An analytical method based on urine formed elements

Abstract

A method for analyzing formed elements in urine includes the following steps: acquiring multiple images of formed elements from a urine sample; segmenting the acquired images, where each segmented image contains only one type of formed element; labeling all the small images to form a dataset; training a deep convolutional neural network model using the dataset to classify the formed elements; classifying images of renal tubular epithelial cells in the sample using the trained deep convolutional neural network model; performing a chi-square test to obtain a significant difference P-value based on the number of positive renal tubular cell samples; performing a rank-sum test to obtain a significant difference P-value based on the detection results of renal tubular epithelial cells; if all obtained P-values ​​are less than 0.001, the classification index of the deep convolutional neural network model is deemed effective, and the classification result obtained by the deep convolutional neural network model can be used as a reference index for AKI diagnosis.

Description

Technical Field

[0001] This invention relates to the field of urine detection and analysis technology, and in particular to an analytical method based on the formed elements of urine. Background Technology

[0002] Routine urinalysis is a non-invasive and widely used testing method, and its results play an important role in the diagnosis of related diseases. By analyzing indicators such as urine color, composition, and physicochemical properties, it helps assess kidney function, urinary system health, and metabolic diseases. The examination mainly includes appearance, chemical analysis such as urine protein, urine glucose, and pH, and microscopic examination of cells and casts. It is suitable for physical examinations, disease diagnosis, and monitoring of treatment efficacy. However, microscopic examination is slow, and the results are also dependent on the doctor's skill level.

[0003] Existing technologies have developed urine testing devices that can detect various formed elements in urine, such as casts, red blood cells, white blood cells, epithelial cells, crystals, etc. However, existing urine testing devices can only detect the aforementioned formed elements, but there is no rapid analytical method to analyze the results represented by these formed elements. Summary of the Invention

[0004] In view of this, the present invention provides an analytical method based on the formed elements of urine to solve the above-mentioned technical problems.

[0005] An analytical method based on formed elements in urine, comprising the following steps: STEP101: Provides a urine formed elements analyzer and acquires multiple formed element images of a urine sample using the urine formed elements analyzer; STEP102: The obtained multiple images of formed elements are segmented using an image segmentation method. Each segmented image contains only one type of formed element, namely red blood cells, white blood cells, casts, crystals, renal tubular epithelium, and squamous epithelium. STEP103: Annotate images with formed elements according to their types to form a dataset, and divide the dataset into training, validation and test sets according to a certain ratio; STEP104: Provide a deep convolutional neural network model, train the deep convolutional neural network model using the dataset, and fix the parameters of the trained network model to classify formed elements; STEP105: Use a trained deep convolutional neural network model to classify the images of renal tubular epithelial cells in the samples, and group them according to the criteria of whether there is a clinical diagnosis, whether it meets KDIGO, and whether USS is greater than or equal to 1. The samples are divided into AKI group and non-AKI group. The AKI group is clinically diagnosed as AKI, meets KDIGO, or has USS greater than or equal to 1. The non-AKI group is clinically diagnosed as non-AKI, does not meet KDIGO, or has USS of 0. STEP106: Use the rank-sum test to evaluate the numerical results of renal tubular epithelial cell detection. Calculate the significant differences in renal tubular epithelial cell detection values ​​for each group using a deep convolutional neural network model. If the calculated difference is less than 0.001, the model results meet the requirements; if the difference is greater than or equal to 0.001, the model does not meet the requirements. STEP106: For the above classification, for each of the three groups, perform a chi-square test according to the number of positive renal tubular cell samples detected to obtain the significant difference P-value between the deep convolutional neural network model and the non-AKI group. STEP107: For the above classification, for each of the three groups, according to the detection results of renal tubular epithelial cells, the rank-sum test is performed to obtain the significant difference P value between the deep convolutional neural network model and the non-AKI group. STEP108: If the P-values ​​obtained in STEP106 are all less than 0.001, and the P-values ​​obtained in STEP107 are also all less than 0.001, then the classification index of the deep convolutional neural network model is determined to be valid.

[0006] Furthermore, the urine formed elements analyzer uses planar flow cytometry to acquire multiple images of the urine.

[0007] Furthermore, the image segmentation method is the Otsu method.

[0008] Furthermore, the proportions of the training set, validation set, and test set are 60%, 20%, and 20%, respectively.

[0009] Furthermore, according to the KDIGO grading criteria, the images containing renal tubular epithelial cells were divided into only the AKI group and the non-AKI group.

[0010] Furthermore, the clinical diagnosis involves using a microscope to determine whether the small images of the formed elements belong to the AKI group or the non-AKI group.

[0011] Furthermore, when the number of renal tubular epithelial cells per high-power field is 0, 1-5, and the presence time is greater than or equal to 6, the USS score is 0, 1, and 2 points respectively.

[0012] Furthermore, when the number of particle tubes per low magnification field of view is 0, 1 to 5, and the existence time is greater than or equal to 6, the USS score is 0, 1, and 2 points respectively.

[0013] Compared with existing technologies, the urine formed element analysis method provided by this invention first captures images using existing instruments, then segments them into small images containing only one type of formed element using existing image segmentation algorithms, and then constructs and trains a deep convolutional neural network model. The trained deep convolutional neural network model is then used to classify the small images containing formed elements. Urine samples from patients in departments with a high incidence of AKI are collected and grouped into AKI and non-AKI groups based on criteria such as clinical diagnosis, KDIGO conformation, and USS greater than or equal to 1. After classification, the chi-square test is used to calculate the p-value of the significant difference between the deep convolutional neural network model and the non-AKI group, and the rank-sum test is used to calculate the p-value of the significant difference between the deep convolutional neural network model and the non-AKI group. If the p-values ​​obtained by both statistical methods are less than 0.001, the classification result by the deep convolutional neural network model can be used as a reference indicator for AKI diagnosis, thereby reducing the reliance on microscopy and improving efficiency. Attached Figure Description

[0014] Figure 1 The flowchart illustrates an analytical method for urine formed elements provided by this invention.

[0015] Figure 2 Images of urine collected using a urine formed elements analyzer. Detailed Implementation

[0016] The following provides a more detailed description of specific embodiments of the present invention. It should be understood that the description of the embodiments of the present invention herein is not intended to limit the scope of protection of the present invention.

[0017] like Figure 1 The diagram shown is a structural schematic of an analytical method for urine formed elements provided by the present invention. The analytical method for urine formed elements includes the following steps: STEP101: Provides a urine formed elements analyzer and acquires multiple formed element images of a urine sample using the urine formed elements analyzer; STEP102: The obtained multiple images of formed elements are segmented using an image segmentation method. The segmented images contain images of only one type of formed element, which is one of the following: red blood cells, white blood cells, casts, crystals, renal tubular epithelium, squamous epithelium, etc. STEP103: Annotate images with formed elements according to their types to form a dataset, and divide the dataset into training, validation and test sets according to a certain ratio; STEP104: Provide a deep convolutional neural network model, train the deep convolutional neural network model using the dataset, the parameters of the trained network model are fixed and it classifies formed elements; STEP105: Use a trained deep convolutional neural network model to classify the images of renal tubular epithelial cells in the samples, collect urine samples, and group them according to the criteria of whether there is a clinical diagnosis, whether it meets KDIGO, and whether USS is greater than or equal to 1. The AKI group is clinically diagnosed as AKI, meets KDIGO, or has a USS greater than or equal to 1. The non-AKI group is clinically diagnosed as non-AKI, does not meet KDIGO, or has a USS of 0. STEP106: For the above classification, for each of the three groups, perform a chi-square test according to the number of positive renal tubular cell samples detected to obtain the significant difference P-value between the deep convolutional neural network model and the non-AKI group. STEP107: For the above classification, for each of the three groups, according to the detection results of renal tubular epithelial cells, the rank-sum test is performed to obtain the significant difference P value between the deep convolutional neural network model and the non-AKI group. STEP108: If the P-values ​​obtained in STEP106 are all less than 0.001, and the P-values ​​obtained in STEP107 are also all less than 0.001, then the classification index of the deep convolutional neural network model is determined to be valid.

[0018] In STEP101, the urine formed element analyzer itself is existing technology, as disclosed in patent number CN202322534470.3, entitled "A Urine Analysis Mechanism and Liquid Path System." The urine formed element analyzer employs planar flow cytometry, capable of acquiring numerous images (e.g., 2000) for each sample, with an image size of 1440x1080 pixels. Each of these images typically contains many formed elements, such as... Figure 2 As shown. It is well known that these formed elements can be red blood cells, white blood cells, casts, crystals, renal tubular epithelium, squamous epithelium, and other epithelial cells, etc.

[0019] In STEP 102, the image segmentation method itself is an existing technology, such as the Otsu method, as disclosed in patent application CN202311153927.4, entitled "A Medical Image Segmentation Method and a Training Method for a Medical Image Segmentation Model." This method, for a large acquired image, first determines a threshold, uses the threshold to binarize the image, and then uses connected component analysis to segment it into many smaller images. In this embodiment, a large image is segmented into multiple smaller images of formed elements. Each smaller image of formed elements has one and only one formed element. For example, one smaller image of formed elements may only show an image of red blood cells, while another smaller image of formed elements may only show an image of renal tubular epithelium.

[0020] In STEP 103, the annotation of the formed element small images can be done manually or with machine assistance. Each formed element small image is labeled with its type, indicating whether the formed elements are renal tubular epithelium or other cells. These labeled images form a dataset. This dataset is then divided into training, validation, and test sets according to a certain ratio. The training, validation, and test sets are the foundational data for the deep convolutional neural network model during training and use. The training set is used to fit the parameters of the deep convolutional neural network model, such as weights and biases. The validation set is used to adjust hyperparameters such as learning rate and number of network layers; it does not directly participate in parameter updates but helps determine whether the model is overfitting or underfitting. The test set is used for the final performance evaluation of the trained model, reflecting its generalization ability in real-world scenarios. In this embodiment, the dataset is divided into training, validation, and test sets in proportions of 60%, 20%, and 20%, respectively.

[0021] In STEP104, the deep convolutional neural network (DNN) model is a mature existing technology widely used in image recognition and vision tasks, and is one of the core models in deep learning. This DNN model automatically extracts local features from the input data through a hierarchical structure of convolutional layers, pooling layers, and fully connected layers, progressively compressing information, reducing redundancy, and improving generalization ability. Since the deep convolutional neural network model is existing technology, its working principle will not be described in detail here. In this embodiment, the deep convolutional neural network model is used to identify cell images of small formed components in the obtained dataset, thereby providing identification and detection results for medical personnel's reference.

[0022] By training the deep convolutional neural network model, it gains the ability to identify formed elements in the detection dataset, and then classifies these formed elements in small images. To ensure this classification meets the diagnostic requirements of medical personnel, the deep convolutional neural network is trained by dividing the dataset into two categories: renal tubular epithelial cells are classified as one category, and all other formed elements are classified as another, forming a binary classification problem. The purpose of this classification is to distinguish renal tubular epithelial cells from the formed elements, while other types of formed elements can be disregarded, thus training a binary classification convolutional neural network. The characteristics of renal tubular epithelial cells can be used by medical personnel to diagnose acute kidney injury. It is conceivable that if other types of cells, such as red blood cells or white blood cells, need to be detected, red blood cells or white blood cells can be classified as one category, and other types of formed elements can be classified as another.

[0023] In STEP 105, grouping is based on three criteria: clinical diagnosis, KDIGO, and USS. Clinical diagnosis is the physician's diagnosis of the patient, categorized into AKI and non-AKI groups. AKI is the abbreviation for Acute Kidney Injury. KDIGO (Kidney Disease: Improving Global Outcomes) is a global non-profit organization established in 2003. This organization grades the severity of acute kidney injury (AKI) using the KDIGO classification, a crucial tool for assessing the severity of acute deterioration in kidney function. In this embodiment, patients are divided into AKI and non-AKI groups according to the KDIGO classification. The USS (urine sediment score) is a standard designed based on granular casts and renal tubular epithelial cell counts for urine sediment scoring, published by Perazella et al. in 2010. The microscopic examination method for this USS standard conforms to the requirements of the 4th edition of the operating procedure. The USS score is the sum of the scores for renal tubular epithelial cells and granular casts. When the number of renal tubular epithelial cells per high-power field is 0, 1-5, or greater than or equal to 6, the score is 0, 1, or 2 points respectively. When the number of granular casts per low-power field is 0, 1-5, or greater than or equal to 6, the score is 0, 1, or 2 points respectively. For example, a score of 2 is awarded when there are 6 or more renal tubular epithelial cells per HPF, and a score of 2 is awarded when there are 6 or more granular casts per LPF. The total score is 2 + 2 = 4 points, i.e., the USS score is 4. In this embodiment, a USS score greater than or equal to 1 indicates the AKI group, and a USS score of 0 indicates the non-AKI group. In STEP 106, chi-square tests were performed on three groups obtained using different criteria to calculate the significant difference (P-value) between the deep convolutional neural network model and the AKI and non-AKI groups. The chi-square test (X²-test) is a hypothesis testing method based on categorical variables. It makes statistical inferences by measuring the deviation between actual observed values ​​and theoretically inferred values ​​of a sample; the degree of deviation is measured by the chi-square value. The larger the chi-square value, the more significant the deviation. This method is mainly applied to comparing two or more rates or proportions, and to correlation analysis between categorical variables. Since the chi-square test is an existing technique, its specific working mechanism and calculation method will not be elaborated here.

[0024] In STEP 107, the rank sum test, also known as the ordinal sum test, is a nonparametric statistical method used to compare whether there is a significant difference in the distributions of two independent samples. It is applicable to data that does not satisfy the normal distribution assumption. The rank sum test is performed by arranging all observations (or the absolute value of the difference between each pair of observations) in ascending order, and numbering each observation (or the absolute value of the difference between each pair of observations) sequentially, which is called the rank (or order). The rank sum is calculated separately for each of the two groups of observations (in a paired design, the observations are divided into two groups based on the sign of the difference) for the test.

[0025] Let's illustrate STEP 106 and STEP 107 with further examples. For instance, regarding the grouping based on whether a clinical diagnosis is present, let's say there are 80 samples clinically diagnosed with AKI and 60 samples without AKI. Of the 80 samples, perhaps 50 have renal tubular epithelial cells, and of the 60, perhaps 10 have them. For clinical diagnosis, renal tubular epithelial cells are observed under a microscope, not directly detected by instruments. Clinically, microscopic examination is the gold standard and is reliable; instrumental examination can only be used as an auxiliary method and cannot be completely trusted. Then, if we test these 80 and 60 samples with an instrument, the results might be: 55 out of the 80 samples have renal tubules because the instrument might misidentify other structures as renal tubular epithelial cells or incorrectly classify renal tubules as other structures; and 8 out of the 60 samples have renal tubules.

[0026] In STEP 108, statistical methods, namely the chi-square test and rank-sum test from STEP 106 and 107, were used to statistically test the above data. If the calculated p-values ​​are all less than 0.001, this statistically proves a strong correlation between the results detected by our instrument and the results observed under a microscope in clinical diagnosis. In other words, the instrument results can replace the microscope results, and therefore, using instrument results as a reference for clinical diagnosis is effective.

[0027] Additionally, it's important to clarify that not all AKI patients have renal tubular epithelial cells, and non-AKI patients may also have them; it's just that AKI patients have a higher probability of having renal tubular epithelial cells. Therefore, the presence or absence of renal tubular epithelial cells is only one of many indicators in the diagnosis of acute kidney injury.

[0028] In other words, the statistical tests of STEP105 and STEP106 show that the results obtained by the instrument are reliable and can be used as a reference for disease diagnosis.

[0029] The following example illustrates this.

[0030] For a single individual's urine sample, a fully automated formed element analyzer using planar flow cytometry was used to acquire 2000 images per sample. Each image is 1440x1080 pixels in size, and each image typically contains many formed elements, such as... Figure 2 As shown, the image was segmented using a segmentation algorithm to obtain smaller images containing formed elements, with each smaller image containing only one formed element. The formed element images were collected, and the formed elements were manually labeled to form a dataset. The dataset consists of 8920 formed element images, including 1960 renal tubular epithelial cells and 6960 other components such as red blood cells, white blood cells, casts, crystals, and other epithelial cells.

[0031] A deep convolutional neural network was constructed, consisting of multiple convolutional layers, pooling layers, and a fully connected layer. The dataset was divided into training, validation, and test sets with proportions of 60%, 20%, and 20%, respectively. The cross-entropy loss function was used as the error function, and the network parameters were optimized using the Adam optimizer. The optimal network model was selected by plotting the ROC curve.

[0032] The model was used to detect renal tubular epithelial cells in urine samples from 328 patients in departments with a high incidence of AKI. Medical records were reviewed, and AKI was grouped according to KDIGO, clinical diagnosis, and USS.

[0033] In the three groups, the detection rates of renal tubular epithelial cells were 96.77% (60 / 62), 100% (9 / 9), and 93.16% (109 / 117), respectively.

[0034] Regarding the number of positive renal tubular epithelial cell samples detected, the model showed significant differences between the AKI group and the non-AKI group, with a c² value of 104.47 for the KDIGO group and 276.5 for the USS group. Regarding the numerical values ​​of renal tubular epithelial cell detection results, the model showed significant differences in the renal tubular epithelial cell detection values ​​between the AKI group and the non-AKI group in all three AKI subgroups (P values ​​were all less than 0.001). Therefore, the model results can be used as a reference indicator for the diagnosis of AKI.

[0035] Compared with existing technologies, the urine formed element analysis method provided by this invention first captures images using existing instruments, then segments them into small images containing only one type of formed element using existing image segmentation algorithms, and then constructs and trains a deep convolutional neural network model. The trained deep convolutional neural network model is then used to classify the small images containing formed elements. Urine samples from patients in departments with a high incidence of AKI are collected and grouped into AKI and non-AKI groups based on criteria such as clinical diagnosis, KDIGO conformation, and USS greater than or equal to 1. After classification, the chi-square test is used to calculate the p-value of the significant difference between the deep convolutional neural network model and the non-AKI group, and the rank-sum test is used to calculate the p-value of the significant difference between the deep convolutional neural network model and the non-AKI group. If the p-values ​​obtained by both statistical methods are less than 0.001, the classification result by the deep convolutional neural network model can be used as a reference indicator for AKI diagnosis, thereby reducing the reliance on microscopy and improving efficiency.

[0036] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions or improvements within the spirit of the present invention are covered within the scope of the claims of the present invention.

Claims

1. An analytical method based on formed elements in urine, characterized in that: The analytical method based on urine formed elements includes the following steps: STEP101: Provides a urine formed elements analyzer and acquires multiple formed element images of a urine sample using the urine formed elements analyzer; STEP102: The obtained multiple images of formed elements are segmented using an image segmentation method. Each segmented image contains a small image of only one type of formed element, namely red blood cells, white blood cells, casts, crystals, renal tubular epithelium, and squamous epithelium. STEP103: Annotate images with formed elements according to their types to form a dataset, and divide the dataset into training, validation and test sets according to a certain ratio; STEP104: Provide a deep convolutional neural network model, train the deep convolutional neural network model using the dataset, and fix the parameters of the trained network model to classify formed elements; STEP105: Use a trained deep convolutional neural network model to classify the images of renal tubular epithelial cells in the samples, and group them according to the criteria of whether there is a clinical diagnosis, whether it meets KDIGO, and whether USS is greater than or equal to 1. The samples are divided into AKI group and non-AKI group. The AKI group is clinically diagnosed as AKI, meets KDIGO, or has USS greater than or equal to 1. The non-AKI group is clinically diagnosed as non-AKI, does not meet KDIGO, or has USS of 0. STEP106: Use the rank-sum test to evaluate the numerical results of renal tubular epithelial cell detection. Calculate the significant differences in renal tubular epithelial cell detection values ​​for each group using a deep convolutional neural network model. If the calculated difference is less than 0.001, the model results meet the requirements; if the difference is greater than or equal to 0.001, the model does not meet the requirements. STEP106: For the above classification, for each of the three groups, perform a chi-square test according to the number of positive renal tubular cell samples detected to obtain the significant difference P-value between the deep convolutional neural network model and the non-AKI group. STEP107: For the above classification, for each of the three groups, according to the detection results of renal tubular epithelial cells, the rank-sum test is performed to obtain the significant difference P value between the deep convolutional neural network model and the non-AKI group. STEP108: If the P-values ​​obtained in STEP106 are all less than 0.001, and the P-values ​​obtained in STEP107 are also all less than 0.001, then the classification index of the deep convolutional neural network model is determined to be valid.

2. The analytical method based on urine formed elements as described in claim 1, characterized in that: The urine formed elements analyzer uses a planar flow cytometry method to acquire multiple images of the urine.

3. The analytical method based on urine formed elements as described in claim 1, characterized in that: The image segmentation method described is the Otsu method.

4. The analytical method based on urine formed elements as described in claim 1, characterized in that: The training set, validation set, and test set each account for 60%, 20%, and 20% of the total, respectively.

5. The analytical method based on urine formed elements as described in claim 1, characterized in that: According to the KDIGO classification criteria, the images containing renal tubular epithelial cells were divided into only the AKI group and the non-AKI group.

6. The analytical method based on urine formed elements as described in claim 1, characterized in that: The clinical diagnosis is made by using a microscope to determine whether the small images of the formed elements belong to the AKI group or the non-AKI group.

7. The analytical method based on urine formed elements as described in claim 1, characterized in that: When the number of renal tubular epithelial cells per high-power field is 0, 1-5, and the presence time is greater than or equal to 6, the USS score is 0, 1, and 2 points respectively.

8. The analytical method based on urine formed elements as described in claim 1, characterized in that: When the number of particle tubes per low magnification field of view is 0, 1 to 5, and the existence time is greater than or equal to 6, the USS score is 0, 1, and 2 points respectively.

Images