagnosis. By observing the morphology, quantity, and distribution of formed elements in urine (such as red blood cells, white blood cells, casts, crystals, and epithelial cells), it can provide crucial information for the diagnosis of urinary system diseases, kidney diseases, and other systemic diseases. Traditional urine formed element analysis mainly relies on manual microscopic examination. This method is labor-intensive, inefficient, and the results are greatly affected by the subjective experience of the laboratory personnel, making it prone to missed diagnoses and misdiagnoses, and difficult to standardize and automate.

[0003] In recent years, with the development of computer vision and artificial intelligence technologies, automated urine formed element analysis systems based on image processing have gradually become a research hotspot. However, existing technologies still face many challenges in urine image analysis. First, the quality of urine sample microscopic images is affected by various factors, such as uneven illumination, noise interference, and insufficient contrast, which makes subsequent image segmentation and feature extraction difficult. Second, urine formed elements are diverse in type and morphology, and often overlap and adhere, making accurate identification and segmentation difficult. Third, existing methods often focus on the identification and counting of individual formed elements, neglecting the spatial distribution, co-occurrence relationships, and statistical correlations between different formed elements. These correlations are crucial for comprehensively judging pathological conditions. For example, an abnormally high white blood cell count combined with a large bacterial aggregation may indicate a urinary tract infection, while the presence of red blood cell casts and granular casts strongly suggests glomerulonephritis. Existing technologies fail to fully utilize these higher-order correlations, resulting in insufficient accuracy and robustness in pathological condition judgment.

[0004] Furthermore, in actual clinical applications, various artifacts (such as bubbles, fibers, oil, etc.) often appear in urine images, and these artifacts may resemble the morphology of formed elements, which can easily lead to misjudgment. The current technology's ability to identify and eliminate artifacts needs to be improved.

[0005] Therefore, the applicant hereby proposes a method, system, terminal device, and readable storage medium for urine image feature extraction and pathological state identification. Summary of the Invention

[0006] This invention proposes a deep learning-based method for urine image feature extraction and identification, aiming to solve the technical problems of existing urine image analysis, such as insufficient image quality, inadequate accuracy in formed element recognition and segmentation, failure to fully utilize the correlation information between formed elements, and lack of artifact identification ability.

[0007] To achieve the above objectives, this invention proposes a method for urine image feature extraction and identification based on deep learning, comprising the following steps:

[0008] S1. Use image acquisition equipment to acquire the original image of the urine sample to be tested, and transmit the acquired image data to the central processing system;

[0009] S2. Preprocess and extract features from the raw urine image obtained in step S1, analyze and determine the analysis mode corresponding to the current urine image. The analysis mode includes at least: conventional physicochemical feature mode, formed element recognition mode, and multi-feature comprehensive diagnostic mode. The conventional physicochemical feature mode is used to make a preliminary judgment on a single physical property based on urine color and transparency. The formed element recognition mode is used to identify formed elements such as cells, casts, and crystals in urine through morphological analysis. The multi-feature comprehensive diagnostic mode is used to coordinate color, turbidity, formed element concentration and distribution characteristics, and perform comprehensive analysis and prompts in combination with preset medical rules. The preset medical rules include at least: basic rules for defining the normal range thresholds of each component, diagnostic rules for defining the correlation between feature combinations and pathological signs, and quality control rules for defining the review process under different testing scenarios.

[0010] S3. Based on the analysis mode determined in step S2, the corresponding deep learning model is invoked for feature recognition and classification to generate urine component analysis results or diagnostic prompts. S3 further includes the following steps: verifying the consistency of the analysis results through a logical reasoning algorithm; if there are conflicts in the conclusions of multi-feature inference, arbitration and review suggestions are made according to preset medical priority rules. The specific medical priority rules are as follows:

[0011] (a) Prioritize acute indications: Features indicating acute pathological changes have the highest priority and trigger immediate review;

[0012] (b) Morphological evidence takes precedence: conclusions with clear morphological evidence take precedence over inferences based solely on color / turbidity;

[0013] (c) Prioritize multiple features that are consistent: When multiple independent features support the same conclusion, the confidence level of that conclusion is increased;

[0014] (d) Prioritize historical comparison: When there is a significant deviation from the user's historical detection data, the current result should be highlighted.

[0015] Preferably, step S2 includes the following steps:

[0016] The original urine image obtained in step S1 is subjected to color correction, noise reduction, and image enhancement.

[0017] Calculate the color histogram, texture features, and edge density of the image;

[0018] Based on the extracted feature vectors, a pre-trained pattern classifier is used to determine the analysis pattern to which the current urine image belongs.

[0019] Preferably, the analysis mode determination process is as follows:

[0020] When the image texture complexity is higher than the preset first texture threshold and the edge density is higher than the preset first edge threshold, it is determined to be a formed element recognition mode; when the color feature deviates from the normal range to a degree exceeding the preset color deviation threshold and the texture complexity is lower than the preset second texture threshold, it is determined to be a conventional physical and chemical feature mode; when color, texture and edge features all trigger multi-dimensional rules and meet the preset comprehensive judgment conditions, it is determined to be a multi-feature comprehensive diagnosis mode.

[0021] Wherein, the first texture threshold is the maximum value of image texture complexity, used to determine whether to enter the formed element recognition mode; the second texture threshold is the minimum value of device texture intensity, used in combination with the first texture threshold to determine whether to enter the conventional physical and chemical feature mode.

[0022] Preferably, step S2 further includes the following step:

[0023] The original urine image was segmented to distinguish the background, liquid areas and formed element clusters, and features were extracted and normalized for each area.

[0024] Preferably, step S3 includes the following steps:

[0025] Based on the analysis mode determined in step S2, select the corresponding deep learning recognition model;

[0026] If it is a formed element recognition mode, the component recognition deep learning model is called, the segmented formed element region image is input into the model, and the probability distribution of each component category and quantity is output.

[0027] If it is a multi-feature comprehensive diagnosis mode, the comprehensive diagnosis deep learning model is called to fuse multi-dimensional features such as color, texture, and morphology, and output possible pathological state prompts after weighting through the attention mechanism.

[0028] If it is a conventional physicochemical feature mode, the corresponding physicochemical property level will be matched from the preset rule library based on the color and transparency features.

[0029] Preferably, the component recognition deep learning model adopts a convolutional neural network structure, with the input being a standardized formed component region image and the output being the category confidence of each component; the comprehensive diagnosis deep learning model adopts a multi-feature fusion network combined with a temporal attention mechanism, with the input being a normalized multi-dimensional feature sequence and the output being a comprehensive diagnostic suggestion probability distribution.

[0030] Preferably, when the comprehensive diagnostic deep learning model outputs diagnostic prompts, the following constraints must be met:

[0031] Clinical indication relevance constraint: The pathological condition suggested must be statistically significantly correlated with the image features;

[0032] Interpretability constraint: The model must output key image feature regions that support the prompt;

[0033] Safety warning constraints: Images with suspected acute pathological signs must trigger a review prompt regardless of confidence level;

[0034] Multiple validation constraints: When a single feature is unclear, multiple features must be used for collaborative judgment to avoid false positives / false negatives.

[0035] The beneficial effects of the technical solution of this invention are as follows:

[0036] By introducing deep learning models, we can extract deep features from complex urine images more effectively, overcome the limitations of traditional methods in feature extraction, and significantly improve the accuracy of urine component identification and overall analysis efficiency.

[0037] It can intelligently determine the analysis mode (routine physical and chemical analysis, formed element identification, and multi-feature comprehensive diagnosis) based on the characteristics of urine images, avoiding the limitations of a single mode and making the analysis process more targeted and intelligent.

[0038] The multi-feature integrated diagnostic mode can coordinate color, turbidity, formed element concentration and distribution characteristics, and combine them with preset medical rules to conduct comprehensive analysis and suggestions, providing more comprehensive and valuable diagnostic advice for clinical practice, and improving the intelligence level and clinical applicability of diagnosis.

[0039] By introducing constraints related to clinical indications, interpretability of results, safety warnings, and multiple validations, the scientific rigor, reliability, and clinical safety of diagnostic suggestions are ensured. Furthermore, the introduction of logical reasoning algorithms further enhances the consistency and credibility of the analysis results.

[0040] Automated and intelligent image feature extraction and deep learning recognition greatly reduce reliance on human experience, reduce errors caused by subjective judgment, and improve the standardization and consistency of detection results. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of a urine image pH value acquisition card in one embodiment of this application;

[0042] Figure 2 This is a schematic diagram of urine crystals observed under a polarizing microscope in one embodiment of this application;

[0043] Figure 3 This is a schematic diagram of red blood cells exhibiting a datura-like or shrunken shape under a bright-field microscope in one embodiment of this application, due to the influence of non-physiological pH values ​​of urine.

[0044] Figure 4 This is a schematic diagram of fungal hyphae observed under a bright-field microscope in one embodiment of this application.

[0045] Figure 5 This is a schematic diagram of phase contrast observation of urine sediment in one embodiment of this application.

[0046] Figure 6 This is a schematic diagram of the terminal device structure for a urine image feature extraction and pathological state identification method in one embodiment of this application.

[0047] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0048] The solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0050] It should also be noted that when a component is described as "fixed to" or "set on" another component, it can be directly on the other component or there may be an intervening component present. When a component is described as "connected to" another component, it can be directly connected to the other component or there may be an intervening component present.

[0051] Furthermore, descriptions using terms such as "first" and "second" in this invention are for descriptive purposes only (e.g., to distinguish identical or similar elements) and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" and "second" may explicitly or implicitly include at least one of those features. Additionally, technical solutions from different embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If a combination of technical solutions is contradictory or impossible to implement, such a combination should be considered nonexistent and not within the scope of protection claimed by this invention.

[0052] See Figure 1 This invention proposes a method for extracting and identifying urine image features based on deep learning, comprising the following steps:

[0053] S1. Use image acquisition equipment to acquire the original image of the urine sample to be tested, and transmit the acquired image data to the central processing system.

[0054] In this step, the image acquisition device can be a high-resolution digital microscope, a urine analyzer equipped with a specialized image sensor, or any device capable of acquiring clear images of urine samples. The acquired raw image data, such as RGB format image files, is transmitted to a central processing system via wired or wireless means. The central processing system can be a high-performance computer, server, or embedded processing unit, responsible for subsequent image processing, feature extraction, pattern determination, and deep learning model inference.

[0055] S2. Preprocess and extract features from the raw urine image obtained in step S1, analyze and determine the analysis mode corresponding to the current urine image. The analysis mode includes at least: conventional physicochemical feature mode, formed element recognition mode and multi-feature comprehensive diagnosis mode.

[0056] Specifically, refer to Figure 2-5 Step S2 includes the following sub-steps:

[0057] First, the raw urine image acquired in step S1 undergoes color correction, denoising, and image enhancement. Color correction aims to eliminate color deviations caused by different acquisition devices or lighting conditions, ensuring the standardization of image colors. Denoising (e.g., using Gaussian filtering, median filtering, etc.) removes random noise from the image, improving image quality. Image enhancement (e.g., contrast stretching, histogram equalization, etc.) highlights details and features in the image, providing better input for subsequent feature extraction.

[0058] Secondly, the color histogram, texture features, and edge density of the image are calculated. The color histogram quantifies the color distribution of the image, reflecting the overall color information of the urine. Texture features (such as gray-level co-occurrence matrix, local binary pattern (LBP), etc.) are used to describe the local structure and details of the image, reflecting the turbidity and graininess of the urine. Edge density reflects the richness of object boundaries in the image, which is crucial for identifying formed elements. These features together constitute the image's feature vector.

[0059] Finally, based on the extracted feature vectors, a pre-trained pattern classifier determines the analysis mode to which the current urine image belongs. This pattern classifier can be a support vector machine (SVM), a random forest, or a small neural network, which is trained on a large amount of labeled data to learn how to classify urine images into different analysis modes based on features such as color, texture, and edge density.

[0060] Specifically, the image texture complexity refers to a feature value obtained by performing texture analysis on a preprocessed urine image, quantifying the irregularity and complexity of its local structure. It reflects the richness of the visual texture presented by the distribution of particulate matter, formed elements, or fluid state in the urine sample.

[0061] Specifically, the gray-level co-occurrence matrix (GLCM) or other texture descriptors are calculated based on regions of the urine image (such as the global or specific regions of interest). Let p texture features (such as contrast, correlation, energy, homogeneity, entropy, etc.) be extracted from the GLCM, denoted as: And the value of each parameter at time point t is... ; and on The parameters in the data are normalized, and the parameters are set as follows: The historical minimum value is The historical maximum value is The normalized value at time point t can be obtained as follows: ;

[0062] Furthermore, for adjacent time points t and t+1, the parameters The texture amount is: Weights are set based on the impact on the environmental state texture. ,satisfy The weighted texture parameters are then: ;

[0063] Finally, the expression for image texture complexity is derived as follows:

[0064] ;

[0065] Furthermore, referring to Figure 3 The specific process for determining the analysis mode is as follows:

[0066] When the image texture complexity exceeds a preset first texture threshold, and the edge density exceeds a preset first edge threshold, it is determined to be in formed element recognition mode. This indicates that the image may contain a large number of formed elements such as cells, tubules, and crystals, requiring detailed morphological analysis. The first texture threshold is the maximum value of the current image texture complexity, used to determine whether to enter the formed element recognition mode, ensuring that this mode is triggered only when the texture information is sufficiently rich. The first edge threshold is used to ensure that there is enough boundary information in the image to support the recognition of formed elements.

[0067] When the color feature deviates from the normal range by more than a preset color deviation threshold, and the texture complexity is lower than a preset second texture threshold, it is determined to be in a conventional physicochemical feature mode. This usually corresponds to abnormal urine color or transparency, but with unclear formed elements. The second texture threshold is the minimum value of the device's texture intensity, used in conjunction with the first texture threshold to determine whether to enter the conventional physicochemical feature mode; that is, when texture information is not rich, a conventional physicochemical judgment is more likely to be made.

[0068] When color, texture, and edge features all trigger multi-dimensional rules and meet preset comprehensive judgment conditions, the image is judged as a multi-feature comprehensive diagnostic mode. This indicates that multiple abnormal features exist in the image, requiring a comprehensive diagnosis by combining various information. The comprehensive judgment condition can be a logical combination rule, such as "color abnormality AND texture abnormality AND edge density moderate to high", or it can be determined by a multi-classifier.

[0069] In a preferred embodiment, S2 further includes the following steps: performing region segmentation on the original urine image to distinguish between background, liquid regions, and formed element aggregation regions, and performing feature extraction and normalization processing on each region. Region segmentation can employ techniques such as thresholding, edge detection, or semantic segmentation to divide the image into different semantic regions. Performing feature extraction and normalization processing on different regions helps subsequent deep learning models focus more on the analysis of key regions, improving the accuracy and robustness of recognition. For example, in formed element aggregation regions, more refined morphological features can be extracted; in liquid regions, more attention is paid to color and turbidity features.

[0070] S3. Based on the analysis mode determined in step S2, call the corresponding deep learning model to perform feature recognition and classification, and generate urine component analysis results or diagnostic prompts.

[0071] S3 includes the following sub-steps:

[0072] Based on the analysis mode determined in step S2, select the corresponding deep learning recognition model.

[0073] If the form element recognition mode is selected, a deep learning model for component recognition is invoked. This model takes the segmented form element region image as input and outputs the probability distribution of each component category (such as red blood cells, white blood cells, epithelial cells, various casts, crystals, etc.) and quantity. For example, this model could be an image classification or object detection model based on a convolutional neural network (CNN), capable of accurately identifying and counting form elements in urine.

[0074] If a multi-feature comprehensive diagnostic mode is used, a comprehensive diagnostic deep learning model is invoked. This model fuses multi-dimensional features such as color, texture, and morphology, and outputs possible pathological state suggestions after weighting through an attention mechanism. The attention mechanism enables the model to automatically focus on the features that contribute most to the diagnostic results when processing multi-dimensional features, thereby improving the accuracy and interpretability of the diagnosis. For example, when the model determines glomerulonephritis, it may focus on features such as red blood cell morphology and protein casts.

[0075] If the mode is a standard physicochemical characteristic mode, the corresponding physicochemical property level is matched from a preset rule base based on color and transparency characteristics. For example, based on the depth of urine color and the level of turbidity, it is judged as "pale yellow and clear" or "dark yellow and cloudy", and corresponding preliminary prompts are given.

[0076] In a preferred embodiment, the component recognition deep learning model employs a convolutional neural network (CNN) structure. The input is a standardized image of formed component regions, and the output is the category confidence score of each component. CNNs are widely used due to their superior performance in image recognition, capable of automatically learning spatial hierarchical features in images. Standardization ensures the consistency of the input data. The comprehensive diagnostic deep learning model employs a multi-feature fusion network combined with a temporal attention mechanism. The input is a normalized multi-dimensional feature sequence, and the output is a probability distribution of comprehensive diagnostic suggestions. The multi-feature fusion network effectively integrates feature information from different modalities (color, texture, shape, etc.). The temporal attention mechanism allows the model to dynamically allocate attention weights when processing feature sequences, thereby better capturing the correlation and importance between features and improving the accuracy of comprehensive diagnosis.

[0077] Furthermore, referring to Figure 5 When the comprehensive diagnostic deep learning model outputs diagnostic prompts, the following constraints must be met:

[0078] Clinical indication relevance constraint: The suggested pathological state must be statistically significantly correlated with the image features. This means that during model training, it needs to ensure that there is a strong statistical correlation between its output diagnostic results and actual clinical diagnoses, avoiding the output of clinically meaningless indications.

[0079] Interpretability constraint: The model must output key image feature regions that support the indication. For example, when indicating glomerulonephritis, the model should highlight abnormally shaped red blood cells or protein casts in the image, enhancing the transparency and credibility of the diagnostic result. This can be achieved through interpretable AI techniques such as Gradient Weighted Class Activation Mapping (Grad-CAM).

[0080] Safety warning constraints: Images with suspected acute pathological indications should trigger a review prompt regardless of confidence level. For example, if the model detects a small number of abnormally shaped red blood cells or white blood cells, even with low confidence, a manual review should be prompted to avoid missing acute pathological conditions.

[0081] Multiple validation constraints: When a single feature is unclear, multiple features must be used for collaborative judgment to avoid false positives / false negatives. For example, if only the color is abnormal but other features are normal, the model should not easily give a serious diagnosis, but should combine multiple features such as texture and formed elements for a comprehensive judgment.

[0082] In a preferred embodiment, the preset medical rules include at least: basic rules for defining the normal range thresholds of each component, diagnostic rules for defining the correlation between feature combinations and pathological indications, and quality control rules for defining the review process under different testing scenarios. Basic rules, for example, define the normal count ranges for red blood cells, white blood cells, casts, etc. Diagnostic rules, for example, define that "a large number of red blood cells + protein casts" may indicate glomerular disease. Quality control rules, for example, define that "manual review is required when the test results significantly differ from historical data."

[0083] In a preferred embodiment, step S3 further includes the following steps: verifying the consistency of the analysis results using a logical reasoning algorithm; if there are conflicts in the multi-feature inference conclusions, arbitration and review recommendations are made according to preset medical priority rules. The logical reasoning algorithm can be based on an expert knowledge base or ontology to perform secondary verification on the output of the deep learning model, ensuring the logical rationality of the results.

[0084] Furthermore, the medical priority rule is specifically as follows:

[0085] (a) Prioritize acute indications: Features indicating acute pathological changes have the highest priority and trigger immediate review. For example, indications related to acute kidney injury and acute infection.

[0086] (b) Morphological evidence takes precedence: Conclusions with clear morphological evidence take precedence over inferences based solely on color / turbidity. For example, clearly identified red blood cell casts are more diagnostically valuable than simply red urine color.

[0087] (c) Prioritize multiple consistent features: When multiple independent features support the same conclusion, the confidence level of that conclusion is increased. For example, if color, turbidity, and formed elements all point to the same pathological condition, the diagnosis has a higher priority.

[0088] (d) Prioritize historical comparisons: When there is a significant deviation from the user's historical test data, the current result should be highlighted. For example, if a patient's urine has been normal for a long time and suddenly develops a serious abnormality, it should be taken very seriously, even if the model's confidence level is not high.

[0089] This application provides a urine image pathological state identification system for implementing any of the above methods, characterized in that it includes:

[0090] The image acquisition module is used to acquire raw microscopic images of urine samples;

[0091] The image preprocessing module is used to preprocess the original image to obtain the target image;

[0092] The component stratification module is used to construct a stratification system for urine formed components based on clinical importance;

[0093] The component segmentation module is used to perform formed component detection and segmentation on the target image, and to identify and segment the components of each layer.

[0094] The feature extraction module is used to extract multi-dimensional feature sets from each segmented urine constituent element;

[0095] The graph construction module is used to construct component association graphs based on the spatial distribution and statistical relationships of components.

[0096] The state discrimination module has a built-in pre-trained pathological state classification model, which is used to receive the multi-dimensional feature set and component association map and output the pathological state discrimination results and differential warning information.

[0097] Furthermore, this application embodiment also provides a terminal device, the internal structure of which can be as follows: Figure 3As shown, the terminal device includes a processor, memory, communication interface, and database connected via a system bus. The processor provides computing and control capabilities. The terminal device's memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The terminal device's database stores data called by the computer programs. The terminal device's communication interface is used for data communication with external terminals. The terminal device's input device receives signals from external devices. When the computer program is executed by the processor, it implements a urine image feature extraction and pathological state identification method as described in the above embodiment.

[0098] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the solution of this application, and does not constitute a limitation on the terminal device to which the solution of this application is applied.

[0099] Furthermore, this application also proposes a readable storage medium comprising a computer program, which, when executed by a processor, implements the steps of the urine image feature extraction and pathological state identification method as described in the above embodiments. It is understood that the readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0100] In summary, the urine image feature extraction and pathological state identification method, device, terminal equipment and readable storage medium provided in the embodiments of this application combine systematic image preprocessing, target recognition under the guidance of a hierarchical system, multi-dimensional feature extraction and component correlation map analysis to construct an efficient, accurate and reliable automated urine analysis scheme.

[0101] Those skilled in the art will understand that implementing all or part of the processes in the above-described methods for urine image feature extraction and pathological state identification can be accomplished by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above-described methods for urine image feature extraction and pathological state identification. Any references to memory, storage, databases, or other media used in this application and in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0102] The above description is only a part or preferred embodiment of the present invention. Neither the text nor the drawings should limit the scope of protection of the present invention. All equivalent structural transformations made using the content of the present invention specification and drawings under the overall concept of the present invention, or direct / indirect applications in other related technical fields, are included within the scope of protection of the present invention.

Claims

1. A method for feature extraction and identification of urine images based on deep learning, characterized in that, Includes the following steps: S1. Use image acquisition equipment to acquire the original image of the urine sample to be tested, and transmit the acquired image data to the central processing system; S2. Preprocess and extract features from the raw urine image obtained in step S1, analyze and determine the analysis mode corresponding to the current urine image. The analysis mode includes at least: conventional physicochemical feature mode, formed element recognition mode, and multi-feature comprehensive diagnostic mode. The conventional physicochemical feature mode is used to make a preliminary judgment on a single physical property based on urine color and transparency. The formed element recognition mode is used to identify formed elements such as cells, casts, and crystals in urine through morphological analysis. The multi-feature comprehensive diagnostic mode is used to coordinate color, turbidity, formed element concentration and distribution characteristics, and perform comprehensive analysis and prompts in combination with preset medical rules. The preset medical rules include at least: basic rules for defining the normal range thresholds of each component, diagnostic rules for defining the correlation between feature combinations and pathological signs, and quality control rules for defining the review process under different testing scenarios. S3. Based on the analysis mode determined in step S2, the corresponding deep learning model is invoked for feature recognition and classification to generate urine component analysis results or diagnostic prompts. S3 further includes the following steps: verifying the consistency of the analysis results through a logical reasoning algorithm; if there are conflicts in the conclusions of multi-feature inference, arbitration and review suggestions are made according to preset medical priority rules. The specific medical priority rules are as follows: (a) Prioritize acute indications: Features indicating acute pathological changes have the highest priority and trigger immediate review; (b) Morphological evidence takes precedence: conclusions with clear morphological evidence take precedence over inferences based solely on color / turbidity; (c) Prioritize multiple features that are consistent: When multiple independent features support the same conclusion, the confidence level of that conclusion is increased; (d) Prioritize historical comparison: When there is a significant deviation from the user's historical detection data, the current result should be highlighted.

2. The method for extracting and identifying urine image features based on deep learning according to claim 1, characterized in that, Step S2 includes the following steps: color correction, denoising, and image enhancement of the original urine image obtained in step S1; calculation of the color histogram, texture features, and edge density of the image; and determination of the analysis mode to which the current urine image belongs based on the extracted feature vector using a pre-trained pattern classifier.

3. The method for extracting and identifying urine image features based on deep learning according to claim 2, characterized in that, The analysis mode determination process is as follows: when the image texture complexity is higher than a preset first texture threshold and the edge density is higher than a preset first edge threshold, it is determined to be a formed element recognition mode; when the color feature deviates from the normal range by more than a preset color deviation threshold and the texture complexity is lower than a preset second texture threshold, it is determined to be a conventional physical and chemical feature mode; when color, texture, and edge features all trigger multi-dimensional rules and meet preset comprehensive determination conditions, it is determined to be a multi-feature comprehensive diagnosis mode; wherein, the first texture threshold is the maximum value of image texture complexity, used to determine whether to enter the formed element recognition mode; the second texture threshold is the minimum value of device texture intensity, used in conjunction with the first texture threshold to determine whether to enter the conventional physical and chemical feature mode.

4. The method for extracting and identifying urine image features based on deep learning according to claim 3, characterized in that, The S2 further includes the following steps: performing region segmentation on the original urine image, distinguishing between background, liquid regions and formed element aggregation regions, and performing feature extraction and normalization processing on each region.

5. The method for extracting and identifying urine image features based on deep learning according to claim 3, characterized in that, S3 includes the following steps: Selecting a corresponding deep learning recognition model based on the analysis mode determined in step S2; wherein, if it is a formed element recognition mode, the component recognition deep learning model is called, the segmented formed element region image is input into the model, and the probability distribution of each component category and quantity is output; if it is a multi-feature comprehensive diagnosis mode, the comprehensive diagnosis deep learning model is called, the multi-dimensional features such as color, texture, and morphology are fused, and the possible pathological state prompts are output after weighting through an attention mechanism; if it is a conventional physicochemical feature mode, the corresponding physicochemical property level is matched from the preset rule base according to the color and transparency features.

6. The method for extracting and identifying urine image features based on deep learning according to claim 5, characterized in that, The component recognition deep learning model adopts a convolutional neural network structure. The input is a standardized formed component region image, and the output is the category confidence of each component. The comprehensive diagnosis deep learning model adopts a multi-feature fusion network combined with a temporal attention mechanism. The input is a normalized multi-dimensional feature sequence, and the output is a comprehensive diagnostic suggestion probability distribution.

7. The method for extracting and identifying urine image features based on deep learning according to claim 5, characterized in that, When the comprehensive diagnostic deep learning model outputs diagnostic prompts, the following constraints must be met: Clinical indication relevance constraint: The pathological condition suggested must be statistically significantly correlated with the image features; Interpretability constraint: The model must output key image feature regions that support the prompt; Safety warning constraints: Images with suspected acute pathological signs must trigger a review prompt regardless of confidence level; Multiple validation constraints: When a single feature is unclear, multiple features must be used for collaborative judgment to avoid false positives / false negatives.

Images