Tumor recognition method based on image processing and related device

Abstract

This application relates to the field of data processing technology, and provides a tumor identification method and related apparatus based on image processing. The method includes: acquiring a set of medical image data, a set of pathological image data, and a set of clinical diagnosis and treatment data of a target object; performing basic attribute extraction processing on each medical image data in the medical image data set to obtain a set of tumor attribute feature data of the target object; performing tumor texture feature recognition based on the set of tumor attribute feature data and a set of sample label data to obtain tumor texture feature data of the target object; determining tumor invasion feature data of the target object based on the set of tumor attribute feature data, tumor texture feature data, medical image data, pathological image data, and sample label data; and identifying tumor lesions based on the set of tumor attribute feature data, tumor texture feature data, tumor invasion feature data, and clinical diagnosis and treatment data to obtain the target tumor identification result.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a tumor recognition method and related apparatus based on image processing. Background Technology

[0002] Current tumor identification methods suffer from numerous shortcomings. For example, in terms of detection technology, routine physical examinations lack targeted screening methods for some early-stage, occult, or specific types of tumors. Furthermore, imaging examinations are easily affected by factors such as tumor location and size, leading to missed diagnoses. The specificity and sensitivity of tumor marker detection are also insufficient. Regarding sample collection, there are problems such as insufficient sample size, poor sample quality, or improper sampling locations. Pathological examinations also largely rely on expert experience, making them susceptible to subjective factors. Moreover, tumor heterogeneity means that the same tumor may exist in multiple types or subtypes, further complicating the judgment. Therefore, how to achieve more accurate and comprehensive tumor identification based on existing multifaceted data has become an urgent problem to be solved. Summary of the Invention

[0003] This application provides a tumor identification method and related apparatus based on image processing, which can perform more accurate and comprehensive tumor identification based on existing multi-faceted data, thereby improving the accuracy of the tumor identification process.

[0004] A first aspect of this application provides a tumor recognition method based on image processing, the method comprising: Acquire the medical imaging data set, pathological image data set, and clinical diagnosis and treatment data set of the target object; Basic attribute extraction processing is performed on each medical image data in the medical image dataset to obtain the tumor attribute feature dataset of the target object. Tumor texture feature recognition is performed based on tumor attribute feature dataset and sample label dataset to obtain tumor texture feature data of the target object; Based on the tumor attribute feature data set, tumor texture feature data, medical image data set, pathological image data set, and sample label data set, the tumor invasion feature data of the target object is determined; Based on the tumor attribute feature data set, tumor texture feature data, tumor invasion feature data, and clinical diagnosis and treatment data set, the tumor lesion identification is performed to obtain the target tumor identification result.

[0005] In this example, by acquiring a set of medical image data, a set of pathological image data, and a set of clinical diagnosis and treatment data for the target object, and performing basic attribute extraction processing on each medical image data in the set of medical image data, a set of tumor attribute feature data for the target object is obtained. Furthermore, tumor texture feature recognition can be performed based on the set of tumor attribute feature data and the set of sample label data to obtain the tumor texture feature data of the target object. Therefore, based on the set of tumor attribute feature data, the tumor texture feature data, the set of medical image data, the set of pathological image data, and the set of sample label data, the tumor invasion feature data of the target object can be determined. Subsequently, based on the set of tumor attribute feature data, the tumor texture feature data, the tumor invasion feature data, and the set of clinical diagnosis and treatment data, a more accurate target tumor identification result can be obtained, thus improving the accuracy of the tumor identification process.

[0006] A second aspect of this application provides a tumor recognition device based on image processing, the device comprising: The acquisition unit is used to acquire the medical image data set, pathological image data set, and clinical diagnosis and treatment data set of the target object; The first processing unit is used to perform basic attribute extraction processing on each medical image data in the medical image data set to obtain the tumor attribute feature data set of the target object. The second processing unit is used to perform tumor texture feature recognition based on the tumor attribute feature data set and the sample label data set to obtain tumor texture feature data of the target object. The first determining unit is used to determine the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set; The identification unit is used to identify tumor lesions and obtain the target tumor identification result based on the tumor attribute feature data set, the tumor texture feature data, the tumor invasion feature data and the clinical diagnosis and treatment data set.

[0007] A third aspect of this application provides a terminal including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute the step instructions as described in the first aspect of this application.

[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the first aspect of this application.

[0009] A fifth aspect of this application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the first aspect of this application. The computer program product may be a software installation package. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This application provides a schematic diagram of the structure of an image processing-based tumor recognition system as an embodiment; Figure 2 This application provides a flowchart illustrating a tumor recognition method based on image processing. Figure 3 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application; Figure 4 This application provides a schematic diagram of the structure of a tumor recognition device based on image processing. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0014] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0015] To better understand the tumor identification method based on image processing provided in this application, a brief introduction to the application scenarios of this method is given below. Currently, an increasing number of cancer patients are emerging, including those in the early stages of their illness and those undergoing long-term cancer treatment requiring regular checkups. Faced with this large patient population and the increasing variety of tumor types, routine physical examinations lack targeted screening methods for some early, occult, or specific types of tumors. Furthermore, imaging examinations are easily affected by factors such as tumor location and size, leading to missed diagnoses. Moreover, tumor marker detection lacks specificity and sensitivity, and sample collection suffers from insufficient sample size, poor quality, or improper sampling locations. Therefore, current tumor identification accuracy is low. This application aims to solve the problem of low accuracy in tumor identification by providing an image processing-based tumor identification method. This method can determine tumor attribute feature data, tumor texture feature data, and tumor invasion feature data through a set of medical imaging data, pathological image data, and clinical diagnosis and treatment data of the target object. Further combining this with the clinical diagnosis and treatment data for tumor lesion identification yields more accurate target tumor identification results, improving the accuracy of tumor identification.

[0016] Image processing-based tumor identification methods can be applied to image processing-based tumor identification systems. Figure 1 A schematic diagram of an image processing-based tumor recognition system is shown. Figure 1As shown, this image processing-based tumor recognition system may include, but is not limited to, a data acquisition module, a basic attribute extraction module, a target generative adversarial image processing module, an invasion feature analysis module (including a target invasion analysis image processing module), and a tumor lesion identification module. Specifically, the data acquisition module can include a set of medical image data, a set of pathological image data, and a set of clinical diagnosis and treatment data for the target object; the basic attribute extraction module is used to perform basic attribute extraction processing on each medical image data in the medical image data set, such as extracting the morphological features, density features, and grayscale features of the tumor; the target generative adversarial image processing module is used to perform tumor texture feature recognition based on the tumor attribute feature data set and the sample label data set to obtain the tumor texture feature data of the target object; the invasion feature analysis module is used to determine the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set; and the tumor lesion identification module is used to perform tumor lesion identification based on the tumor attribute feature data set, the tumor texture feature data, the tumor invasion feature data, and the clinical diagnosis and treatment data set to obtain a more accurate target tumor identification result.

[0017] Please see Figure 2 , Figure 2 This application provides a flowchart illustrating an image processing-based tumor recognition method, which includes: S10: Obtain the medical image data set, pathological image data set, and clinical diagnosis and treatment data set of the target object.

[0018] The target object can be understood as one or more patients who need to identify and recognize tumor lesions using the image processing-based tumor identification method provided in this application; this application does not limit this. The medical image dataset may include one or more medical image datasets, which can be used to indicate image data acquired through various medical imaging techniques, such as computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET).

[0019] The pathological image dataset may include one or more pathological image datasets, which can be used to indicate the image data obtained by taking pictures with equipment such as microscopes after pathological sections of diseased tissue are prepared. The clinical diagnosis and treatment dataset may include one or more clinical diagnosis and treatment datasets, which can be used to indicate various data generated during the clinical diagnosis and treatment of the target subject, such as the patient's medical history, symptom description, laboratory test results (such as blood tests, tumor marker detection, etc.), treatment plan, and related records during the treatment process, etc., without any limitations on these.

[0020] S20: Perform basic attribute extraction processing on each medical image data in the medical image data set to obtain the tumor attribute feature data set of the target object.

[0021] The tumor attribute feature dataset may include one or more tumor attribute feature data, which can be used to indicate the tumor attribute-related feature data obtained after basic attribute extraction from medical image data. Optionally, the tumor attribute feature dataset may include, but is not limited to, tumor morphological feature data, tumor density feature data, and tumor grayscale feature data, etc., and this application does not impose any limitations on this.

[0022] Specifically, in two-dimensional medical images (such as CT slices), the size of a tumor can be approximated by calculating the number of pixels occupied by the tumor region, and further converted by combining the pixel spacing information of the image to obtain the actual size. In three-dimensional images, the volume of the tumor can be calculated using voxel counting. The characteristics of the tumor boundary can be described by calculating indicators such as boundary roughness and fractal dimension; among them, roughness can be measured by calculating the gray value changes of boundary pixels, and fractal dimension can be used to describe the complexity of the boundary, which is not limited in this application.

[0023] In medical imaging, each pixel or voxel has a corresponding grayscale value or density value. The average density of the tumor can be obtained by calculating the average density value of all pixels or voxels within the tumor region. Optionally, the distribution of density within the tumor can be described by calculating statistical measures such as the standard deviation, skewness, and kurtosis of the density. Among these, the standard deviation reflects the dispersion of density, the skewness describes the asymmetry of the density distribution, and the kurtosis indicates the sharpness of the density distribution. This application does not impose any limitations on these measures.

[0024] S30: Based on the tumor attribute feature data set and the sample label data set, tumor texture feature recognition is performed to obtain the tumor texture feature data of the target object.

[0025] Tumor texture feature data can be used to describe the texture characteristics of tumor tissue. The sample label dataset can include one or more sample label datasets, which can be used to indicate data containing accurate annotation information, such as medical image data annotating the specific location of the tumor, and / or pathological image data annotating the specific stage of the tumor. This sample label data can be used for supervised training of image processing modules (such as the initial generative adversarial image processing module and the initial infiltration analysis image processing module, which will be mentioned below), to help the modules learn correct features and / or classification relationships, thereby improving the accuracy of the tumor lesion identification process.

[0026] Optionally, a target generative adversarial image processing module can be trained based on a tumor attribute feature dataset and a sample label dataset. This module can then be used to perform tumor texture feature recognition on each tumor attribute feature in the tumor attribute feature dataset, thereby obtaining tumor texture feature data. This target generative adversarial image processing module can be a logical module capable of implementing image processing functions related to Generative Adversarial Networks (GANs), and this application does not impose any limitations on it.

[0027] Optionally, the tumor attribute feature dataset can be divided into a training set and a test set, so as to train the initial generative adversarial image processing module using the training set and the sample label dataset, and to test the trained generative adversarial image processing module using the test set. This application does not impose any restrictions on this.

[0028] This application uses a target-generative adversarial image processing module trained on a target to perform tumor texture feature recognition as an example, and does not constitute a limitation on this application. The target-generative adversarial image processing module can effectively mine the complex and potential relationship between tumor attribute features and texture through an adversarial learning mechanism between the generator and discriminator. It can generate a rich variety of texture samples to compensate for the shortcomings of real data and enhance the model's learning and understanding of tumor texture features, thereby improving the accuracy of recognition. Furthermore, its powerful generative capabilities help to discover subtle texture features that are difficult to capture by traditional methods, providing a more comprehensive perspective for tumor texture recognition. This can more accurately assist in the judgment of tumor conditions, thereby improving diagnostic efficiency and quality.

[0029] It is important to understand that, in order to obtain the tumor texture feature data of the target object by performing tumor texture feature recognition based on the tumor attribute feature data set and the sample label data set, the process refers to how a trained target generative adversarial image processing module performs tumor texture feature recognition on each tumor attribute feature data in the tumor attribute feature data set, thereby obtaining the tumor texture feature data. Specifically, step S30, which involves performing tumor texture feature recognition based on the tumor attribute feature data set and the sample label data set to obtain the tumor texture feature data of the target object, may include the following steps: S31: The generator of the initial generative adversarial image processing module generates tumor texture features based on each sample label data in the sample label data set, and obtains the generated texture feature data set; S32: Using the discriminator of the initial generative adversarial image processing module, perform texture feature discrimination processing based on each generated texture feature data in the generated texture feature data set to obtain a discriminative texture feature data set; S33: Determine the texture feature loss function based on the discriminative texture feature data set, the generated texture feature data set, and the sample label data set; S34: Train the initial generative adversarial image processing module according to the texture feature loss function to obtain the target generative adversarial image processing module; S35: Using the target generative adversarial image processing module, tumor texture feature recognition is performed on each tumor attribute feature data in the tumor attribute feature data set to obtain the tumor texture feature data of the target object.

[0030] The initial generative adversarial image processing module can be understood as an initial image processing module built based on GAN. The parameters of this initial generative adversarial image processing module are randomly initialized and need to be adjusted and optimized through subsequent training processes to obtain a target generative adversarial image processing module that can accurately identify tumor texture features.

[0031] The generative adversarial image processing module (such as the initial generative adversarial image processing module mentioned above and the target generative adversarial image processing module to be mentioned below) can contain two core parts: a generator and a discriminator. The generator can be used to continuously learn and simulate the distribution of real data in order to generate increasingly realistic generated data, such as the texture features of tumors; the discriminator can be used to distinguish between real data and generated data as accurately as possible.

[0032] The generated texture feature dataset can include one or more generated texture feature datasets. These datasets can be used to indicate the texture feature data of the tumor generated by the generator of the initial generative adversarial image processing module based on the sample label data. The generated texture feature data is the result of the generator simulating the texture features of a real tumor. In the early stages of training, it may differ significantly from the real tumor texture features, but it can gradually approach the real tumor texture as training progresses.

[0033] It is important to understand that the process of generating tumor texture features based on each sample label data in the sample label dataset using the generator of the initial generative adversarial image processing module, resulting in a generated texture feature dataset, refers to the process by which the generator of the initial generative adversarial image processing module generates the texture feature dataset based on the tumor attribute feature dataset. Specifically, step S31, which involves generating tumor texture features based on each sample label data in the sample label dataset using the generator of the initial generative adversarial image processing module, can include the following steps: S311: The generator of the initial generative adversarial image processing module encodes each sample label data in the sample label data set with a noise vector to obtain a tumor noise feature data set. S312: Using the generator of the initial generative adversarial image processing module, feature mapping processing is performed on each tumor noise feature data in the tumor noise feature data set to obtain an abstract feature data set; S313: Using the generator of the initial generative adversarial image processing module, perform feature reshaping processing on each abstract feature data in the abstract feature data set to obtain a reshaped feature data set; S314: Using the generator of the initial generative adversarial image processing module, upsample each reshaped feature data in the reshaped feature data set to obtain an upsampled feature data set; S315: Using the generator of the initial generative adversarial image processing module, perform nonlinear enhancement processing on each upsampled feature data in the upsampled feature data set to obtain an enhanced feature data set; S316: Using the generator of the initial generative adversarial image processing module, normalize each enhanced feature data in the enhanced feature data set to obtain a generated texture feature data set.

[0034] In this context, the noise vector can be understood as a vector obtained by randomly sampling from distributed data (such as normally distributed data or uniformly distributed data). Introducing a noise vector into a generative adversarial network (GAN) adds randomness and diversity to the generation process, enabling the generator to produce tumor texture features of different styles and avoiding the uniformity of the generated results. Encoding the sample label data with the noise vector can be understood as an integration and transformation operation, which can fuse data of different types and dimensions to obtain a tumor noise feature dataset, thus preparing for the subsequent generation of diverse tumor texture features.

[0035] Feature mapping can utilize the layered structure of neural networks (such as fully connected layers and convolutional layers) to map tumor noise feature data to a new abstract feature space. This captures deeper features of the data, thereby extracting more representative and abstract features, resulting in an abstract feature data set. This abstract feature data set can include one or more abstract feature data points, which can be used to indicate the feature data after abstraction and refinement of the tumor noise feature data through feature mapping. This abstract feature data can contain more advanced and complex features, which helps in the subsequent generation of more realistic tumor texture features.

[0036] Feature reshaping can further adjust the dimensions and shape of abstract feature data to obtain a reshaped feature data set. For example, feature reshaping can convert abstract feature data into a suitable format, such as reshaping a one-dimensional abstract feature vector into a two-dimensional or three-dimensional feature map for subsequent processing. The reshaped feature data set can include one or more reshaped feature data, which can be used to indicate the feature data after feature reshaping, based on the abstract feature data, with adjustments to its dimensions and shape.

[0037] Upsampling can be understood as enlarging the spatial size of reconstructed feature data to gradually restore its spatial resolution, resulting in an upsampled feature data set that more closely approximates the size of the actual tumor texture features. The upsampled feature data set can include one or more upsampled feature data points, which can be understood as feature data that has undergone upsampling, resulting in a spatial enlargement of the reconstructed feature data.

[0038] Nonlinear enhancement processing can be understood as the process of processing upsampled feature data using a nonlinear activation function. This nonlinear activation function can be a rectified linear unit (ReLU), a leaky rectified linear unit (LeakyReLU), or a hyperbolic tangent function (Tanh), and this application does not impose any limitation on this. The nonlinear activation function can introduce nonlinear factors into the generative adversarial image processing module, enabling it to learn more complex features and patterns, thus enhancing the expressive power of the data.

[0039] The enhanced feature dataset may include one or more enhanced feature datasets. These enhanced feature datasets can be understood as feature datasets that have undergone nonlinear enhancement processing, resulting in richer feature information and stronger expressive power, and can better reflect the texture features and details of tumors.

[0040] Standardizing the enhanced feature data yields a set of generated texture feature data with a relatively normalized and stable distribution. Specifically, methods such as batch normalization or layer normalization can be used to achieve the normalization process described above. Normalization accelerates the training process of the generative adversarial image processing (GAP) module, improves its stability and generalization ability, and ultimately results in a higher quality and more realistic set of generated texture feature data.

[0041] The discriminative texture feature dataset may include one or more discriminative texture feature data. These discriminative texture feature data can be used to instruct the discriminator of the initial generative adversarial image processing module on the resulting data after processing the generated texture feature data. Specifically, the discriminator assigns a probability to each generated texture feature data, indicating the likelihood that the generated texture feature data is real data.

[0042] Specifically, the discriminator can receive each generated texture feature data from the generated texture feature data set as input, and can preprocess the generated texture feature data before input, such as scaling the value range of the data to a specific interval through normalization; further, it can perform feature extraction processing on the preprocessed data, such as performing feature extraction processing on the preprocessed data through multiple convolutional layers; and it can insert pooling layers (such as max pooling or average pooling) after the convolutional layers to downsample the feature maps after feature extraction, which can reduce the size of the feature maps while retaining important feature information and reducing the amount of computation; it can also use non-linear activation functions. Introducing nonlinear factors enables the discriminator to learn more complex decision boundaries, enhancing the model's expressive power. The discriminator can also fuse features at different levels and scales after the aforementioned multi-layer convolution, pooling operations, and nonlinear transformations to comprehensively utilize these feature information. The fused feature vector can be input into a fully connected layer for further linear transformation and combination of features, thereby mapping the high-dimensional feature vector to a lower-dimensional space. Finally, an activation function can be applied to the output of the fully connected layer to output the probability that the generated texture feature data is real data, thus obtaining the discriminative texture feature data set.

[0043] The texture feature loss function can be understood as a function used to measure the difference between the texture feature data generated by the generator and the real texture feature data. By combining the discriminant texture feature data set, the generated texture feature data set, and the sample label data set, a loss function that is helpful for tumor texture feature recognition can be determined. This provides a specific optimization target for model training, allowing the generator and discriminator to continuously adjust their parameters during adversarial training to reduce the gap between the generated data and the real data.

[0044] Specifically, the texture feature loss function can include a discriminator loss function and a generator loss function. The discriminator loss function can use a binary cross-entropy loss function to enable the discriminator to more accurately distinguish between real and generated images. Optionally, the discriminator loss function can be constructed as shown in the following formula: in, This represents the discriminator loss function; Expressing expectations; This represents the sample label data in the sample label dataset; This indicates the distribution of the sample label data; This indicates that the sample label data was sampled from the sample label data distribution; This indicates an operation that performs logarithmic calculations; This indicates that the input data (i.e., sample label data) is processed by the discriminator. The operation performs a discrimination test, and its output is a probability value between 0 and 1, which can represent the input data (i.e., sample label data). ) represents the probability of the actual data; Represents noise data; This represents the distribution of noise data; This indicates that the noise data was sampled from the noise data distribution. This indicates an operation that generates data based on noisy data using a generator. This indicates an operation that uses a discriminator to discriminate the input data (i.e., the data generated by the generator, such as generated texture feature data).

[0045] It should be noted that, This section deals with the processing of sample label data (i.e., real data). The goal of the discriminator is to classify the input sample label data as real data as accurately as possible. To get as close to 1. Because of the logarithmic function. It is monotonically increasing on the domain (0,1], when When it approaches 1, Approaching 0; when When away from 1, It is a negative number with a large absolute value. Adding a negative sign means that the objective of this part of the loss function is to minimize it, thereby encouraging the discriminator to correctly identify the true data. Get as close to 1 as possible.

[0046] for This part deals with the processing of generated data (i.e., generated texture feature data). The discriminator's goal is to classify the data generated by the generator as accurately as possible. Data that is identified as generated, i.e. To get close to 0. It is close to 1, that is Approaching 0. Similarly, by adding a negative sign, the objective of this part of the loss function is also to minimize, in order to encourage the discriminator to judge the probability of generating data. Get as close to 0 as possible.

[0047] Optionally, the generator loss function can also be based on the binary cross-entropy loss function, so that the generator can fool the discriminator. The generator loss function can be constructed as shown in the following formula: in, Represents the generator loss function; Expressing expectations; Represents noise data; Represents the distribution of noise data; This indicates that the noise data was sampled from the noise data distribution. This indicates an operation that generates data based on noisy data using a generator. This indicates an operation that uses a discriminator to discriminate the input data (i.e., the data generated by the generator, such as generated texture feature data).

[0048] It should be noted that, This refers to the operation of generating based on noise data by a generator, which is the generator mentioned above that uses the initial generative adversarial image processing module to encode each tumor attribute feature data in the tumor attribute feature data set with a noise vector to obtain a tumor noise feature data set, and finally obtain the generated texture feature data set. This application will not elaborate further here.

[0049] The target generative adversarial image processing module can be understood as a generative adversarial image processing module obtained after training. That is to say, the parameters of the generator and discriminator of this target generative adversarial image processing module have been optimized and adjusted, so that it can more accurately generate tumor texture feature data that is close to the real situation based on tumor attribute feature data.

[0050] By employing a trained target generative adversarial image processing module to process each tumor attribute feature data in the tumor attribute feature dataset, tumor texture feature data of the target object can be obtained. This tumor texture feature data can be closer to the tumor texture features in real-world situations, which helps improve the accuracy of subsequent tumor analysis and diagnosis.

[0051] As can be seen from the foregoing, during the initial training of the generative adversarial image processing module, a set of generated texture feature data for the training process (such as the training generated texture feature data set) can be generated. Furthermore, based on this training generated texture feature data set and the discriminator's discriminative texture feature data set, a texture feature damage function that better fits the tumor texture feature recognition can be determined, thereby obtaining a more accurate target generative adversarial image processing module.

[0052] It should be noted that after obtaining the target generative adversarial image processing module, the output of the generator of the target generative adversarial image processing module, i.e., the tumor texture feature data obtained by using the target generative adversarial image processing module to perform tumor texture feature recognition on tumor attribute feature data, can be used as the recognition result of the tumor texture feature recognition process. For details regarding the generation of tumor texture feature data using the generator of the target generative adversarial image processing module during the tumor texture feature recognition process, please refer to the aforementioned detailed description of the generator of the initial generative adversarial image processing module and the generation of the texture feature data set during the training process; this application will not repeat it here.

[0053] S40: Determine the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set.

[0054] Tumor invasion feature data can be used to describe the characteristics of tumor invasion into surrounding tissues. Tumor invasion feature data can be determined by analyzing tumor invasion from multiple dimensions (such as invasion space dimension, heterogeneity dimension, global dimension, and local dimension) based on tumor attribute feature data, tumor texture feature data, medical imaging data, pathological image data, and sample label data. This results in more comprehensive and accurate feature data indicating the extent of tumor invasion.

[0055] It is important to understand that determining the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set refers to the process of judging the invasion status of the tumor and obtaining the tumor invasion feature data. Specifically, step S40, which involves determining the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set, may include the following steps: S41: Perform invasion range analysis on each tumor attribute feature data in the tumor attribute feature data set to obtain invasion space feature data; S42: Perform infiltration difference analysis on the tumor texture feature data to obtain infiltration heterogeneity feature data; S43: Perform global infiltration analysis based on the medical image data set and the sample label data set to obtain global infiltration feature data; S44: Perform local infiltration analysis on each pathological image data in the pathological image dataset to obtain infiltration local feature data; S45: Perform multi-scale fusion processing on the infiltration spatial feature data, the infiltration heterogeneous feature data, the infiltration global feature data, and the infiltration local feature data to obtain the tumor infiltration feature data of the target object.

[0056] The infiltration spatial feature data can be understood as the feature data that describes the tumor infiltration from a spatial dimension after analyzing the infiltration range of the tumor attribute feature data. The infiltration spatial feature data may include, but is not limited to, invasion breadth parameters (which can reflect the size of the tumor's spatial spread), infiltration boundary parameters (which can describe the clarity and morphology of the boundary between the tumor and the surrounding normal tissue), and geometric angles with surrounding tissues (which can reflect the relative positional relationship and invasion direction of the tumor with the surrounding tissues), etc., and this application does not impose any limitations on them.

[0057] Infiltration heterogeneity feature data can be understood as feature data obtained after performing infiltration difference analysis on the tumor texture feature data, which can be used to reflect the differences in texture among different regions within the tumor. This infiltration heterogeneity feature data can reflect the heterogeneity in the tumor invasion process, a characteristic that may be related to factors such as different biological behaviors of tumor cells and differences in the microenvironment.

[0058] Global invasion feature data can be understood as the data on tumor invasion characteristics at the macroscopic level obtained after performing global invasion analysis based on the medical image dataset and the sample label dataset. Optionally, this application uses a target invasion analysis image processing module to achieve global invasion analysis of the tumor, which does not constitute a limitation on this application.

[0059] It is important to understand that performing global infiltration analysis based on the medical image dataset and the sample label dataset to obtain global infiltration feature data refers to the process of conducting infiltration analysis at a macroscopic level to obtain global infiltration feature data. Specifically, step S43, which involves performing global infiltration analysis based on the medical image dataset and the sample label dataset to obtain global infiltration feature data, may include the following steps: S431: Perform comprehensive feature extraction processing on each medical image data in the medical image dataset to obtain a set of medical image feature maps; S432: Perform sliding window processing on each medical image feature map in the medical image feature map set to obtain a medical image anchor frame data set; S433: Classify each medical image anchor frame data in the medical image anchor frame data set to obtain a medical image foreground data set; S434: Perform offset regression calculation on each medical image foreground data in the medical image foreground data set to obtain a medical image regression data set; S435: Perform focused feature extraction processing on each medical image regression data in the medical image regression dataset to obtain a set of infiltration image feature maps; S436: Determine the infiltration feature loss function based on the medical image anchor frame data set, the medical image foreground data set, the infiltration image feature map set, and the sample label data set; S437: The initial infiltration analysis image processing module is trained according to the infiltration feature loss function to obtain the target infiltration analysis image processing module; S438: Using the target infiltration analysis image processing module, infiltration analysis is performed on each medical image data in the medical image dataset to obtain global infiltration feature data.

[0060] The medical image feature map set may include one or more medical image feature maps. These medical image feature maps can be understood as feature maps used to indicate various tumor-related features after comprehensive feature extraction processing of each medical image data. Specifically, features can be extracted using methods such as Convolutional Neural Networks (CNNs) to extract various tumor-related features in the image, such as edge features, texture features, shape features, and density features. This application does not impose any limitations on these features.

[0061] By performing a sliding window operation on each medical image feature map in the medical image feature map set, each medical image feature map can be divided into multiple local regions to obtain medical image anchor box data that can represent the feature information of a local region. The medical image anchor box data set can include one or more medical image anchor boxes. These anchor boxes can be understood as data representing a local region of a specific location and size on the medical image feature map after the sliding window process. The anchor box data can help the model locate and identify target objects in subsequent steps.

[0062] The medical image foreground data set may include one or more medical image foreground data. This medical image foreground data can be understood as data obtained after classifying and processing medical image anchor frame data, used to describe areas where tumors may exist. It is understood that this application refers to areas where tumors may exist as foreground areas and areas where tumors do not exist as background areas, solely for the purpose of clearer region differentiation and description, and does not constitute a limitation on this application.

[0063] The medical image regression dataset may include one or more medical image regression datasets. These datasets can be understood as data obtained by performing offset regression calculations on foreground data of medical images, specifically, data that offsets the tumor location within the foreground data. Specifically, for each foreground data point in the medical image foreground dataset, the offset (such as coordinate offset, size offset, etc.) between it and the actual tumor location can be calculated. Based on this offset, medical image regression data can be obtained to more accurately locate the tumor's position and size. Optionally, a regression model (such as a CNN-based regression model) can be typically used to implement the above offset regression steps; this application does not limit this approach.

[0064] The infiltration image feature map set may include one or more infiltration image feature maps. These infiltration image feature maps can be understood as feature maps containing tumor invasion-related features obtained after focusing feature extraction processing of medical image regression data. Specifically, for each regression data point in the medical image regression data set, key features focusing on tumor invasion, such as invasion boundaries and invasion direction, can be further extracted and represented in the form of feature maps, thereby constituting the aforementioned infiltration image feature map set. This application does not impose any limitations on this.

[0065] The infiltration feature loss function is a function constructed based on the medical image anchor box dataset, medical image foreground dataset, infiltration image feature map dataset, and sample label dataset. It measures the difference between the prediction results of the model (i.e., the infiltration analysis image processing module) and the true labels. Specifically, the infiltration feature loss function can be constructed based on the cross-entropy loss function or the mean squared error loss function, etc., and this application does not impose any restrictions on this.

[0066] The initial invasion analysis image processing module can be understood as an initial model built based on deep learning or other image processing techniques for tumor invasion analysis. It's understood that the parameters of this initial invasion analysis image processing module are randomly initialized before training, and can be optimized through training to improve model performance. The target invasion analysis image processing module can be understood as the invasion analysis image processing module obtained after training. Its parameters have been adjusted according to the invasion feature loss function, enabling more accurate invasion analysis of medical image data and obtaining more reliable global invasion feature data.

[0067] Specifically, this application can construct an initial immersion analysis image processing module based on a Fast Region-based Convolutional Neural Network (Faster R-CNN) model. Optionally, the network layer structure of this initial immersion analysis image processing module may include, but is not limited to, convolutional layers (used for comprehensive feature extraction of the input medical image data to obtain medical image feature maps), Region Proposal Network (RPN) layers (used for sliding window processing of the medical image feature maps to obtain medical image anchor box data), Region of Interest (ROI) pooling layers (used for focused feature extraction of the medical image regression data output by the RPN layers to obtain immersion image feature maps), and Fast R-CNN layers (used for fully connected operations on the pooled features and outputting the final recognition result, i.e., immersion global feature data). The RPN layers can output two branches: a classification branch (used for classification processing to obtain medical image foreground data) and a regression branch (used for offset regression calculation to obtain medical image regression data).

[0068] Optionally, based on the network structure of the initial immersion analysis image processing module described above, the immersion feature loss function may include four parts: RPN classification loss, RPN regression loss, Fast R-CNN classification loss, and Fast R-CNN regression loss. This application does not impose any limitations on this.

[0069] Specifically, the RPN classification loss can be constructed based on the binary cross-entropy loss function to calculate the classification error between the foreground and background in medical image anchor box data. Optionally, the construction process of the RPN classification loss can be found in the following formula: in, Represents the RPN classification loss; Represents medical image anchor frame data; Represents a set of medical image anchor frames; This represents the sample label data in the sample label dataset. This can represent the sample label data as the foreground, i.e., the feature map corresponding to the tumor. It can represent the sample label data as background, that is, the feature map corresponding to other tissues; Represents logarithmic operations; This represents the predicted probability.

[0070] Specifically, the RPN regression loss can be constructed based on the Smooth L1 loss function to calculate the positional offset error of the anchor box in the foreground data of medical images. Optionally, the construction process of the RPN regression loss can be found in the following formula: in, This represents the RPN regression loss; Represents foreground data in medical imaging; This represents a set of foreground data in medical imaging. This indicates that the operation is performed using the Smooth L1 loss function; Indicates the location of the predicted foreground data in medical imaging; This indicates the actual tumor location in the sample label data; Indicates a temporary variable; This represents the absolute value of a temporary variable.

[0071] Specifically, the Fast R-CNN classification loss can be constructed based on binary or multi-class cross-entropy loss functions to measure the difference between the predicted class probability distribution of the medical image foreground data and the true class. Optionally, the construction process of the Fast R-CNN classification loss can be found in the following formula: in, This represents the classification loss of Fast R-CNN; These represent the input parameters of the Fast R-CNN classification loss function. This represents the category probability distribution for predicting foreground data in medical images. Indicates the true category in the sample label data; Represents logarithmic operations; This indicates the probability that the predicted foreground data of the target medical image belongs to the true category u; This indicates the probability that the predicted foreground data of the target medical image belongs to the true category 1; This indicates the probability that the predicted foreground data of the target medical image belongs to the true category 2; This represents the probability that the predicted foreground data of the target medical image belongs to the true category k; This indicates the total number of categories.

[0072] Specifically, the Fast R-CNN regression loss can be constructed based on the Smooth L1 loss function to calculate the positional refinement error of the foreground data in medical images. Optionally, the construction process for the Fast R-CNN regression loss can be found in the following formula: in, This represents the regression loss of Fast R-CNN; These represent the input parameters of the Fast R-CNN regression loss function. The bounding box regression parameters represent the predicted foreground data of the nth medical image. Represents the regression parameters of the true bounding box of the nth medical image foreground data; Represents dimension q; These represent the four dimensions: x-axis scaling, y-axis scaling, width scaling, and height scaling. This indicates that the operation is performed using the Smooth L1 loss function; Represents the regression parameters of the predicted bounding box. The difference between the regression parameters of the true bounding box in dimension q, such as when q=x, This represents the difference between the predicted x-axis offset of the bounding box center and the actual x-axis offset. This indicates the offset of the predicted bounding box center in the x-axis direction relative to the reference box (such as anchor box data); This indicates the offset of the predicted bounding box center relative to the reference box in the y-axis direction; This represents the scaling ratio of the predicted bounding box width relative to the reference box width (usually expressed in logarithmic form). This represents the scaling ratio of the predicted bounding box height relative to the reference box height (usually expressed in logarithmic form). This represents the offset of the actual bounding box center in the x-axis direction relative to the reference box (such as anchor box data); This represents the offset of the actual bounding box center relative to the reference box in the y-axis direction; This represents the scaling ratio of the actual bounding box width relative to the reference box width (usually expressed in logarithmic form). This represents the scaling ratio of the actual bounding box height relative to the reference box height (usually expressed in logarithmic form).

[0073] After determining the RPN classification loss, RPN regression loss, Fast R-CNN classification loss, and Fast R-CNN regression loss, these four losses can be further integrated to obtain the infiltrated feature loss function. Optionally, the construction process of the infiltrated feature loss function can be found in the following formula: in, Represents the infiltration feature loss function; Hyperparameters representing the classification loss of RPN; Represents the RPN classification loss; The hyperparameters representing the RPN regression loss; This represents the RPN regression loss; The hyperparameters representing the classification loss of Fast R-CNN; This represents the classification loss of Fast R-CNN; This represents the hyperparameters of the regression loss in Fast R-CNN; This represents the regression loss of Fast R-CNN.

[0074] Local invasion feature data can be understood as data on the microscopic tumor invasion characteristics obtained by performing local invasion analysis on each pathological image in a pathological image dataset. Since pathological images can provide detailed information at the cellular level, local invasion analysis can focus on the invasion of tumor cells at the microscopic level, such as the invasion mode of tumor cells and their interaction with surrounding cells, thereby achieving microscopic analysis and obtaining local invasion feature data.

[0075] Furthermore, the previously obtained spatial, heterogeneous, global, and local invasion feature data can be fused. Specifically, methods such as feature concatenation and weighted summation can be used to integrate feature information at different scales, ultimately yielding the tumor invasion feature data of the target object. By integrating and fusing invasion feature data obtained at different scales (spatial, microscopic, and macroscopic scales), more comprehensive and accurate tumor invasion feature data can be obtained by fully utilizing information from various aspects.

[0076] S50: Based on the tumor attribute feature data set, the tumor texture feature data, the tumor invasion feature data, and the clinical diagnosis and treatment data set, tumor lesion identification is performed to obtain the target tumor identification result.

[0077] The target tumor identification result can be understood as an identification result obtained after a comprehensive analysis and identification of the tumor's lesion status based on a variety of tumor-related characteristic data (such as tumor attribute characteristic data set, tumor texture characteristic data, and tumor invasion characteristic data), combined with the clinical diagnosis and treatment data of the target patient. This target tumor identification result may include, but is not limited to, the name, type, stage, and degree of lesion of the tumor; this application does not impose any limitations on this.

[0078] Specifically, multimodal data fusion can be performed on the tumor attribute feature data set, the tumor texture feature data, and the tumor invasion feature data. For example, feature selection algorithms (such as correlation analysis and principal component analysis) can be used to screen out the most valuable features for tumor lesion identification, reduce data dimensionality, and improve identification efficiency.

[0079] Furthermore, a comprehensive judgment can be made based on the fused characteristics combined with clinical diagnostic and treatment data. For example, it can be combined with the patient's basic information, such as the incidence and type of certain tumors being closely related to the patient's age and gender, or the risk of developing certain tumors being increased in people with a family history of cancer; or it can be combined with symptom and sign information, such as the patient's symptoms (e.g., pain, cough, weight loss, etc.) reflecting the growth location and development of the tumor, or physical signs found by the doctor through physical examination (e.g., the size, texture, and mobility of the mass) being an important basis for tumor identification; or it can be combined with treatment history and follow-up information, such as the patient's previous treatment history (e.g., surgery, chemotherapy, radiotherapy, etc.) affecting the current status and prognosis of the tumor, or regular follow-up data (e.g., changes in tumor markers, imaging examination results, etc.) reflecting the development trend of the tumor, etc. This application does not impose any restrictions on this.

[0080] As can be seen, in the above scheme, by acquiring the medical image data set, pathological image data set, and clinical diagnosis and treatment data set of the target object, and performing basic attribute extraction processing on each medical image data in the medical image data set, a tumor attribute feature data set of the target object is obtained. Furthermore, tumor texture feature recognition can be performed based on the tumor attribute feature data set and the sample label data set to obtain the tumor texture feature data of the target object. Thus, based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set, the tumor invasion feature data of the target object can be determined. Furthermore, based on the tumor attribute feature data set, the tumor texture feature data, the tumor invasion feature data, and the clinical diagnosis and treatment data set, more accurate target tumor identification results can be obtained, thus improving the accuracy of the tumor identification process.

[0081] For examples consistent with the above embodiments, please refer to... Figure 3 , Figure 3 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application, such as... Figure 3 As shown, it includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions. The program includes instructions for performing the following steps. Obtain user product requirement information, including functional information of the target business product; Based on a preset product acquisition function module group, the function module group includes at least one functionally independent function module; Based on the product requirement information, one or more functional modules matching the product requirement information are determined in the functional module group, and the target business product is generated based on the one or more functional modules.

[0082] In this example, by acquiring a set of medical image data, a set of pathological image data, and a set of clinical diagnosis and treatment data for the target object, and performing basic attribute extraction processing on each medical image data in the set of medical image data, a set of tumor attribute feature data for the target object is obtained. Furthermore, tumor texture feature recognition can be performed based on the set of tumor attribute feature data and the set of sample label data to obtain the tumor texture feature data of the target object. Therefore, based on the set of tumor attribute feature data, the tumor texture feature data, the set of medical image data, the set of pathological image data, and the set of sample label data, the tumor invasion feature data of the target object can be determined. Subsequently, based on the set of tumor attribute feature data, the tumor texture feature data, the tumor invasion feature data, and the set of clinical diagnosis and treatment data, a more accurate target tumor identification result can be obtained, thus improving the accuracy of the tumor identification process.

[0083] The above mainly describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the terminal includes the corresponding hardware structure and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0084] This application embodiment can divide the terminal into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0085] For those consistent with the above, please refer to Figure 4 , Figure 4 This application provides a schematic diagram of the structure of a tumor recognition device based on image processing. For example... Figure 4 As shown, the device includes: The acquisition unit 101 is used to acquire the medical image data set, pathological image data set, and clinical diagnosis and treatment data set of the target object; The first processing unit 102 is used to perform basic attribute extraction processing on each medical image data in the medical image data set to obtain the tumor attribute feature data set of the target object. The second processing unit 103 is used to perform tumor texture feature recognition based on the tumor attribute feature data set and the sample label data set to obtain tumor texture feature data of the target object. The first determining unit 104 is used to determine the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set; The identification unit 105 is used to identify tumor lesions and obtain the target tumor identification result based on the tumor attribute feature data set, the tumor texture feature data, the tumor invasion feature data and the clinical diagnosis and treatment data set.

[0086] In one possible implementation, the second processing unit 103 is configured to perform tumor texture feature recognition based on the tumor attribute feature data set and the sample label data set to obtain tumor texture feature data of the target object, specifically for: The generator of the initial generative adversarial image processing module generates tumor texture features based on each sample label data in the sample label dataset, resulting in a generated texture feature dataset. The discriminator of the initial generative adversarial image processing module is used to perform texture feature discrimination processing based on each generated texture feature data in the generated texture feature data set to obtain a discriminative texture feature data set; Based on the discriminative texture feature data set, the generated texture feature data set, and the sample label data set, determine the texture feature loss function; The initial generative adversarial image processing module is trained according to the texture feature loss function to obtain the target generative adversarial image processing module. The target generative adversarial image processing module is used to perform tumor texture feature recognition on each tumor attribute feature data in the tumor attribute feature data set to obtain the tumor texture feature data of the target object.

[0087] In one possible implementation, the second processing unit 103 is used to generate tumor texture features based on each sample label data in the sample label data set using the generator of the initial generative adversarial image processing module, to obtain a generated texture feature data set, specifically for: The generator of the initial generative adversarial image processing module encodes each sample label data in the sample label data set with a noise vector to obtain a tumor noise feature data set. The generator of the initial generative adversarial image processing module is used to perform feature mapping processing on each tumor noise feature data in the tumor noise feature data set to obtain an abstract feature data set. The generator of the initial generative adversarial image processing module is used to perform feature reshaping processing on each abstract feature data in the abstract feature data set to obtain a reshaped feature data set; The generator of the initial generative adversarial image processing module is used to upsample each reshaped feature data in the reshaped feature data set to obtain an upsampled feature data set. The generator of the initial generative adversarial image processing module is used to perform nonlinear enhancement processing on each upsampled feature data in the upsampled feature data set to obtain an enhanced feature data set. The generator of the initial generative adversarial image processing module is used to normalize each enhanced feature data in the enhanced feature data set to obtain a generated texture feature data set.

[0088] In one possible implementation, the first determining unit 104 is configured to determine the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set, specifically for: Infiltration range analysis is performed on each tumor attribute feature data in the tumor attribute feature data set to obtain infiltration space feature data; Infiltration difference analysis was performed on the tumor texture feature data to obtain infiltration heterogeneity feature data; Global infiltration analysis is performed based on the medical image dataset and the sample label dataset to obtain global infiltration feature data. Local infiltration analysis is performed on each pathological image in the pathological image dataset to obtain local infiltration feature data; Multi-scale fusion processing is performed on the infiltration spatial feature data, the infiltration heterogeneous feature data, the infiltration global feature data, and the infiltration local feature data to obtain the tumor infiltration feature data of the target object.

[0089] In one possible implementation, the first determining unit 104 is configured to perform global infiltration analysis based on the medical image dataset and the sample label dataset to obtain global infiltration feature data, specifically for: A comprehensive feature extraction process is performed on each medical image data in the medical image dataset to obtain a set of medical image feature maps. Sliding window processing is performed on each medical image feature map in the medical image feature map set to obtain a medical image anchor frame data set. Each medical image anchor frame in the medical image anchor frame dataset is classified to obtain a medical image foreground dataset. Shift regression calculation is performed on each medical image foreground data in the medical image foreground data set to obtain a medical image regression data set. Focusing feature extraction is performed on each medical image regression data in the medical image regression dataset to obtain a set of infiltration image feature maps; Based on the medical image anchor box data set, the medical image foreground data set, the infiltration image feature map set, and the sample label data set, determine the infiltration feature loss function; Based on the infiltration feature loss function, the initial infiltration analysis image processing module is trained to obtain the target infiltration analysis image processing module. The target infiltration analysis image processing module is used to perform infiltration analysis on each medical image data in the medical image dataset to obtain global infiltration feature data.

[0090] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the image processing-based tumor identification methods described in the above method embodiments.

[0091] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the image processing-based tumor identification methods described in the above method embodiments.

[0092] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0093] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0094] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0095] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0096] Furthermore, the functional units in the various embodiments of the application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software program module.

[0097] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0098] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.

[0099] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A tumor identification method based on image processing, characterized in that, The method includes: Acquire the medical imaging data set, pathological image data set, and clinical diagnosis and treatment data set of the target object; Basic attribute extraction processing is performed on each medical image data in the medical image dataset to obtain the tumor attribute feature dataset of the target object; Based on the tumor attribute feature data set and the sample label data set, tumor texture feature recognition is performed to obtain the tumor texture feature data of the target object; Based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set, the tumor invasion feature data of the target object is determined; Based on the tumor attribute feature data set, the tumor texture feature data, the tumor invasion feature data, and the clinical diagnosis and treatment data set, the tumor lesion is identified to obtain the target tumor identification result.

2. The tumor recognition method based on image processing according to claim 1, characterized in that, The step of identifying tumor texture features based on the tumor attribute feature data set and the sample label data set to obtain tumor texture feature data of the target object includes: The generator of the initial generative adversarial image processing module generates tumor texture features based on each sample label data in the sample label dataset, resulting in a generated texture feature dataset. The discriminator of the initial generative adversarial image processing module is used to perform texture feature discrimination processing based on each generated texture feature data in the generated texture feature data set to obtain a discriminative texture feature data set; Based on the discriminative texture feature data set, the generated texture feature data set, and the sample label data set, determine the texture feature loss function; The initial generative adversarial image processing module is trained according to the texture feature loss function to obtain the target generative adversarial image processing module. The target generative adversarial image processing module is used to perform tumor texture feature recognition on each tumor attribute feature data in the tumor attribute feature data set to obtain the tumor texture feature data of the target object.

3. The tumor recognition method based on image processing according to claim 2, characterized in that, The generator employing the initial generative adversarial image processing module generates tumor texture features based on each sample label data in the sample label data set, resulting in a generated texture feature data set, including: The generator of the initial generative adversarial image processing module encodes each sample label data in the sample label data set with a noise vector to obtain a tumor noise feature data set. The generator of the initial generative adversarial image processing module is used to perform feature mapping processing on each tumor noise feature data in the tumor noise feature data set to obtain an abstract feature data set. The generator of the initial generative adversarial image processing module is used to perform feature reshaping processing on each abstract feature data in the abstract feature data set to obtain a reshaped feature data set; The generator of the initial generative adversarial image processing module is used to upsample each reshaped feature data in the reshaped feature data set to obtain an upsampled feature data set. The generator of the initial generative adversarial image processing module is used to perform nonlinear enhancement processing on each upsampled feature data in the upsampled feature data set to obtain an enhanced feature data set. The generator of the initial generative adversarial image processing module is used to normalize each enhanced feature data in the enhanced feature data set to obtain a generated texture feature data set.

4. The tumor identification method based on image processing according to claim 3, characterized in that, The step of determining the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set includes: Infiltration range analysis is performed on each tumor attribute feature data in the tumor attribute feature data set to obtain infiltration space feature data; Infiltration difference analysis was performed on the tumor texture feature data to obtain infiltration heterogeneity feature data; Global infiltration analysis is performed based on the medical image dataset and the sample label dataset to obtain global infiltration feature data. Local infiltration analysis is performed on each pathological image in the pathological image dataset to obtain local infiltration feature data; Multi-scale fusion processing is performed on the infiltration spatial feature data, the infiltration heterogeneous feature data, the infiltration global feature data, and the infiltration local feature data to obtain the tumor infiltration feature data of the target object.

5. The tumor identification method based on image processing according to claim 4, characterized in that, The step of performing global infiltration analysis based on the medical image dataset and the sample label dataset to obtain global infiltration feature data includes: A comprehensive feature extraction process is performed on each medical image data in the medical image dataset to obtain a set of medical image feature maps. Sliding window processing is performed on each medical image feature map in the medical image feature map set to obtain a medical image anchor frame data set. Each medical image anchor frame in the medical image anchor frame dataset is classified to obtain a medical image foreground dataset. Shift regression calculation is performed on each medical image foreground data in the medical image foreground data set to obtain a medical image regression data set. Focusing feature extraction is performed on each medical image regression data in the medical image regression dataset to obtain a set of infiltration image feature maps; Based on the medical image anchor box data set, the medical image foreground data set, the infiltration image feature map set, and the sample label data set, determine the infiltration feature loss function; Based on the infiltration feature loss function, the initial infiltration analysis image processing module is trained to obtain the target infiltration analysis image processing module. The target infiltration analysis image processing module is used to perform infiltration analysis on each medical image data in the medical image dataset to obtain global infiltration feature data.

6. A tumor recognition device based on image processing, characterized in that, The device includes: The acquisition unit is used to acquire the medical image data set, pathological image data set, and clinical diagnosis and treatment data set of the target object; The first processing unit is used to perform basic attribute extraction processing on each medical image data in the medical image data set to obtain the tumor attribute feature data set of the target object. The second processing unit is used to perform tumor texture feature recognition based on the tumor attribute feature data set and the sample label data set to obtain tumor texture feature data of the target object. The first determining unit is used to determine the tumor invasion feature data of the target object based on the tumor attribute feature data set, the tumor texture feature data, the medical image data set, the pathological image data set, and the sample label data set; The identification unit is used to identify tumor lesions and obtain the target tumor identification result based on the tumor attribute feature data set, the tumor texture feature data, the tumor invasion feature data and the clinical diagnosis and treatment data set.

7. The tumor recognition device based on image processing according to claim 6, characterized in that, The second processing unit is used to perform tumor texture feature recognition based on the tumor attribute feature data set and the sample label data set to obtain tumor texture feature data of the target object, specifically for: The generator of the initial generative adversarial image processing module generates tumor texture features based on each sample label data in the sample label dataset, resulting in a generated texture feature dataset. The discriminator of the initial generative adversarial image processing module is used to perform texture feature discrimination processing based on each generated texture feature data in the generated texture feature data set to obtain a discriminative texture feature data set; Based on the discriminative texture feature data set, the generated texture feature data set, and the sample label data set, determine the texture feature loss function; The initial generative adversarial image processing module is trained according to the texture feature loss function to obtain the target generative adversarial image processing module. The target generative adversarial image processing module is used to perform tumor texture feature recognition on each tumor attribute feature data in the tumor attribute feature data set to obtain the tumor texture feature data of the target object.

8. The tumor recognition device based on image processing according to claim 7, characterized in that, The second processing unit is used to generate tumor texture features based on each sample label data in the sample label data set using the generator of the initial generative adversarial image processing module, thereby obtaining a generated texture feature data set, specifically for: The generator of the initial generative adversarial image processing module encodes each sample label data in the sample label data set with a noise vector to obtain a tumor noise feature data set. The generator of the initial generative adversarial image processing module is used to perform feature mapping processing on each tumor noise feature data in the tumor noise feature data set to obtain an abstract feature data set. The generator of the initial generative adversarial image processing module is used to perform feature reshaping processing on each abstract feature data in the abstract feature data set to obtain a reshaped feature data set; The generator of the initial generative adversarial image processing module is used to upsample each reshaped feature data in the reshaped feature data set to obtain an upsampled feature data set. The generator of the initial generative adversarial image processing module is used to perform nonlinear enhancement processing on each upsampled feature data in the upsampled feature data set to obtain an enhanced feature data set. The generator of the initial generative adversarial image processing module is used to normalize each enhanced feature data in the enhanced feature data set to obtain a generated texture feature data set.

9. A terminal, characterized in that, The system includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to invoke the program instructions to perform the method as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-5.

Images