Medical data processing method and device, equipment and storage medium

Abstract

The invention relates to the technical field of data processing, and discloses a medical data processing method and device, equipment and a storage medium, and the method comprises the steps: obtaining ultrasonic images of a plurality of target objects; for each ultrasonic image, segmenting the ultrasonic image into a plurality of target areas; performing character recognition on the target area; and if a plurality of continuously appearing similar characters exist in the target area, identifying the characters in the target area by using a fine-grained classification OCR model. The efficiency and accuracy of medical data collection and arrangement can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a medical data processing method, device, equipment and storage medium. Background Art

[0002] In modern medical practice, the collection and processing of a large amount of medical data including ultrasound images face significant challenges. On the one hand, the problem of low efficiency is particularly prominent. Traditional data collection methods often rely on manual input and collation, which not only consumes a large amount of time of medical staff, but also delays the diagnosis and treatment process of patients.

[0003] On the other hand, the problem of low accuracy cannot be ignored either. Due to the intervention of human factors (such as fatigue, misoperation, etc.), errors are likely to occur during data recording and transcription, and these errors may seriously affect the accuracy and reliability of subsequent analysis. Especially when dealing with medical data such as ultrasound images that highly depend on detailed information, any small error may lead to misdiagnosis or missed diagnosis.

[0004] In summary, the current process of collecting and processing a large amount of medical data including ultrasound images is both time-consuming and error-prone, which constitutes a major obstacle to improving the quality and efficiency of medical services. Summary of the Invention

[0005] In view of this, the present invention provides a medical data processing method, device, equipment and storage medium to solve the problems of low efficiency and low accuracy in collecting and collating a large amount of medical data including ultrasound images.

[0006] In a first aspect, the present invention provides a medical data processing method, and the method includes:

[0007] Obtain ultrasound images of multiple target objects;

[0008] For each of the ultrasound images, segment the ultrasound image into multiple target regions;

[0009] Perform character recognition on the target regions;

[0010] If there are multiple similar characters that appear continuously in the target region, use a fine-grained classification OCR model to recognize the characters in the target region.

[0011] In an optional implementation manner, the fine-grained classification OCR model includes a multi-scale feature extraction network, an attention mechanism, a recurrent neural network and a recognition network;

[0012] Among them, the multi-scale feature extraction network is used to extract feature information of different scales in the target region; the feature information extracted by the multi-scale feature extraction network serves as the input feature map of the attention mechanism, and the attention mechanism is used to weight the feature map; the recurrent neural network is used to convert the target region into a data sequence and process it to capture the long-term dependencies in the data sequence; the recognition network is used to perform character recognition based on the feature information, the weighted feature map, and the output of the recurrent neural network.

[0013] In an alternative embodiment, the ultrasonic image includes a two-dimensional gray-scale scanned ultrasonic image;

[0014] After obtaining the ultrasonic images of multiple target objects, it further includes:

[0015] Construct a multi-scale image pyramid for the ultrasonic image;

[0016] Perform Laplace transform on each layer of the multi-scale image pyramid respectively to obtain the Laplace image corresponding to the scale;

[0017] For each scale of the Laplace image, calculate the Laplace variance corresponding to the scale;

[0018] Based on the Laplace variances of each scale, determine the total sharpness of the ultrasonic image;

[0019] Determine the blurriness of the ultrasonic image based on the total sharpness.

[0020] In an alternative embodiment, the determining the blurriness of the ultrasonic image based on the total sharpness includes:

[0021] Perform edge detection on the ultrasonic image to obtain the edge intensity of each pixel;

[0022] Based on the edge intensities of each pixel, perform weighted adjustment on the total sharpness to obtain a global sharpness score;

[0023] Determine the blurriness of the ultrasonic image based on the global sharpness score.

[0024] In an alternative embodiment, after obtaining the ultrasonic images of multiple target objects, it further includes:

[0025] Obtain the tracing result of the target structure in the ultrasonic image;

[0026] Set an edge area at the edge of the target structure based on the tracing result.

[0027] In an alternative embodiment, the medical data processing method further includes:

[0028] Obtain medical data other than the ultrasound image, and construct a target medical data set including the ultrasound image;

[0029] Construct a pseudo-random number generator that combines values processed by chaotic mapping, fractal transformation, and fuzzy logic;

[0030] For the sensitive information in the target medical data set, use the pseudo-random number generator and different control parameters to generate a pseudo-random sequence;

[0031] Process the pseudo-random sequence using the generation method in fractal geometry and / or fuzzy logic to obtain the final pseudo-random number;

[0032] Use the final pseudo-random number to replace the corresponding sensitive information in the target medical data set.

[0033] In an alternative embodiment, the processing of the pseudo-random sequence using the generation method in fractal geometry includes: performing trigonometric function operations on the pseudo-random sequence and introducing a decay term;

[0034] And / or,

[0035] The processing of the pseudo-random sequence using fuzzy logic includes: performing arithmetic processing on the pseudo-random sequence based on defined fuzzy rules and membership functions.

[0036] In a second aspect, the present invention provides a medical data processing device, which includes:

[0037] An ultrasound image acquisition module for acquiring ultrasound images of multiple target objects;

[0038] A segmentation model module for segmenting each ultrasound image into multiple target regions;

[0039] A character recognition module for performing character recognition on the target region;

[0040] A fine-grained recognition module for, if there are multiple similar characters continuously appearing in the target region, using a fine-grained classification OCR model to recognize the characters in the target region.

[0041] In a third aspect, the present invention provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the medical data processing method according to the first aspect or any corresponding embodiment thereof.

[0042] Fourthly, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to make a computer execute the medical data processing method according to the first aspect or any corresponding embodiment thereof.

[0043] Fifthly, the present invention provides a computer program product, including computer instructions, and the computer instructions are used to make a computer execute the medical data processing method according to the first aspect or any corresponding embodiment thereof.

[0044] The medical data processing method, device, equipment and storage medium provided by the embodiments of the present invention automatically recognize characters in an ultrasound image by region, and then for a region where multiple similar characters appear continuously, use a fine-grained classification OCR model to recognize the characters in the target region, which not only improves the sorting efficiency of a large amount of medical data including ultrasound images, but also improves the accuracy, thereby enhancing the efficiency, accuracy and reliability of subsequent data analysis results. Description of the Drawings

[0045] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related art, the following will briefly introduce the drawings required to be used in the description of the specific embodiments or the related art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0046] Figure 1 is a schematic flowchart of the medical data processing method according to the embodiment of the present invention;

[0047] Figure 2 is a schematic diagram of the region (part) division of the ultrasound image according to the embodiment of the present invention;

[0048] Figure 3 is a schematic diagram of the breast cancer clinical research data warehouse according to the embodiment of the present invention;

[0049] Figure 4 is a schematic overall flowchart of the collection and collation of medical data according to the embodiment of the present invention;

[0050] Figure 5 is a structural block diagram of the medical data processing device according to the embodiment of the present invention;

[0051] Figure 6 is a schematic hardware structure diagram of the computer device according to the embodiment of the present invention. Detailed Embodiments

[0052] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0053] According to an embodiment of the present invention, an embodiment of a medical data processing method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of executable computer instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0054] In this embodiment, a medical data processing method is provided, which can be used in various computer devices, and the computer device can be used as a medical device, etc. Figure 1 is a flowchart of the medical data processing method according to an embodiment of the present invention, as Figure 1 shown, the process includes the following steps:

[0055] Step S101, obtain ultrasonic images of multiple target objects.

[0056] Specifically, the ultrasonic image can be one or more of an elastogram, a two-dimensional grayscale ultrasound (2D Grayscale Ultrasound) (i.e., B-mode ultrasound (Brightness Mode, B-Mode)) image, an M-mode ultrasound (Motion Mode, M-Mode) image, a Doppler ultrasound image, etc.

[0057] For example, for each patient, 60 shear wave elastograms I 1 、I 2 ...I 60 are collected using the shear wave elastography mode. Taking the patient as the target object, multiple ultrasonic images of multiple target objects are obtained. After collecting the shear wave elastograms of the patient, the doctor will make an ultrasonic diagnosis record form (Case Report Form, CRF) based on the collected shear wave elastograms. Then, the ultrasonic images of each target object and the corresponding ultrasonic diagnosis record form can be combined into a folder Subsequently, the images in the corresponding folders can be read for multiple patients respectively to obtain the ultrasonic images of multiple target objects.

[0058] Step S102: For each of the ultrasonic images, segment the ultrasonic image into multiple target regions.

[0059] Specifically, when segmenting an ultrasonic image, it is divided according to the content display rule of the ultrasonic image, not necessarily according to an equally sized grid. The content display rule of the ultrasonic image is pre-set or default in the ultrasonic imaging device. Therefore, the content display rule of the ultrasonic image can be obtained in advance, and then the ultrasonic image segmentation method can be set according to the content display rule of the ultrasonic image, and finally the ultrasonic image is segmented according to the set segmentation method. In other embodiments, the entire ultrasonic image can also be directly recognized to identify the image display area and the character display area therein, that is, segmented according to the identified image display area and character display area. Or, identify the blank areas in the ultrasonic image. The blank areas include the interval areas between image displays, between image display and character display, and between character displays, and then screen the blank areas according to the size of the blank areas, and segment the ultrasonic image based on the screened blank areas.

[0060] For example, when segmenting the ultrasonic image as shown in Figure 2 the shear wave elasticity value, viscosity value, and strain elasticity value display area 201 of the tumor area on the ultrasonic image, the shear wave elasticity value, viscosity value, and strain elasticity value display area 202 of the Shell shell area, and the shear wave elasticity value, viscosity value, and strain elasticity value display area 203 of the area inside the Shell shell are segmented into three independent target regions.

[0061] In the embodiments of the present invention, the regions of the ultrasonic image are divided according to the content categories (such as character content, image content) in the ultrasonic image, so that character recognition can be performed only on the regions including character content, avoiding character recognition of image content, and improving the efficiency of recognizing a large amount of ultrasonic image content.

[0062] Step S103: Perform character recognition on the target regions.

[0063] Step S104: If there are multiple similar characters continuously appearing in the target region, use a fine-grained classification optical character recognition (OCR) model to recognize the characters in the target region.

[0064] For example, if there are adjacent similar characters "1" and "7", or adjacent similar characters "5" and "7", or adjacent similar characters "2" and ".", etc., use the OCR model to recognize the characters in the target region.

[0065] The process of character recognition can be specifically as follows: First, find all connected pixel groups, each group representing a possible character, and determine the position of the character according to the pixel distribution in the horizontal or vertical direction. Then, for each segmented character, OCR (Optical Character Recognition) technology can be used for recognition. For the recognized characters, the following methods can be used to judge their similarity: Edit Distance: Measures the minimum number of edit operations (insertion, deletion, replacement) required to convert one string into another; Feature Vector Comparison: If a deep learning model is used for character recognition, the feature vectors of each character can be extracted, and methods such as cosine similarity and Euclidean distance can be used to compare the distances of these vectors; Template Matching: Define standard character templates in advance, and then compare them with the recognition results to calculate the similarity score.

[0066] The medical data processing method provided by the embodiments of the present invention uses a fine-grained classification OCR recognition model to recognize similar characters continuously appearing in ultrasonic images, ensuring the accuracy and effectiveness of individual characters.

[0067] In some optional specific embodiments, the fine-grained classification OCR model includes a multi-scale feature extraction network, an attention mechanism, a recurrent neural network, and a recognition network;

[0068] Among them, the multi-scale feature extraction network is used to extract feature information of different scales in the target area; in the embodiments of the present invention, a multi-scale feature extraction network, such as a Multi-Scale Convolutional Neural Network (MSCNN), is introduced to extract features of different scales in ultrasonic images, so as to capture more information in fine-grained classification. Specifically, if the size of the input image of the target area is H×W, a multi-layer convolutional layer is used to extract multi-scale features F scale , which can be expressed as:

[0069] F scale ={f conv (I,k)∣k∈{k 1 ,k 2 ,…,k m}}

[0070] Among them, k is the size of different convolutional kernels, and m is the number of different convolutional kernels.

[0071] The feature information extracted by the multi-scale feature extraction network is used as the input feature map of the attention mechanism, and the attention mechanism is used to weight the feature map. Specifically, the attention weight A of the feature map can be determined according to the following formula:

[0072]

[0073] Among them, F is the input feature map, and d k is the feature dimension.

[0074] Then, the feature map is weighted using the determined attention weight A to enhance important features: F att = A · F, where F att is the weighted feature map.

[0075] In the embodiments of the present invention, the self-attention mechanism is used to enhance the attention of the fine-grained classification OCR model to important features and improve the sensitivity to details.

[0076] A Recurrent Neural Network (RNN) is used to convert the target region into a data sequence and process it to capture long-term dependencies in the data sequence. Specifically, the target region R j is converted into a data sequence S i : S j = flatten(R j ), and then, the data sequence is processed using RNN-LSTM to generate a hidden state H j : H j = RNN(S j ).

[0077] In the embodiments of the present invention, a recurrent neural network is used to process sequence data so as to consider the context relationship between characters when recognizing text characters.

[0078] The recognition network is used to perform character recognition based on the feature information, the weighted feature map, and the output of the recurrent neural network. In the embodiments of the present invention, the recognition network in the fine-grained classification OCR model combines the feature F scale extracted through multi-scale feature extraction, the feature F att processed by the attention mechanism, and the output H j of the RNN for final classification recognition. Specifically, features from different sources can be concatenated and fused to form a final feature representation F final : F final = concat(F scale , F att , H j ), and then the fully connected layer is used to classify the final feature, and the class probability P is output: P = softmax(W · F final + b).

[0079] In addition, after character recognition is performed on each target region of an ultrasound image, the character recognition results of each target region of the ultrasound image can be integrated to obtain a final text output:

[0080]

[0081] where N is the number of target regions of the ultrasound image.

[0082] The medical data processing method provided by the embodiments of the present invention introduces the combination of a multi-scale feature extraction network (MSCNN) and a self-attention mechanism, which improves the ability to extract detailed information in ultrasound images. In addition, a recurrent neural network (RNN) is used to process character sequences, improving the accuracy of OCR recognition, especially enhancing the sensitivity to details and the accuracy of character recognition, and reducing the risk of misrecognition caused by character similarity.

[0083] In some optional specific embodiments, the ultrasound image includes a two-dimensional gray-scale scanning ultrasound image; the two-dimensional gray-scale scanning ultrasound image can be acquired by a high-frequency ultrasound device;

[0084] After obtaining the ultrasound images of multiple target objects, the following steps are further included:

[0085] Step S105, constructing a multi-scale image pyramid for the ultrasound image.

[0086] Specifically, a Gaussian pyramid can be constructed. The Gaussian pyramid is a method of representing an image by downsampling. By performing multiple smoothing and shrinking operations on the original image (for example, a two-dimensional grayscale matrix with a size of H×W), a series of images with gradually decreasing resolutions are generated. Each layer of the image is a low-resolution version of the previous layer of the image.

[0087] Specifically, for each layer of the image pyramid (i.e., each layer of the image in the multi-scale image pyramid) I s , Gaussian blurs with different scales are used, and the scale factor for each layer is s (taking 1, 2, 4).

[0088] Step S106, performing Laplace transform on each layer of the image in the multi-scale image pyramid respectively to obtain the Laplace image L s .

[0089]

[0090] where (x, y) are pixel coordinates.

[0091] Step S107, for each scale of the Laplace image, calculating the Laplace variance V s : Vs = Var(L s (x,y)).

[0092] Step S108: Determine the overall sharpness of the ultrasonic image based on the Laplacian variances at each scale. Specifically, the Laplacian variances at each scale can be combined to obtain the overall sharpness V multi : where the weight w s is set according to experience or adaptively adjusted according to the differences in sharpness at different scales.

[0093] Step S109: Determine the blurriness of the ultrasonic image based on the overall sharpness. The blurriness indicates whether the ultrasonic image is blurry or clear.

[0094] In some optional specific embodiments, Step S109, that is, determining the blurriness of the ultrasonic image based on the overall sharpness, includes:

[0095] Step S1091: Perform edge detection on the ultrasonic image to obtain the edge intensity of each pixel.

[0096] Specifically, the Sobel operator is used to obtain the edge intensity of each pixel. Additionally, before performing edge detection on the ultrasonic image, noise reduction processing can also be performed on the ultrasonic image, and the ultrasonic image after noise reduction processing is denoted by I smooth . The edge intensity of each pixel is:

[0097]

[0098] where (x,y) is the pixel coordinate.

[0099] Step S1092: Perform weighted adjustment on the overall sharpness based on the edge intensities of each pixel to obtain a global sharpness score.

[0100] Specifically, the global sharpness score V final obtained after edge weighting is:

[0101]

[0102] where α is a parameter for adjusting the edge weight, reflecting the influence of the edge on the image sharpness.

[0103] In the embodiments of the present invention, through edge weighting, the character edges can be made clearer, thereby improving the recognition accuracy of characters. Especially for similar characters that appear adjacent to each other, when the edges are clear, the probability of recognition errors can be greatly reduced.

[0104] Step S1093: Determine the blurriness of the ultrasonic image based on the global sharpness score.

[0105] Specifically, a threshold T can be used for judgment. If V final < T, it is considered that the image is blurred; otherwise, it is considered clear.

[0106] In the embodiments of the present invention, by combining the adaptive multi-scale Laplace transform and the edge detection algorithm, the clarity and blurriness of the ultrasonic image are automatically evaluated. Subsequently, based on the blurriness of the ultrasonic image, the ultrasonic images can be screened to ensure that high-quality ultrasonic images are used for subsequent analysis.

[0107] In some optional specific embodiments, after obtaining the ultrasonic images of multiple target objects, the following steps are further included:

[0108] Step S110: Obtain the tracing result of the target structure in the ultrasonic image. Specifically, a software package in related technologies can be used to trace the contour of the target structure (such as a tumor) to obtain the tracing result.

[0109] Step S111: Set an edge area at the edge of the target structure based on the tracing result.

[0110] For example, regarding the target result of a tumor, what clinical research aims to explore is within what range around the tumor edge the viscoelasticity of its tissue can best predict the invasiveness of the tumor, where the invasiveness includes intravascular tumor thrombus, perineural invasion, and distant lymph node metastasis. Based on the above requirements, two-dimensional ultrasonic images of breast masses of breast cancer patients are collected on an ultrasonic instrument, and the edges of the tumor nodules are traced. Multiple sections are selected for each mass to reduce the error of viscoelasticity. On the clearest and most standard section image among them, the tumor area (A) is marked, and a "Shell ring" (which can also be called a Shell shell) is set at the traced tumor edge using the ultrasonic system tool package. The Shell ring represents the area of the tissue around the tumor. The thickness of the Shell ring can be manually adjusted with an interval difference of 0.5 mm, making the Shell extend 3 mm into the tumor (represented by "-") and 5 mm outside the tumor (represented by "+"). The marking result is used for subsequent analysis and data processing.

[0111] In addition, according to the set rules, the results obtained from the above processing (including the recognized characters, the blurriness of the ultrasonic image, the edge area of the target structure, etc.) can be structured and stored in a data table. The data table includes but is not limited to the basic patient information, ultrasonic image information (the corresponding shear wave elastography values (i.e., shear wave elasticity value E (kPa)), strain elastography values (i.e., strain elasticity value) Strain, and viscosity value Vi (Pa·s), each including MEAN (mean), MAX (maximum), MIN (minimum), SD (standard deviation)), the ultrasonic record form diagnosed by the doctor, etc. According to ultrasonic knowledge and clinical experience, a mechanism for auditing numerical values is set to further verify and screen the structured data. Among them, the auditing mechanism includes but is not limited to numerical range checking, outlier detection, and logical consistency verification, etc.

[0112] In the embodiment of the present invention, the unstructured image data is converted into a structured format, which is convenient for continuous processing and analysis of large-scale data. The collected ultrasonic image data and related diagnostic information are structured and stored, and an auditing mechanism is set. Through numerical range checking and logical consistency verification, the accuracy and consistency of the image analysis results are ensured, human errors are reduced, and the reliability and usability of the data are improved.

[0113] In some optional specific embodiments, the medical data processing method further includes:

[0114] Step S112, obtaining medical data other than the ultrasonic image, and constructing a target medical data set including the ultrasonic image;

[0115] For example, after collecting the ultrasonic image of the target object, the doctor will make an ultrasonic diagnosis record form (Case Report Form, CRF) based on the collected ultrasonic image. The data in this form is the medical data other than the ultrasonic image. Combining the ultrasonic images of each target object with the corresponding ultrasonic diagnosis record form together can construct the target medical data set.

[0116] Step S113, constructing a pseudo-random number generator that combines values processed by chaotic mapping, fractal transformation, and fuzzy logic.

[0117] A pseudo-random number generator (PRNG) that combines values processed by chaotic mapping, fractal transformation, and fuzzy logic is a complex system that combines multiple mathematical techniques to enhance randomness and unpredictability. This combined approach aims to utilize the unique properties of each technique to generate high-quality pseudo-random numbers. Chaotic mapping is a method of generating seemingly random but actually deterministic sequences through simple non-linear equations. Common chaotic mappings include the Logistic Map, Henon Map, etc. These mappings are extremely sensitive to initial conditions, making them very suitable for use in PRNGs. Fractal transformation involves using recursive algorithms to generate geometric shapes with self-similarity, and its recursive nature can be applied to the generation of numerical sequences, thereby increasing the complexity and randomness of the output. Fuzzy logic allows for the processing of imprecise or uncertain information, which is very useful for simulating the uncertainties in natural phenomena. In a PRNG, fuzzy set theory can be used to adjust or mix data from different sources to achieve a smooth transition and increase randomness.

[0118] The constructed pseudo-random number generator can be, for example:

[0119] x n+1 =f(x n )=r·x n ·(1 - x n )

[0120] where x n is the input information, such as the characters corresponding to sensitive information, or the sequence of chaotic behaviors (also a pseudo-random sequence) generated by the characters corresponding to sensitive information through one or more iterations, and r is the control parameter.

[0121] Step S114, for the sensitive information in the target medical dataset, use the pseudo-random number generator and different control parameters to generate a pseudo-random sequence.

[0122] Specifically, different control parameters r can be used to obtain a pseudo-random sequence through multiple iterations: x n =f n (x 0 )。

[0123] The specific process of generating a pseudo-random sequence can be, for example:

[0124] 1. Initialization: Select appropriate chaotic mapping parameters and initial values.

[0125] 2. Generate sequence: Use the selected chaotic mapping to generate a preliminary sequence.

[0126] 3. Fractal transformation: Use the sequence generated by the chaotic mapping as input and further transform it through a certain form of fractal algorithm (such as the Iterated Function System IFS) to generate a more complex pattern.

[0127] 4. Fuzzy logic adjustment: Design a set of fuzzy rules to adjust the weights and combine the results based on the results of chaotic mapping and fractal transformation.

[0128] 5. Output: Generate the final pseudo-random number sequence.

[0129] In step S115, use the generation method in fractal geometry and / or fuzzy logic to process the pseudo-random sequence to obtain the final pseudo-random number.

[0130] In the embodiment of the present invention, in order to further enhance randomness, a fractal structure and / or fuzzy logic is introduced to transform the pseudo-random sequence generated by the pseudo-random number generator, increasing its complexity.

[0131] In some optional specific embodiments, the use of the generation method in fractal geometry to process the pseudo-random sequence includes: performing trigonometric function operations on the pseudo-random sequence and introducing a decay term. The specific processing formula can be, for example:

[0132]

[0133] In the embodiment of the present invention, the generation method in fractal geometry is used to perform trigonometric function operations and introduce a decay term (fractal characteristics) to increase the randomness and unpredictability of the generated numbers.

[0134] In some optional specific embodiments, the use of fuzzy logic to process the pseudo-random sequence includes: performing arithmetic operations on the pseudo-random sequence based on the defined fuzzy rules and membership functions. The specific arithmetic operation formula is: y n = Fuzzify(x n ) = μ a (x n ) + μ b (x n )·(1 - x n ), where μa and μb are the defined fuzzy membership functions, representing different fuzzy sets.

[0135] In the embodiment of the present invention, fuzzy logic is applied to the generated pseudo-random sequence. By defining fuzzy rules and membership functions, the output result has greater uncertainty.

[0136] In addition, if both the fractal generation method and fuzzy logic are used to transform the pseudo-random sequence generated by the pseudo-random number generator. Then, the processing results of fractal generation and the results of using fuzzy logic can be combined. For example, they can be combined according to the following formula:

[0137]

[0138] where r n is the final pseudo-random number, and α is an adjustment parameter to control the final output characteristics.

[0139] Step S116: Replace the corresponding sensitive information in the target medical dataset with the final pseudo-random number. In other embodiments, the sensitive information in the target medical dataset may also be encrypted using the final pseudo-random number.

[0140] In the embodiments of the present invention, a data security processing technology based on a pseudo-random number generator (PRNG) is adopted to ensure the desensitization processing of patient sensitive information. By combining a chaotic map, fractal, and fuzzy logic in the pseudo-random number generator, the security of data processing and the protection of patient privacy are enhanced, which helps to meet the security-related requirements for medical data processing.

[0141] In the embodiments of the present invention, the sensitive information of patients in the data table obtained by the above-mentioned structured processing can be desensitized to ensure the privacy and security of patients. The processed data table includes the relevant numerical values of all ultrasound images and an ultrasound diagnosis record form (Case Report Form, CRF) for subsequent clinical research and scientific analysis. Specifically, the structured data in the desensitized data table can be output in multiple formats for subsequent clinical research and analysis. For example, as Figure 3 shown, a metadata knowledge base, data governance, ad-hoc query, online analytical processing (OLAP), data mining tools are provided to automatically generate statistical charts, correlation analysis charts, ROC (Receiver Operating Characteristic) curves, etc. to support clinical decision-making.

[0142] In the embodiments of the present invention, multiple data output formats are supported, and a metadata knowledge base, data governance, ad-hoc query, OLAP analysis, and data mining tools are provided, which helps doctors quickly interpret the data results and assist in clinical decision-making.

[0143] The medical data processing method provided in this embodiment automatically recognizes characters in ultrasound images by region, and then for regions where multiple similar characters appear continuously, a fine-grained classification OCR model is used to recognize the characters in the target region, which not only improves the collation efficiency of a large amount of medical data including ultrasound images, but also improves the accuracy, thereby enhancing the efficiency, accuracy, and reliability of subsequent data analysis results.

[0144] In summary, the medical data processing method provided by the embodiments of the present invention realizes the automated collection and processing of medical data through Robotic Process Automation (RPA), greatly improving the efficiency of medical data collection and processing. It solves the problem in the related art that the collection of medical clinical and research data mainly relies on manual operations, such as manually selecting images and manually inputting data, resulting in low collection efficiency.

[0145] In addition, the embodiments of the present invention provide strong support for data analysis by setting up an audit value mechanism and structured storage, making the data analysis process more efficient, accurate, and reliable. It solves the problem in the related art that the medical data analysis process may be more cumbersome and requires professional medical knowledge and data analysis skills.

[0146] Moreover, the embodiments of the present invention introduce technical means such as a clarity evaluation algorithm and the setting of an edge tracing Shell ring for the target structure to ensure the quality and accuracy of the collected images. For similar characters, fine-grained classification OCR models are used for recognition and structured storage, improving the accuracy and consistency of the data.

[0147] The embodiments of the present invention provide a medical data processing method, which can be used to realize intelligent RPA (Robotic Process Automation) collection and processing of breast cancer clinical and research data. As Figure 4 shown, the specific process is as follows:

[0148] Step S401: Perform two-dimensional gray-scale scanning (imaging) and evaluate the clarity of the obtained two-dimensional gray-scale ultrasound images.

[0149] Step S402: Image tracing and Shell ring setting.

[0150] Step S403: Shear wave elastography and data integration (collection).

[0151] Step S404: Automatic image segmentation and fine-grained classification OCR recognition.

[0152] Step S405: Structured storage and setting of the audit value mechanism.

[0153] Step S406: Secure processing and data desensitization based on PRNG.

[0154] Step S407: Data output and visualization.

[0155] In this embodiment, a medical data processing device is further provided. This device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0156] This embodiment provides a medical data processing device, as Figure 5 shown, including:

[0157] An ultrasonic image acquisition module 501, configured to acquire ultrasonic images of multiple target objects;

[0158] A segmentation model module 502, configured to segment each of the ultrasonic images into multiple target regions;

[0159] A character recognition module 503, configured to perform character recognition on the target regions;

[0160] A fine recognition module 504, configured to, if there are multiple similar characters that appear continuously in the target region, use a fine-grained classification OCR model to recognize the characters in the target region.

[0161] In some alternative implementation manners, the fine-grained classification OCR model includes a multi-scale feature extraction network, an attention mechanism, a recurrent neural network, and a recognition network;

[0162] Among them, the multi-scale feature extraction network is configured to extract feature information of different scales in the target region; the feature information extracted by the multi-scale feature extraction network is used as the input feature map of the attention mechanism, and the attention mechanism is configured to weight the feature map; the recurrent neural network is configured to convert the target region into a data sequence and process it to capture long-term dependencies in the data sequence; the recognition network is configured to perform character recognition based on the feature information, the weighted feature map, and the output of the recurrent neural network.

[0163] In some alternative implementation manners, the ultrasonic image includes a two-dimensional gray-scale scanned ultrasonic image;

[0164] The medical data processing device further includes:

[0165] An image pyramid construction module, configured to construct a multi-scale image pyramid for the ultrasonic image;

[0166] A transformation module, configured to perform Laplace transformation on each layer of the image in the multi-scale image pyramid respectively to obtain a Laplace image corresponding to the scale;

[0167] A variance calculation module, configured to calculate the Laplacian variance corresponding to each scale for the Laplacian image of each scale;

[0168] A total sharpness determination module, configured to determine the total sharpness of the ultrasonic image based on the Laplacian variances of each scale;

[0169] An ambiguity determination module, configured to determine the ambiguity of the ultrasonic image based on the total sharpness.

[0170] In some alternative embodiments, the ambiguity determination module includes:

[0171] An edge intensity acquisition unit, configured to perform edge detection on the ultrasonic image to acquire the edge intensity of each pixel;

[0172] A global sharpness score acquisition unit, configured to perform weighted adjustment on the total sharpness based on the edge intensities of each pixel to obtain a global sharpness score;

[0173] An ambiguity determination unit, configured to determine the ambiguity of the ultrasonic image based on the global sharpness score.

[0174] In some alternative embodiments, the medical data processing device further includes:

[0175] A tracing result acquisition unit, configured to acquire the tracing result of a target structure in the ultrasonic image;

[0176] An edge area setting unit, configured to set an edge area at the edge of the target structure based on the tracing result.

[0177] In some alternative embodiments, the medical data processing device further includes:

[0178] A medical data set construction module, configured to acquire medical data other than the ultrasonic image and construct a target medical data set including the ultrasonic image;

[0179] A pseudo-random number generator construction module, configured to construct a pseudo-random number generator by merging values processed by chaotic mapping, fractal transformation, and fuzzy logic processing;

[0180] A pseudo-random sequence generation module, configured to generate a pseudo-random sequence for sensitive information in the target medical data set by using the pseudo-random number generator with different control parameters;

[0181] A processing module, configured to process the pseudo-random sequence by using a generation method in fractal geometry and / or fuzzy logic to obtain a final pseudo-random number;

[0182] A replacement module for replacing corresponding sensitive information in the target medical dataset with the final pseudo-random numbers.

[0183] In some alternative embodiments, the processing module includes:

[0184] A fractal processing unit for performing trigonometric function operations and introducing attenuation terms on the pseudo-random sequence;

[0185] And / or,

[0186] A fuzzy processing unit for performing arithmetic processing on the pseudo-random sequence based on defined fuzzy rules and membership functions.

[0187] The further functional descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above, and will not be elaborated here.

[0188] The medical data processing device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0189] The embodiment of the present invention further provides a computer device having the above-mentioned Figure 5 shown medical data processing device.

[0190] Please refer to Figure 6 , Figure 6 which is a schematic structural diagram of a computer device provided by an alternative embodiment of the present invention. As shown in Figure 6 , the computer device includes: one or more processors 10, a memory 20, and an interface for connecting each component, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as an array of servers, a set of blade servers, or a multi-processor system). Figure 6 Taking one processor 10 as an example in

[0191] The processor 10 may be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 may further include a hardware chip. The above-mentioned hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above-mentioned programmable logic device may be a complex programmable logic device, a field-programmable gate array, a generic array logic, or any combination thereof.

[0192] Among them, the memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiments.

[0193] The memory 20 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may further include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0194] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 may further include a combination of the above types of memories.

[0195] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 may be connected through a bus or other means. Figure 6 Taking connection through a bus as an example.

[0196] The input device 30 can receive input digital or character information, and generate key signal inputs related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (such as an LED), and a tactile feedback device (such as a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light-emitting diode, a display, and a plasma display. In some alternative embodiments, the display device may be a touch screen.

[0197] The computer device further includes a communication interface for the computer device to communicate with other devices or communication networks.

[0198] Embodiments of the present invention also provide a computer-readable storage medium. The method according to the embodiments of the present invention can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0199] A part of the present invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should be able to understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Herein, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to the computer.

[0200] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A medical data processing method, characterized in that: The method comprises: acquiring ultrasound images of a plurality of target objects; For each of the ultrasound images, segment the ultrasound image into a plurality of target areas; Performing character recognition on the target area; If there are multiple similar characters appearing consecutively in the target area, a fine-grained classification OCR model is used to recognize the characters in the target area.

2. The method according to claim 1, characterized in that The fine-grained classification OCR model includes a multi-scale feature extraction network, an attention mechanism, a recursive neural network and a recognition network; Among them, the multi-scale feature extraction network is used to extract feature information of different scales in the target area; the feature information extracted by the multi-scale feature extraction network is used as the input feature map of the attention mechanism, and the attention mechanism is used to weight the feature map; the recursive neural network is used to convert the target area into a data sequence and process it to capture the long-term dependencies in the data sequence; the recognition network is used to perform character recognition based on the feature information, the weighted feature map and the output of the recursive neural network.

3. The method according to claim 1, characterized in that The ultrasonic image includes a two-dimensional grayscale scanning ultrasonic image; After acquiring the ultrasonic images of the plurality of target objects, the method further includes: constructing a multi-scale image pyramid for the ultrasound image; Performing Laplace transform on each layer of the image in the multi-scale image pyramid to obtain a Laplace image of a corresponding scale; For the Laplace image of each scale, calculating the Laplace variance of the corresponding scale; Determining the overall clarity of the ultrasound image based on the Laplace variance at each scale; A blurriness of the ultrasound image is determined based on the overall sharpness.

4. The method according to claim 3, characterized in that The determining the blurriness of the ultrasound image based on the overall clarity comprises: Performing edge detection on the ultrasound image to obtain edge strength of each pixel; Performing weighted adjustment on the total clarity based on the edge strength of each pixel to obtain a global clarity score; Based on the global clarity score, a blurriness of the ultrasound image is determined.

5. The method according to claim 1, characterized in that After acquiring the ultrasonic images of the plurality of target objects, the method further includes: Acquire a tracing result of a target structure in the ultrasound image; An edge region is set at the edge of the target structure based on the tracing result.

6. The method according to claim 1, characterized in that Also includes: Acquire medical data other than the ultrasound image, and construct a target medical data set including the ultrasound image; Construct a pseudo-random number generator based on the combination of chaotic mapping, fractal transformation and fuzzy logic processed values; For sensitive information in the target medical data set, using the pseudo-random number generator and different control parameters to generate a pseudo-random sequence; Processing the pseudo-random sequence using a generation method in fractal geometry and / or fuzzy logic to obtain a final pseudo-random number; The final pseudo-random number is used to replace the corresponding sensitive information in the target medical data set.

7. The method according to claim 6, characterized in that The method of using a generation method in fractal geometry to process the pseudo-random sequence includes: performing trigonometric function operations on the pseudo-random sequence and introducing an attenuation term; and / or, The using fuzzy logic to process the pseudo-random sequence includes: performing operation processing on the pseudo-random sequence based on defined fuzzy rules and membership functions.

8. A medical data processing device, characterized in that: The device comprises: An ultrasonic image acquisition module, used to acquire ultrasonic images of multiple target objects; A segmentation model module, used for segmenting each of the ultrasound images into a plurality of target areas; A character recognition module, used for performing character recognition on the target area; The fine recognition module is used to use a fine-grained classification OCR model to recognize the characters in the target area if there are multiple similar characters appearing continuously in the target area.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the medical data processing method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the medical data processing method according to any one of claims 1 to 7.

Images