Urinary calculus image recognition method based on machine vision

Abstract

The invention belongs to the technical field of urinary calculus image recognition, and relates to a urinary calculus image recognition method based on machine vision, which comprises the following steps: acquiring medical image data of urinary calculus in real time, performing multi-modal image preprocessing and feature preliminary extraction, and generating a standardized calculus image data set; based on the standardized calculus image data set, extracting a multi-dimensional calculus feature index, and comparing the multi-dimensional calculus feature index with a preset calculus feature threshold to obtain a preliminary judgment result of calculus recognition; based on the preliminary judgment result, the position and the type of the calculus are determined in combination with anatomical structure information in the image, and a corresponding feature enhancement model is called according to the calculus type. Standardization and feature optimization of the urinary calculus image are achieved through multi-modal image preprocessing and feature fusion, and the accuracy of the urinary calculus image is improved. And the accuracy and consistency of data processing are improved. On the basis of comparison of the characteristic indexes and the threshold values, the response speed and the accuracy degree of initial calculus recognition are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of urinary stone image recognition, and particularly relates to a urinary stone image recognition method based on machine vision. BACKGROUND

[0002] Urinary stone is a common disease, and its accurate diagnosis is crucial for treatment plan. Currently, the clinical diagnosis of urinary stone mainly relies on doctors' manual interpretation of medical images (such as ultrasound, CT, X-ray film, etc.). This approach has the following shortcomings:

[0003] Manual interpretation is low in efficiency and difficult to meet the needs of large-scale screening;

[0004] The diagnosis result is greatly influenced by doctors' experience and subjective factors, and there is a risk of missed diagnosis and misdiagnosis;

[0005] For complex cases (such as multiple stones and stones with blurred boundaries with surrounding tissues), the recognition is difficult;

[0006] Existing computer-aided diagnosis algorithms need to be improved in the comprehensiveness of stone feature extraction and the robustness of classification models, especially in the adaptability when dealing with images obtained by different imaging devices.

[0007] Therefore, there is an urgent need for a computer-aided method that can efficiently and accurately recognize urinary stones to improve the diagnosis efficiency and accuracy. SUMMARY

[0008] The purpose of the present application is to provide a urinary stone image recognition method based on machine vision, which solves the problems of low efficiency and poor accuracy of manual interpretation in the prior art and the insufficient robustness of existing algorithms.

[0009] The purpose of the present application can be achieved by the following technical solutions:

[0010] A urinary stone image recognition method based on machine vision, comprising the following steps:

[0011] Real-time acquisition of medical image data of urinary stones, and multi-modal image preprocessing and preliminary feature extraction to generate a standardized stone image dataset;

[0012] Based on the standardized stone image dataset, multi-dimensional stone feature indicators are extracted and compared with preset stone feature thresholds to obtain a preliminary determination result of stone recognition;

[0013] Based on the preliminary determination result, the position and type of the stone are determined in combination with the anatomical structure information in the image, and the corresponding feature enhancement model is called according to the type of the stone to generate stone feature enhancement data;

[0014] According to the stone characteristic enhancement data, a deep learning classification model is called to analyze the texture, shape and density characteristics of the stone, and a stone confidence factor of the current image is calculated;

[0015] According to the stone confidence factor and the stone type, the classification parameters of stone recognition are optimized to form a final stone recognition result;

[0016] Based on the final stone recognition result, a diagnosis report is generated in combination with the clinical information of the patient, and the recognition result is visually labeled.

[0017] Preferably, the acquisition step of the standardized stone image data set is:

[0018] Based on the original image obtained by the multi-modal imaging device, image registration and gray scale standardization are performed, noise reduction processing is performed on the image, and fuzzy areas are removed to obtain a pre-processed multi-modal image;

[0019] Based on the pre-processed multi-modal image, edge and texture features of the image are extracted;

[0020] Based on the edge and texture features, the modal type and acquisition timestamp of the image are matched, the features are integrated with the image identifier, and a standardized stone image data set is generated.

[0021] Preferably, the acquisition step of the preliminary determination result of the stone recognition is:

[0022] Based on the standardized stone image data set, the basic features of each image recorded in the data are retrieved, and the image feature sequence is formed in ascending order according to the feature significance;

[0023] Based on the image feature sequence, a preset stone feature threshold value is called, and the difference between each image feature and the threshold value is compared one by one, the absolute value of the difference is calculated, the difference absolute value is sorted, and the sorting result is determined.

[0024] Based on the sorting result, it is judged whether there is a feature difference absolute value exceeding the threshold value, if there is, it is determined as a suspected stone area, otherwise it is determined as normal tissue, and a preliminary determination result is generated.

[0025] Preferably, the acquisition step of the stone characteristic enhancement data is:

[0026] Based on the preliminary determination result, the anatomical structure identifier, the gray scale distribution and the texture features of the stone suspected area in the standardized data set are retrieved, the features are aggregated according to a unified timestamp, and a feature adaptation data group is obtained;

[0027] Based on the feature adaptation data group, a stone type adaptation value is calculated;

[0028] Based on the type adaptation value, set the stone type label corresponding to different intervals, if the adaptation value is in a certain interval, call the corresponding feature enhancement model to generate stone feature enhancement data.

[0029] Preferably, the stone confidence factor acquisition step is:

[0030] Based on the stone feature enhancement data, retrieve the texture feature vector, shape descriptor and density distribution value in the data, eliminate abnormal values by checking the time stamp and data integrity, and form a feature data set;

[0031] Based on the feature data set, extract standardized texture features, shape features and density features, and normalize them with reference values;

[0032] Based on the normalized features, input them into the pre-trained deep learning classification model to calculate the stone confidence factor of the current image.

[0033] Preferably, the final stone recognition result acquisition step is:

[0034] According to the stone confidence factor and the stone type, call the current image classification parameters recorded in the standardized data set, compare the parameters with the standard reference values corresponding to the types item by item, mark the classification parameters with differences exceeding the preset range, and form a parameter difference set;

[0035] Based on the parameter difference set, adjust the weight parameters and classification threshold of the deep learning model for the regions with confidence lower than the threshold, and execute correction region by region to generate a corrected classification parameter set;

[0036] Based on the corrected classification parameter set, verify the feature matching degree of each suspected stone region item by item, if the matching degree exceeds the allowable threshold, adjust the feature extraction weight to reduce the matching degree to within the threshold, and form the final stone recognition result.

[0037] Preferably, the diagnostic report generation step is:

[0038] Based on the final stone recognition result, extract the position, size, number, type and other information of the stone, call the clinical information of the patient, compare the association of stone features and clinical symptoms one by one, and generate a diagnostic report containing a diagnostic conclusion and suggestions;

[0039] According to the diagnostic report, color mark the stone region through the visual interface, and superimpose the anatomical structure diagram to form a visual diagnostic result.

[0040] The application has the beneficial effects that the application realizes standardization and feature optimization of urinary stone images through multi-modal image preprocessing and feature fusion, and improves the accuracy and consistency of data processing. Based on comparison of feature indexes and threshold values, the response speed and accuracy of preliminary stone identification are improved, which helps to reduce missed diagnosis. Through integration of preliminary determination results and anatomical structure information, accurate positioning of stone position and type is realized, and the global analysis capability of the system is enhanced. In the deep learning identification process, multi-dimensional features are analyzed and a confidence factor is calculated, and the classification parameters are optimized based on the factor, so that the identification result is more in line with clinical needs, and the diagnostic accuracy is improved. The diagnostic report is generated combined with clinical information and visualized, which further ensures the readability and clinical practicability of the diagnostic result, and improves the diagnostic efficiency and decision accuracy of doctors. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0042] Figure 1 The flowchart of the urinary stone image recognition method based on machine vision of the application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the application will be described clearly and completely in the following with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0044] Please refer to Figure 1 The application is a urinary stone image recognition method based on machine vision, which comprises the following steps:

[0045] Real-time acquisition of medical image data of urinary stones, and multi-modal image preprocessing and feature preliminary extraction to generate a standardized stone image data set.

[0046] Based on the standardized stone image data set, multi-dimensional stone feature indexes are extracted and compared with preset stone feature threshold values to obtain preliminary determination results of stone recognition.

[0047] Based on the preliminary determination results, the position and type of the stone are determined in combination with the anatomical structure information in the image, and the stone feature enhancement data is generated according to the corresponding feature enhancement model called according to the stone type.

[0048] According to the stone feature enhancement data, when further improvement of recognition accuracy is needed, a deep learning classification model is called to analyze the texture, shape and density features of the stone, and a stone confidence factor of the current image is calculated.

[0049] According to the stone confidence factor and the stone type, the classification parameters of stone recognition are optimized to form the final stone recognition result.

[0050] Based on the final stone recognition result, a diagnosis report is generated in combination with the clinical information of the patient, and the recognition result is visually marked.

[0051] The acquisition step of the standardized stone image data set is:

[0052] Based on the original images obtained by the multi-modal imaging equipment (ultrasound, CT, X-ray), image registration and gray scale standardization are performed, noise reduction processing is performed on the images, and fuzzy areas are removed to obtain pre-processed multi-modal images.

[0053] Based on the pre-processed multi-modal images, the basic features (edges, contours) of the images are extracted.

[0054] Based on the basic features, the modal type and acquisition timestamp of the image are matched, the basic features are integrated with the image identification, and the standardized stone image data set is generated.

[0055] The acquisition step of the preliminary determination result of the stone recognition is:

[0056] Based on the standardized stone image data set, the basic features of each image recorded in the data are retrieved, and the features are arranged in ascending order according to their significance to form an image feature sequence.

[0057] Based on the image feature sequence, a preset stone feature threshold is called, and the differences between each image feature and the threshold are compared one by one, the absolute values of the differences are calculated, the differences are sorted, and the sorting result is determined.

[0058] Based on the sorting result, it is determined whether there is a feature difference absolute value that exceeds the threshold value, if there is, it is determined as a suspected stone area, otherwise it is determined as normal tissue, and a preliminary determination result is generated.

[0059] The acquisition step of the stone feature enhancement data is:

[0060] Based on the preliminary determination result, the anatomical structure identification, the gray scale distribution and the texture features of the stone suspected area in the standardized data set are retrieved, the features are aggregated according to a unified timestamp, and a feature adaptation data group is obtained.

[0061] Based on the feature adaptation data set, a stone type adaptation value is calculated.

[0062] Based on the type adaptation value, stone type labels corresponding to different intervals are set, and if the adaptation value is in a specific interval, a corresponding feature enhancement model (such as a texture enhancement model for uric acid stones or an edge enhancement model for calcium oxalate stones) is called to generate stone feature enhancement data.

[0063] The stone confidence factor acquisition step is:

[0064] Based on the stone feature enhancement data, texture feature vectors, shape descriptors, and density distribution values in the data are retrieved, abnormal values are eliminated by checking the time stamp and data integrity, and a feature data set is formed.

[0065] Based on the feature data set, standardized texture features, shape features, and density features are extracted and normalized using reference values.

[0066] Based on the normalized features, input into a pre-trained deep learning classification model (such as an improved U-Net or ResNet), the stone confidence factor of the current image is calculated.

[0067] The final stone recognition result acquisition step is:

[0068] According to the stone confidence factor and the stone type, the current image classification parameters recorded in the standardized data set are called, and the parameters are compared with the standard reference values corresponding to the type, and the classification parameters with differences exceeding the preset range are marked to form a parameter difference set.

[0069] Based on the parameter difference set, for regions with a confidence factor below the threshold, the weight parameters and classification threshold of the deep learning model are adjusted, and the correction is performed region by region to generate a corrected classification parameter set.

[0070] Based on the corrected classification parameter set, the feature matching degree of each suspected stone region is verified, and if the matching degree exceeds the allowable threshold, the matching degree is reduced to within the threshold by adjusting the feature extraction weight to form the final stone recognition result.

[0071] The diagnostic report generation step is:

[0072] Based on the final stone recognition result, the position, size, number, type, and other information of the stone are extracted, the clinical information (age, gender, medical history) of the patient is called, and the relevance of the stone features and clinical symptoms is compared one by one to generate a diagnostic report containing the stone diagnosis conclusion and suggestions.

[0073] According to the diagnostic report, the stone area is color-labeled through a visual interface, and an anatomical structure diagram is superimposed to form a visual diagnostic result.

[0074] Specifically, the application further includes the following embodiments.

[0075] A urinary stone image recognition method based on machine vision

[0076] Image data acquisition and preprocessing

[0077] Use an ultrasonic device (frequency 3-5 MHz), a 64-slice spiral CT (slice thickness 1 mm), and a digital X-ray machine to obtain the original images of the patient's urinary system. Perform spot noise reduction (use Lee filter) on the ultrasound images, metal artifact correction on the CT images, and contrast enhancement on the X-ray films.

[0078] Use Elastix software to perform rigid registration on the multi-modal images, standardize the image gray values to the [0, 255] interval, and remove fuzzy areas with a signal-to-noise ratio less than 3 to obtain the preprocessed images.

[0079] Extract the Canny edge features and LBP texture features of the preprocessed images, combine the image modal labels (such as CT, US) and the acquisition timestamps (such as 2025-06-25T10:30:00), and generate a standardized stone image dataset.

[0080] Preliminary determination result acquisition

[0081] Retrieve the LBP texture feature values of each image from the dataset, arrange them in ascending order of feature value size, and form a feature sequence.

[0082] The texture feature threshold of calcium oxalate stones is preset to 120 (mean value + 1.5 times standard deviation based on 1000 clinical sample statistics), and the absolute difference between each image feature and the threshold is calculated one by one.

[0083] If there is a feature with an absolute difference > 10, it is determined to be a suspected stone area, otherwise it is determined to be normal tissue, and a preliminary determination result is generated.

[0084] Stone feature enhancement data acquisition

[0085] Based on the preliminary determination result, retrieve the kidney contour label, CT value distribution (unit: HU) and GLCM texture features of the suspected area in the CT image, and aggregate to obtain a feature adaptation data set.

[0086] Calculate the type adaptation value: Where C k is the CT value of the kth region, F k is the mean CT value of the fat tissue in the region, and U kFor texture consistency, For overall texture consistency mean, W k For area of region (mm 2 ), p is the total number of regions.

[0087] If Λ < 0.8, it is determined to be calcium oxalate stones, and the edge enhancement model (based on the improved Canny operator) is called; if 0.8≤Λ<1.2, it is determined to be uric acid stones, and the texture enhancement model (Gabor filter set) is called to generate feature enhancement data.

[0088] Stone confidence factor calculation

[0089] From the feature enhancement data, Gabor texture features (5 scales, 8 directions), ellipse fitting shape parameters (major axis, minor axis, eccentricity), and CT value histogram are extracted, and abnormal points with CT value <-100HU or >300HU are removed to form a feature data set.

[0090] PCA dimensionality reduction is performed on the texture features (95% variance is retained), and Min-Max normalization is performed on the shape parameters and CT values.

[0091] The normalized features are input into the pre-trained ResNet50 model (trained on 10,000 stone images), and the stone confidence factor (range 0-1) is output.

[0092] Final recognition result generation

[0093] Set the confidence threshold to 0.7, if the confidence of a region is <0.7, adjust the last three layers of convolution kernel weights of ResNet50 (learning rate 0.001, iteration 50 times), and update the classification threshold.

[0094] Calculate the feature matching degree of the adjusted region: Where f i is the region feature, f i,ref is the standard feature of the corresponding type of stone, and n is the feature dimension. If M>0.3, adjust the frequency parameter of the Gabor filter to make M≤0.3, and form the final recognition result.

[0095] Diagnosis report generation

[0096] Extract the stone location (such as left lower calyx), size (long diameter 5mm), number (1), and type (calcium oxalate) in the final recognition result, call the patient's age (45 years old), gender (male), and history of hematuria, etc. Information, generate a diagnosis report: "left lower calyx single calcium oxalate stone, suggest extracorporeal shock wave lithotripsy treatment".

[0097] The stone area is marked with red using ITK-SNAP software, and a kidney anatomical diagram is superimposed on the CT image to form a visual diagnosis result.

[0098] In the description of the specification, the description of the terms "one embodiment", "an example", "a specific example" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0099] The above is only an example and a description of the concept of the present application, and those skilled in the art can make various modifications or supplements or replace with similar ways, as long as it does not deviate from the concept of the invention or exceed the scope defined by the present claims, which shall belong to the protection scope of the present application.

Claims

1. A machine vision-based image recognition method for urinary stones, characterized in that: Includes the following steps: Real-time acquisition of medical image data of urinary stones, multimodal image preprocessing and preliminary feature extraction, and generation of standardized stone image dataset; Based on the standardized stone image dataset, multi-dimensional stone feature indicators are extracted and compared with preset stone feature thresholds to obtain preliminary judgment results for stone identification. Based on the preliminary judgment results, the location and type of the stone are determined by combining the anatomical structure information in the image, and the corresponding feature enhancement model is called according to the stone type to generate stone feature enhancement data; Based on the enhanced stone feature data, a deep learning classification model is invoked to analyze the texture, shape, and density features of the stone, and to calculate the stone confidence factor of the current image. Based on the stone confidence factor and stone type, the classification parameters for stone identification are optimized to form the final stone identification result; Based on the final stone identification results, a diagnostic report is generated in conjunction with the patient's clinical information, and the identification results are visually labeled.

2. The image recognition method for urinary stones based on machine vision according to claim 1, characterized in that: The steps for obtaining the standardized stone image dataset are as follows: Based on the original image acquired by the multimodal imaging device, image registration and grayscale standardization are performed, noise reduction is applied to the image and blurred areas are removed to obtain the preprocessed multimodal image; Based on the preprocessed multimodal image, the edge and texture features of the image are extracted; Based on the aforementioned edge and texture features, the images are matched with their modality type and acquisition timestamp, and the features are integrated with image identifiers to generate a standardized stone image dataset.

3. The image recognition method for urinary stones based on machine vision according to claim 1, characterized in that: The steps for obtaining the preliminary determination result of the stone identification are as follows: Based on the standardized stone image dataset, the basic features of each image recorded in the data are retrieved and sorted in ascending order according to feature saliency to form an image feature sequence; Based on the image feature sequence, a preset stone feature threshold is called, and the difference between each image feature and the threshold is compared one by one. The absolute value of the difference is calculated, the absolute values ​​of the difference are sorted, and the sorting result is determined. Based on the sorting results, it is determined whether there are cases where the absolute value of the feature difference exceeds the threshold. If so, it is determined to be a suspected stone area; otherwise, it is determined to be normal tissue, and a preliminary judgment result is generated.

4. The image recognition method for urinary stones based on machine vision according to claim 1, characterized in that: The steps for obtaining the enhanced stone feature data are as follows: Based on the preliminary judgment results, the anatomical structure identifiers, grayscale distribution and texture features of suspected stone areas are retrieved from the standardized dataset. The features are then aggregated according to a unified timestamp to obtain a feature-fitted data set. Based on the feature-matching data set, calculate the stone type matching value; Based on the type adaptation value, stone type labels are set for different intervals. If the adaptation value is in a specific interval, the corresponding feature enhancement model is called to generate stone feature enhancement data.

5. The image recognition method for urinary stones based on machine vision according to claim 1, characterized in that: The steps for obtaining the stone confidence factor are as follows: Based on the stone feature enhancement data, texture feature vectors, shape descriptors and density distribution values ​​are retrieved from the data. Outliers are removed by verifying timestamps and data integrity to form a feature dataset. Based on the aforementioned feature dataset, standardized texture features, shape features, and density features are extracted and normalized using reference values. Based on the normalized features, the data is input into a pre-trained deep learning classification model to calculate the stone confidence factor of the current image.

6. The image recognition method for urinary stones based on machine vision according to claim 1, characterized in that: The steps for obtaining the final stone identification result are as follows: Based on the stone confidence factor and stone type, the current image classification parameters recorded in the standardized dataset are called, and the parameters are compared with the standard reference values ​​corresponding to the types. Classification parameters with differences exceeding the preset range are marked to form a parameter difference set. Based on the parameter difference set, for regions with confidence levels below the threshold, the weight parameters and classification threshold of the deep learning model are adjusted, and the correction is performed region by region to generate a corrected set of classification parameters. Based on the modified classification parameter set, the feature matching degree of each suspected stone area is verified one by one. If the matching degree exceeds the allowable threshold, the matching degree is reduced to within the threshold by adjusting the feature extraction weight, thus forming the final stone identification result.

7. The image recognition method for urinary stones based on machine vision according to claim 1, characterized in that: The steps for generating the diagnostic report are as follows: Based on the final stone identification results, information such as the location, size, number, and type of the stones are extracted. The patient's clinical information is called up, and the correlation between stone characteristics and clinical symptoms is compared one by one to generate a diagnostic report containing diagnostic conclusions and suggestions. Based on the diagnostic report, the stone area is colored and overlaid with an anatomical diagram through a visualization interface to form a visual diagnostic result.

Images