A medical imaging data segmentation method and system

Abstract

The present invention belongs to the field of image segmentation technology and discloses a method and system for segmenting medical image data. The method comprises the following steps: collecting a number of historical medical image data and performing image processing on the historical medical image data to obtain a number of image-processed historical medical image data with real labels; constructing a medical image data segmentation model based on the image-processed historical medical image data using an attention multi-scale fusion deep learning algorithm; collecting real-time medical image data and performing medical image data segmentation on the real-time medical image data using the medical image data segmentation model to obtain real-time medical image data segmentation results. The present invention solves the problems of insufficient segmentation accuracy, poor model generalization ability, and low automation and intelligence levels in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of image segmentation, and in particular relates to a method and system for segmenting medical image data. Background Art

[0002] Medical imaging data refers to image data acquired through imaging technology during medical diagnosis and treatment. This data is typically generated by various medical imaging devices, such as X-rays, computed tomography (CT), magnetic resonance imaging (MRI), ultrasound, and positron emission tomography (PET). Medical image data segmentation is a crucial step in medical image analysis and is crucial for disease diagnosis, treatment, and prognosis.

[0003] Existing medical image data segmentation technology has the following defects:

[0004] 1) Insufficient segmentation accuracy: Existing medical image data segmentation methods cannot accurately identify and segment subtle structures in complex medical images, resulting in segmentation results that cannot meet the high-precision requirements of clinical diagnosis;

[0005] 2) Poor model generalization: Existing medical image data segmentation models perform well on specific datasets, but have poor generalization capabilities on new datasets and are difficult to adapt to medical image data from different sources or types. Segmentation models also have limited adaptability to different disease states or different patients, requiring retraining or adjustment for specific situations.

[0006] 3) Low level of automation and intelligence: Model parameters are manually adjusted by professionals, which requires a lot of human intervention and is not only time-consuming but also prone to inconsistent results due to differences in personal experience. Parameter selection and optimization often rely on trial and error, lacking an automated and intelligent parameter adjustment mechanism. Summary of the Invention

[0007] In order to solve the problems of insufficient segmentation accuracy, poor model generalization ability, and low automation and intelligence in the existing technology, the present invention aims to provide a method and system for segmenting medical image data.

[0008] The technical solution adopted in the present invention is:

[0009] A method for segmenting medical image data comprises the following steps:

[0010] Collecting a number of historical medical image data, and performing image processing on the number of historical medical image data to obtain a number of image-processed historical medical image data with real labels;

[0011] Based on several historical medical imaging data after image processing, a medical imaging data segmentation model was constructed using the attention multi-scale fusion deep learning algorithm;

[0012] Real-time medical image data is collected, and a medical image data segmentation model is used to perform medical image data segmentation on the real-time medical image data to obtain a real-time medical image data segmentation result.

[0013] Furthermore, a plurality of historical medical image data are collected and image processed to obtain a plurality of image-processed historical medical image data with real labels, including the following steps:

[0014] Collecting a number of historical medical imaging data and corresponding data information, and performing data cleaning on the number of historical medical imaging data to obtain a number of cleaned historical medical imaging data;

[0015] performing normalization processing on a number of cleaned historical medical imaging data to obtain a number of normalized historical medical imaging data;

[0016] Performing Gaussian denoising on a number of normalized historical medical image data to obtain a number of denoised historical medical image data;

[0017] Performing image enhancement on a number of denoised historical medical image data to obtain a number of enhanced historical medical image data;

[0018] According to the data information of the enhanced historical medical image data, a true label is set for each enhanced historical medical image data to obtain a plurality of image-processed historical medical image data set with the true label.

[0019] Furthermore, based on several historical medical image data after image processing, a medical image data segmentation model is constructed using an attention multi-scale fusion deep learning algorithm, which includes the following steps:

[0020] Use the CNN algorithm to build a segmentation network, and introduce a multi-scale feature fusion mechanism to improve the segmentation network to obtain a multi-scale feature fusion network;

[0021] The attention mechanism is introduced to improve the multi-scale feature fusion network and obtain the initial medical image data segmentation model;

[0022] Using a number of image-processed historical medical image data, optimizing and training an initial medical image data segmentation model to obtain an optimized medical image data segmentation model, and generating a number of historical medical image data segmentation results;

[0023] Based on several historical medical image data segmentation results, the optimized medical image data segmentation model is analyzed for segmentation effect, and real-time segmentation effect analysis results are obtained;

[0024] According to the real-time segmentation effect analysis results, the swarm intelligence optimization algorithm is used to optimize the model parameters of the optimized medical image data segmentation model to obtain the final medical image data segmentation model.

[0025] Furthermore, the multi-scale feature fusion network includes a multi-scale input layer, a multi-scale feature extraction layer, and a multi-scale feature output layer connected in sequence;

[0026] By introducing the attention mechanism and improving the multi-scale feature fusion network, we can obtain the initial medical image data segmentation model, which includes the following steps:

[0027] An attention generation layer is set between the multi-scale input layer and the multi-scale feature extraction layer of the multi-scale feature fusion network;

[0028] An attention weighted fusion layer is set between the multi-scale feature extraction layer and the multi-scale feature output layer of the multi-scale feature fusion network;

[0029] After the multi-scale feature output layer, an attention adjustment layer is set to obtain the initial medical image data segmentation model.

[0030] Furthermore, using a number of image-processed historical medical image data, the initial medical image data segmentation model is optimized and trained to obtain an optimized medical image data segmentation model, and a number of historical medical image data segmentation results are generated, including the following steps:

[0031] The processed historical medical imaging data were divided into a model training set and a model test set in a ratio of 7:3;

[0032] Using the model training set, the initial medical image data segmentation model is optimized and trained to obtain a trained medical image data segmentation model, and several historical medical image data segmentation results are generated;

[0033] Use the model test set to test the trained medical image data segmentation model and obtain several predicted labels;

[0034] The model test accuracy is obtained by comparing the predicted labels with the real labels of the historical medical imaging data after image processing in the model test set;

[0035] If the model test accuracy is greater than the accuracy threshold, the optimized medical image data segmentation model is output; otherwise, the optimization training continues.

[0036] Furthermore, based on several historical medical image data segmentation results, the optimized medical image data segmentation model is subjected to segmentation effect analysis to obtain real-time segmentation effect analysis results, including the following steps:

[0037] Extracting historical feature vectors of a number of historical medical image data segmentation results, and performing standardization processing on the number of historical feature vectors to obtain a number of standardized historical feature vectors;

[0038] Based on several standardized historical feature vectors, a feature space is constructed, and the k-NN algorithm is used to obtain the k neighbors of each historical feature vector in the feature space;

[0039] According to the k neighbors of each historical feature vector, the segmentation effect analysis is performed to obtain the real-time segmentation effect analysis results of the optimized medical image data segmentation model.

[0040] Furthermore, segmentation effect analysis is performed based on the k neighbors of each historical feature vector to obtain real-time segmentation effect analysis results of the optimized medical image data segmentation model, including the following steps:

[0041] Obtain the Euclidean distance between each historical feature vector and its k neighbors, and obtain the segmentation effect consistency score based on the average distance of several Euclidean distances;

[0042] According to the standard deviation of several Euclidean distances, the uncertainty score of the segmentation effect is obtained;

[0043] According to the segmentation effect consistency score and the segmentation effect uncertainty score, the real-time segmentation effect analysis results of the optimized medical image data segmentation model are obtained.

[0044] Furthermore, based on the real-time segmentation effect analysis results, a swarm intelligence optimization algorithm is used to optimize the model parameters of the optimized medical image data segmentation model to obtain the final medical image data segmentation model, including the following steps:

[0045] Based on the real-time segmentation effect analysis results, the optimization objectives of the swarm intelligence optimization algorithm are defined, and the model parameters of the optimized medical image data segmentation model are encoded into individual vectors of the swarm intelligence optimization algorithm;

[0046] According to the optimization target and individual vector, a swarm intelligence optimization algorithm is used to iteratively search for the optimal individual in the search space.

[0047] Decoding the individual vector of the optimal individual to obtain the optimal model parameters of the optimized medical image data segmentation model;

[0048] According to the optimal model parameters, the optimized medical image data segmentation model is optimized to obtain the final medical image data segmentation model.

[0049] Furthermore, real-time medical image data is collected, and a medical image data segmentation model is used to perform medical image data segmentation on the real-time medical image data to obtain a real-time medical image data segmentation result, including the following steps:

[0050] Collecting real-time medical image data and performing preprocessing to obtain preprocessed real-time medical image data, and inputting the preprocessed real-time medical image data into a medical image data segmentation model;

[0051] Using a multi-scale input layer, the pre-processed real-time medical imaging data is sampled at multiple scales to obtain several real-time feature maps at different scales.

[0052] Use the attention generation layer to generate corresponding real-time attention weight values ​​based on several real-time feature maps;

[0053] Use a multi-scale feature extraction layer to extract real-time image features of each real-time feature map;

[0054] Use the attention weighted fusion layer to perform weighted fusion on several real-time image features according to the real-time attention weight value to obtain real-time weighted fusion features;

[0055] Use a multi-scale feature output layer to output the initial real-time medical image data segmentation results based on real-time weighted fusion features;

[0056] Using the attention adjustment layer, the initial real-time medical image data segmentation result is fine-tuned according to the real-time attention weight value to obtain the final real-time medical image data segmentation result.

[0057] A medical image data segmentation system is used to implement a segmentation method. The system comprises an image processing unit, a model building unit and a data segmentation unit which are connected in sequence.

[0058] The beneficial effects of the present invention are:

[0059] The present invention provides a medical image data segmentation method and system. By performing image processing on historical medical image data and setting real labels, high-quality training data can be provided for the segmentation model, thereby improving the segmentation accuracy of the segmentation model on real-time medical image data, accurately identifying and segmenting subtle structures in complex medical images, and ensuring that the segmentation results meet the high-precision requirements of clinical diagnosis. The model constructed using the attention multi-scale fusion deep learning algorithm can better capture key features in medical images and improve the model's generalization ability for medical image data of different types and sources, without the need to retrain or adjust the model for specific situations. The medical image data segmentation model is automatically constructed, reducing the need for manual intervention and parameter adjustment, realizing the automation and intelligence of the segmentation process, reducing errors caused by inconsistent human judgment, and improving the reliability of medical image segmentation.

[0060] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flowchart of the medical image data segmentation method in the present invention.

[0062] Figure 2 It is a structural block diagram of the medical image data segmentation system in the present invention. DETAILED DESCRIPTION

[0063] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0064] Example 1:

[0065] like Figure 1 As shown, this embodiment provides a method for segmenting medical image data, comprising the following steps:

[0066] S1: Collecting a number of historical medical image data and performing image processing on the historical medical image data to obtain a number of processed historical medical image data with real labels, including the following steps:

[0067] S1-1: collecting a number of historical medical imaging data and corresponding data information, and performing data cleaning on the number of historical medical imaging data to obtain a number of cleaned historical medical imaging data;

[0068] Data information includes patient information, diagnosis results, image acquisition parameters, etc.; preliminary screening of collected historical medical imaging data is carried out to exclude imaging data that does not meet quality standards, has incomplete information, or is not in a uniform format, thereby improving data quality;

[0069] S1-2: performing normalization processing on a number of cleaned historical medical imaging data to obtain a number of normalized historical medical imaging data;

[0070] Normalization includes size normalization and pixel value normalization to eliminate the size and image intensity differences caused by different imaging devices;

[0071] S1-3: performing Gaussian denoising on a number of normalized historical medical image data to obtain a number of denoised historical medical image data;

[0072] To reduce random noise in the image, Gaussian denoising helps improve image quality, making subsequent feature extraction and segmentation more accurate;

[0073] S1-4: performing image enhancement on a number of denoised historical medical image data to obtain a number of enhanced historical medical image data;

[0074] Perform enhancement processing on the denoised image data, such as contrast enhancement and sharpening, to highlight the key features in the image. Image enhancement helps improve the visibility and interpretability of the image, providing a better foundation for subsequent segmentation and annotation;

[0075] S1-5: setting a true label for each enhanced historical medical image data according to the data information of the enhanced historical medical image data, thereby obtaining a plurality of image-processed historical medical image data with the true labels set;

[0076] The true label may include the outline of the lesion area, the location of the organ, the type of lesion, etc., providing supervised learning samples for subsequent model training;

[0077] S2: Based on some historical medical imaging data after image processing, a medical imaging data segmentation model is constructed using the attention multi-scale fusion deep learning algorithm, including the following steps:

[0078] S2-1: Use the Convolutional Neural Networks (CNN) algorithm to build a segmentation network and introduce a multi-scale feature fusion mechanism to improve the segmentation network to obtain a multi-scale feature fusion network;

[0079] CNN extracts hierarchical features of images through structures such as convolutional layers and pooling layers, and introduces a multi-scale feature fusion mechanism to capture image details at different scales. The multi-scale feature fusion network consists of a multi-scale input layer, a multi-scale feature extraction layer, and a multi-scale feature output layer connected in sequence. By designing a multi-scale feature fusion network, it can extract and fuse features at different scales of medical imaging data, thereby capturing details and global information in medical imaging data.

[0080] S2-2: Introducing the attention mechanism, improving the multi-scale feature fusion network, and obtaining the initial medical image data segmentation model, including the following steps:

[0081] S2-2-1: An attention generation layer is set between the multi-scale input layer and the multi-scale feature extraction layer of the multi-scale feature fusion network to enable the network to pay more attention to important areas in the image. This layer will generate an attention weight map, which will affect the processing of image features by subsequent feature extraction layers;

[0082] S2-2-2: An attention weighted fusion layer is set between the multi-scale feature extraction layer and the multi-scale feature output layer of the multi-scale feature fusion network. The weighted fusion features will be more representative, which will help improve the accuracy of segmentation and ensure that the network can utilize multi-scale information, which is particularly important for processing lesions with different sizes and shapes.

[0083] S2-2-3: After the multi-scale feature output layer, an attention adjustment layer is set to obtain the initial medical image data segmentation model;

[0084] Introducing an attention mechanism into the network enables the network to automatically learn and emphasize key areas in medical imaging data, improving segmentation accuracy and efficiency;

[0085] S2-3: Using a number of image-processed historical medical image data, optimizing and training the initial medical image data segmentation model to obtain an optimized medical image data segmentation model, and generating a number of historical medical image data segmentation results, including the following steps:

[0086] S2-3-1: Divide the processed historical medical imaging data into a model training set and a model test set in a ratio of 7:3;

[0087] S2-3-2: Use the model training set to optimize and train the initial medical image data segmentation model to obtain a trained medical image data segmentation model and generate several historical medical image data segmentation results;

[0088] S2-3-3: Use the model test set to test the trained medical image data segmentation model and obtain several predicted labels;

[0089] S2-3-4: Compare and statistically analyze the predicted labels with the real labels of the historical medical imaging data after image processing in the model test set to obtain the model test accuracy;

[0090] S2-3-5: If the model test accuracy is greater than the accuracy threshold, the optimized medical image data segmentation model is output; otherwise, the optimization training continues;

[0091] The medical image data segmentation result includes a plurality of segmented images obtained by segmenting the medical image data;

[0092] S2-4: Based on several historical medical image data segmentation results, the optimized medical image data segmentation model is analyzed for segmentation effect, and a real-time segmentation effect analysis result is obtained, including the following steps:

[0093] S2-4-1: extracting historical feature vectors of a number of historical medical image data segmentation results, and performing standardization processing on the historical feature vectors to obtain a number of standardized historical feature vectors;

[0094] Feature vectors include texture, shape, and edge information of medical imaging data;

[0095] S2-4-2: Construct a feature space based on several standardized historical feature vectors and use the k-Nearest Neighbors (k-NN) algorithm to obtain the k neighbors of each historical feature vector in the feature space;

[0096] The feature vectors of all segmentation results are constructed into a feature space, where each feature vector corresponds to a point in the feature space. The construction of the feature space helps to analyze the relationship between feature vectors in a multidimensional space.

[0097] S2-4-3: Perform segmentation effect analysis based on the k neighbors of each historical feature vector to obtain the real-time segmentation effect analysis results of the optimized medical image data segmentation model, including the following steps:

[0098] S2-4-3-1: Obtain the Euclidean distance between each historical feature vector and its k neighbors, and obtain the segmentation effect consistency score based on the average distance of several Euclidean distances;

[0099] The average distance is used to characterize the similarity between the segmented image of historical medical imaging data and its neighbors. It can evaluate the consistency of the segmented image. If the similarity between neighbors is high, it means that the segmentation results are relatively consistent; otherwise, there may be uncertainty.

[0100] S2-4-3-2: Get the uncertainty score of the segmentation effect based on the standard deviation of several Euclidean distances;

[0101] The distance standard deviation is used to characterize the neighbor distribution of the segmented image of historical medical imaging data. By analyzing the neighbor distribution of the segmented image in the feature space, the uncertainty of the segmentation result can be quantified;

[0102] S2-4-3-3: Based on the segmentation effect consistency score and the segmentation effect uncertainty score, obtain the real-time segmentation effect analysis results of the optimized medical image data segmentation model;

[0103] In this embodiment, the segmentation effect consistency score is less than the consistency score threshold, and the real-time segmentation effect analysis result includes poor segmentation result consistency; the segmentation effect uncertainty score is less than the uncertainty score threshold, and the real-time segmentation effect analysis result includes high segmentation result uncertainty;

[0104] S2-5: Based on the real-time segmentation effect analysis results, a swarm intelligence optimization algorithm is used to optimize the model parameters of the optimized medical image data segmentation model to obtain the final medical image data segmentation model, including the following steps:

[0105] S2-5-1: Based on the real-time segmentation effect analysis results, define the optimization objectives of the swarm intelligence optimization algorithm and encode the model parameters of the optimized medical image data segmentation model into individual vectors of the swarm intelligence optimization algorithm;

[0106] In this embodiment, the real-time segmentation effect analysis result is that the segmentation result consistency is poor and the segmentation result uncertainty is high, so the optimization goal is to maximize consistency and minimize uncertainty;

[0107] The swarm intelligence optimization algorithm is the Improved Sparrow Search Algorithm (ISSA). The ISSA algorithm avoids the drawback of traditional swarm intelligence optimization algorithms, which are prone to falling into local optimal values ​​and being unable to escape, thereby improving the efficiency of model training and the accuracy of model prediction.

[0108] S2-5-2: Based on the optimization objective and individual vector, a swarm intelligence optimization algorithm is used to iteratively search for the optimal individual in the search space, including the following steps:

[0109] S2-5-2-1: Set the algorithm parameters and maximum number of iterations of the ISSA algorithm, and set the fitness function according to the optimization goal;

[0110] The algorithm parameters of the ISSA algorithm include the search space dimension, determine the search space food and the sparrow position is ;in, is the search space food matrix, They are all elements of the search space food matrix, is the sparrow position matrix, are all elements of the sparrow position matrix, is the number of ISSA individuals, Optimize the dimensions of the problem for your model; h is the total number of sparrows;

[0111] The formula is:

[0112]

[0113] Where, ISSA individual The fitness value of ISSA individual uncertainty function; ISSA individual The consistency function of All are weight coefficients; is the smallest real number that is not 0; c is the ISSA individual indicator;

[0114] S2-5-2-2: Initialize using the Circle chaotic map sequence according to the algorithm parameters and individual vectors to obtain an initial ISSA population consisting of several initial ISSA individuals;

[0115] The formula is:

[0116]

[0117] Where, is the initial ISSA individual of the Circle chaotic map; is the randomly generated initial ISSA individual;

[0118] S2-5-2-3: Use the fitness function to obtain the initial fitness values ​​of all initial ISSA populations, and sort the initial ISSA individuals according to the initial fitness values ​​to obtain the initial discoverers, initial joiners, and initial predators;

[0119] S2-5-2-4: Update the initial ISSA population to obtain an updated ISSA population; the updated ISSA population includes updated discoverers, updated joiners, and updated predators;

[0120] The update formula of the discoverer is:

[0121]

[0122] Where, Respectively t +1, t The first iteration c ISSA individuals who discovered is the maximum number of iterations; A random number between 0 and 1; is a normally distributed random number; for A matrix whose elements are all 1; is the warning value; is the safety threshold;

[0123] The update formula for the joiner is:

[0124]

[0125] Where, Respectively t +1, t The first iteration c ISSA individuals who join; The best position for those who reveal their identities; is the current worst position; is the maximum iteration threshold; A random number between 0 and 1; for A matrix whose elements are all 1 or -1; Update parameters for position;

[0126] The update formula of the predator is:

[0127]

[0128] Where, Respectively t +1, t The first iteration c predator ISSA individuals; is the step size control parameter, and , is the convergence factor, A positive real number that controls the step size to be non-zero; is the current best position; are the current, best and worst fitness of the ISSA individual respectively; is the minimum constant to prevent the denominator from being 0;

[0129]

[0130] Where, is the convergence factor; tanh(.) is the hyperbolic tangent function; is the iteration indicator; is the maximum number of iterations; a max 、 a min are the maximum and minimum values ​​of the convergence factor respectively; λ is the deceleration rate parameter, is the decrement period parameter, λ =-2 π , =π ;

[0131] S2-5-2-5: Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the updated ISSA population to generate a dynamic reverse ISSA population;

[0132] The formula is:

[0133]

[0134] Where, is a dynamically reversed ISSA individual; is the decreasing inertia coefficient; is the upper limit of the search space; is the lower limit of the search space; For the updated ISSA individual;

[0135] S2-5-2-6: Use the fitness function to obtain the updated fitness values ​​of all ISSA individuals in the updated ISSA population and the dynamically reversed ISSA population, and obtain the optimal individual based on the updated fitness values;

[0136] S2-5-2-7: If the current number of iterations is greater than the maximum iteration test or the fitness value of the optimal individual is less than the fitness threshold, the optimal individual is output;

[0137] S2-5-3: Decode the individual vector of the optimal individual to obtain the optimal model parameters of the optimized medical image data segmentation model;

[0138] S2-5-4: Optimizing the optimized medical image data segmentation model according to the optimal model parameters to obtain a final medical image data segmentation model;

[0139] S3: collecting real-time medical image data and using a medical image data segmentation model to perform medical image data segmentation on the real-time medical image data to obtain a real-time medical image data segmentation result, including the following steps:

[0140] S3-1: collecting real-time medical image data and performing preprocessing to obtain preprocessed real-time medical image data, and inputting the preprocessed real-time medical image data into the medical image data segmentation model;

[0141] S3-2: Use a multi-scale input layer to perform multi-scale sampling on the pre-processed real-time medical imaging data to obtain several real-time feature maps at different scales;

[0142] S3-3: Use the attention generation layer to generate corresponding real-time attention weight values ​​based on several real-time feature maps;

[0143] S3-4: Use the multi-scale feature extraction layer to extract the real-time image features of each real-time feature map;

[0144] S3-5: Use the attention weighted fusion layer to perform weighted fusion on several real-time image features according to the real-time attention weight value to obtain real-time weighted fusion features;

[0145] S3-6: Use the multi-scale feature output layer to output the initial real-time medical image data segmentation results based on the real-time weighted fusion features;

[0146] S3-7: Use the attention adjustment layer to fine-tune the initial real-time medical image data segmentation result according to the real-time attention weight value to obtain the final real-time medical image data segmentation result.

[0147] Example 2:

[0148] like Figure 2 As shown, this embodiment provides a medical image data segmentation system for implementing a segmentation method. The system includes an image processing unit, a model building unit, and a data segmentation unit connected in sequence.

[0149] An image processing unit, configured to collect a plurality of historical medical image data and perform image processing on the plurality of historical medical image data to obtain a plurality of image-processed historical medical image data with real labels;

[0150] A model building unit, configured to build a medical image data segmentation model based on a plurality of image-processed historical medical image data using an attention multi-scale fusion deep learning algorithm;

[0151] The data segmentation unit is used to collect real-time medical image data and use a medical image data segmentation model to perform medical image data segmentation on the real-time medical image data to obtain a real-time medical image data segmentation result.

[0152] The present invention provides a medical image data segmentation method and system. By performing image processing on historical medical image data and setting real labels, high-quality training data can be provided for the segmentation model, thereby improving the segmentation accuracy of the segmentation model on real-time medical image data, accurately identifying and segmenting subtle structures in complex medical images, and ensuring that the segmentation results meet the high-precision requirements of clinical diagnosis. The model constructed using the attention multi-scale fusion deep learning algorithm can better capture key features in medical images and improve the model's generalization ability for medical image data of different types and sources, without the need to retrain or adjust the model for specific situations. The medical image data segmentation model is automatically constructed, reducing the need for manual intervention and parameter adjustment, realizing the automation and intelligence of the segmentation process, reducing errors caused by inconsistent human judgment, and improving the reliability of medical image segmentation.

[0153] The present invention is not limited to the above optional embodiments. Anyone can derive various other forms of products based on the teachings of the present invention. The above specific embodiments should not be construed as limiting the scope of protection of the present invention. The scope of protection of the present invention shall be based on the scope defined in the claims, and the description can be used to interpret the claims.

Claims

1. A method for segmenting medical image data, characterized by: The steps include: Collecting a number of historical medical image data, and performing image processing on the number of historical medical image data to obtain a number of image-processed historical medical image data with real labels; Based on several historical medical imaging data after image processing, a medical imaging data segmentation model is constructed using the attention multi-scale fusion deep learning algorithm, including the following steps: Use the CNN algorithm to build a segmentation network, and introduce a multi-scale feature fusion mechanism to improve the segmentation network to obtain a multi-scale feature fusion network; The multi-scale feature fusion network includes a multi-scale input layer, a multi-scale feature extraction layer and a multi-scale feature output layer connected in sequence; By introducing the attention mechanism and improving the multi-scale feature fusion network, we can obtain the initial medical image data segmentation model, which includes the following steps: An attention generation layer is set between the multi-scale input layer and the multi-scale feature extraction layer of the multi-scale feature fusion network; An attention weighted fusion layer is set between the multi-scale feature extraction layer and the multi-scale feature output layer of the multi-scale feature fusion network; After the multi-scale feature output layer, an attention adjustment layer is set to obtain the initial medical image data segmentation model; Using a number of image-processed historical medical image data, optimizing and training an initial medical image data segmentation model to obtain an optimized medical image data segmentation model, and generating a number of historical medical image data segmentation results; Based on several historical medical image data segmentation results, the optimized medical image data segmentation model is analyzed for segmentation effect, and real-time segmentation effect analysis results are obtained, including the following steps: Extracting historical feature vectors of a number of historical medical image data segmentation results, and performing standardization processing on the number of historical feature vectors to obtain a number of standardized historical feature vectors; Based on several standardized historical feature vectors, a feature space is constructed, and the k-NN algorithm is used to obtain the k neighbors of each historical feature vector in the feature space; Based on the k neighbors of each historical feature vector, segmentation effect analysis is performed to obtain the real-time segmentation effect analysis results of the optimized medical image data segmentation model, including the following steps: Obtain the Euclidean distance between each historical feature vector and its k neighbors, and obtain the segmentation effect consistency score based on the average distance of several Euclidean distances; According to the standard deviation of several Euclidean distances, the uncertainty score of the segmentation effect is obtained; According to the segmentation effect consistency score and segmentation effect uncertainty score, the real-time segmentation effect analysis results of the optimized medical image data segmentation model are obtained; Based on the real-time segmentation effect analysis results, the swarm intelligence optimization algorithm is used to optimize the model parameters of the optimized medical image data segmentation model to obtain the final medical image data segmentation model, including the following steps: Based on the real-time segmentation effect analysis results, the optimization objectives of the swarm intelligence optimization algorithm are defined, and the model parameters of the optimized medical image data segmentation model are encoded into individual vectors of the swarm intelligence optimization algorithm; According to the optimization objective and individual vector, a swarm intelligence optimization algorithm is used to iteratively search for the optimal individual in the search space, including the following steps: Set the algorithm parameters and maximum number of iterations of the ISSA algorithm, and set the fitness function according to the optimization goal; The algorithm parameters of the ISSA algorithm include the search space being N×D dimensions, and determining the search space food as F=[F1,F2,...,F D ] T The sparrow position is X=[X h1 ,X h2 ,...,X hD ] T ; Where F is the search space food matrix, F1, F2, ..., F D are all search space food matrix elements, X is the sparrow position matrix, X h1 ,X h2 ,...,X hD are all elements of the sparrow position matrix, N is the number of ISSA individuals, D is the dimension of the model optimization problem; h is the total number of sparrows; The formula is: Where, f(X c ) is the ISSA individual X c The fitness value of A(X c ) is the ISSA individual X c The uncertainty function of T(X c ) is the ISSA individual X c The consistency function of ; α and β are weight coefficients; is the smallest real number that is not 0; c is the ISSA individual indicator; According to the algorithm parameters and individual vectors, the Circle chaotic map sequence is used for initialization to obtain an initial ISSA population including several initial ISSA individuals. The formula is: Where, X' c is the initial ISSA individual of Circle chaotic mapping; X c * is the randomly generated initial ISSA individual; Using the fitness function, the initial fitness values ​​of all initial ISSA populations are obtained, and according to the initial fitness values, the initial ISSA individuals are sorted to obtain the initial discoverers, initial joiners, and initial predators; The initial ISSA population is updated to obtain an updated ISSA population; the updated ISSA population includes updated discoverers, updated joiners and updated predators; The update formula of the discoverer is: Where, are the cth discoverer ISSA individuals of the t+1th and tth iterations respectively; iter max is the maximum number of iterations; ξ is a random number between 0 and 1; Q is a normally distributed random number; L is a 1×D matrix whose elements are all 1; R2 is the warning value; ST is the safety threshold; The update formula for the joiner is: Where, are the cth joiner ISSA individuals in the t+1th and tth iterations respectively; The best position for those who reveal their identities; is the current worst position; iter max is the maximum iteration threshold; ξ is a random number between 0 and 1; L is a 1×D matrix whose elements are all 1 or -1; A + Update parameters for position; The update formula of the predator is: Where, are the c-th predator ISSA individuals in the t+1th and tth iterations respectively; δ is the step-size control parameter, and δ=a"·γ", where a" is the convergence factor and γ" is a non-zero positive real number for step-size control; is the current best position; f c 、f g 、f w are the current, best and worst fitness of the ISSA individual respectively; γ is the minimum constant to prevent the denominator from being 0; Where a" is the convergence factor; tanh(.) is the hyperbolic tangent function; t is the iteration indicator; t max is the maximum number of iterations; a max 、a min are the maximum and minimum values ​​of the convergence factor respectively; λ is the decreasing rate parameter, and k" is the decreasing period parameter; Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the updated ISSA population to generate a dynamic reverse ISSA population; The formula is: Where, is the ISSA individual with dynamic reverse motion; γ* is the decreasing inertia coefficient; ub is the upper limit of the search space; lb is the lower limit of the search space; For the updated ISSA individual; Using the fitness function, obtain the updated fitness values ​​of all ISSA individuals in the updated ISSA population and the dynamically reversed ISSA population, and obtain the optimal individual based on the updated fitness values; If the current number of iterations is greater than the maximum number of iterations or the fitness value of the optimal individual is less than the fitness threshold, the optimal individual is output; Decoding the individual vector of the optimal individual to obtain the optimal model parameters of the optimized medical image data segmentation model; According to the optimal model parameters, the optimized medical image data segmentation model is optimized to obtain the final medical image data segmentation model; Real-time medical image data is collected, and a medical image data segmentation model is used to perform medical image data segmentation on the real-time medical image data to obtain a real-time medical image data segmentation result.

2. The medical image data segmentation method according to claim 1, characterized in that: Collecting a number of historical medical image data and performing image processing on the number of historical medical image data to obtain a number of image-processed historical medical image data with real labels includes the following steps: Collecting a number of historical medical imaging data and corresponding data information, and performing data cleaning on the number of historical medical imaging data to obtain a number of cleaned historical medical imaging data; performing normalization processing on a number of cleaned historical medical imaging data to obtain a number of normalized historical medical imaging data; Performing Gaussian denoising on a number of normalized historical medical image data to obtain a number of denoised historical medical image data; Performing image enhancement on a number of denoised historical medical image data to obtain a number of enhanced historical medical image data; According to the data information of the enhanced historical medical image data, a true label is set for each enhanced historical medical image data to obtain a plurality of image-processed historical medical image data set with the true label.

3. The medical image data segmentation method according to claim 2, characterized in that: Using a number of image-processed historical medical image data, optimizing and training the initial medical image data segmentation model to obtain an optimized medical image data segmentation model, and generating a number of historical medical image data segmentation results, including the following steps: The processed historical medical imaging data were divided into a model training set and a model test set in a ratio of 7:3; Using the model training set, the initial medical image data segmentation model is optimized and trained to obtain a trained medical image data segmentation model, and several historical medical image data segmentation results are generated; Use the model test set to test the trained medical image data segmentation model and obtain several predicted labels; The model test accuracy is obtained by comparing the predicted labels with the real labels of the historical medical imaging data after image processing in the model test set; If the model test accuracy is greater than the accuracy threshold, the optimized medical image data segmentation model is output; otherwise, the optimization training continues.

4. The medical image data segmentation method according to claim 3, characterized in that: Collecting real-time medical image data and using a medical image data segmentation model to perform medical image data segmentation on the real-time medical image data to obtain a real-time medical image data segmentation result includes the following steps: Collecting real-time medical image data and performing preprocessing to obtain preprocessed real-time medical image data, and inputting the preprocessed real-time medical image data into a medical image data segmentation model; Using a multi-scale input layer, the pre-processed real-time medical imaging data is sampled at multiple scales to obtain several real-time feature maps at different scales. Use the attention generation layer to generate corresponding real-time attention weight values ​​based on several real-time feature maps; Use a multi-scale feature extraction layer to extract real-time image features of each real-time feature map; Use the attention weighted fusion layer to perform weighted fusion on several real-time image features according to the real-time attention weight value to obtain real-time weighted fusion features; Use a multi-scale feature output layer to output the initial real-time medical image data segmentation results based on real-time weighted fusion features; Using the attention adjustment layer, the initial real-time medical image data segmentation result is fine-tuned according to the real-time attention weight value to obtain the final real-time medical image data segmentation result.

5. A medical image data segmentation system, for implementing the segmentation method according to any one of claims 1 to 4, characterized in that: The system includes an image processing unit, a model building unit and a data segmentation unit which are connected in sequence.

Images