Granular computing based method, system and device for low-quality medical image classification

Abstract

The application discloses a low-quality medical image classification method, system and device based on granular computing, belongs to the field of low-quality medical image classification, and aims to solve the technical problem of low classification accuracy of existing technologies for low-quality images containing noise, blur and insufficient contrast. The method comprises the following steps: acquiring sample data, converting probability language, constructing and training a classification network model, and classifying images in real time; the constructed probability language term set describes different light and dark degrees of images through five probability languages, each pixel value in each single-channel gray image of a low-quality medical sample image is mapped and converted into a probability language representation through a trapezoidal membership function, and a multi-channel feature map of each sample image is obtained. Through the probability language term set based on granular computing and the trapezoidal membership function, the fuzzy expression can be adaptively adjusted according to the actual characteristics of the input data, so that the fuzzy information can be modeled more flexibly and accurately.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence medical image processing technology, and relates to low-quality medical image classification, and particularly to a low-quality medical image classification method, system and device based on granular computing. Background Technology

[0002] With the widespread adoption of imaging equipment and the continuous growth in demand for medical services, the quality of medical images has become increasingly prominent, posing a key challenge to diagnostic accuracy and the efficiency of medical services. Low-quality images not only affect doctors' ability to identify lesions but may also lead to misdiagnosis or missed diagnosis, and even delay patient treatment. Against this backdrop, low-quality medical image processing has become a crucial research focus in the field of medical imaging. Therefore, the development of efficient image processing technologies has significant theoretical and practical value in improving the quality of medical services and reducing misdiagnosis and missed diagnosis.

[0003] To address the challenges of low-quality medical images, existing technologies have explored various image processing and enhancement techniques. Traditional image enhancement methods include horizontal rotation, random rotation, brightness perturbation, histogram equalization, and filtering algorithms. Patent application number 202410466402.4 discloses a deep learning-based blood glucose prediction method, device, equipment, and storage medium. In the training process of the image quality enhancement model, the CycleGAN model can be selected as the generative adversarial network model in the diffusion model. The CycleGAN model first synthesizes low-quality images based on high-quality images, and then pairs these synthesized low-quality images with high-quality images to form a training image pair. Based on the synthesized low-quality images, the U-Net model is selected as the basic enhancement model in the diffusion model. The diffusion model uses the U-Net model to iteratively learn the inverse mapping from low-quality images to high-quality images in multiple steps, minimizing the difference between the output image and the original high-quality image. In practical applications of the trained image quality enhancement model, the model continuously adds Gaussian noise to the input eye image to destroy low-quality components. Then, it uses a base enhancement model to iteratively remove the noise to restore a higher-quality eye image, thereby achieving the goal of enhancing the image quality of low-quality eye images.

[0004] Besides network model structure, existing technologies also enhance the linguistic expressiveness of low-quality medical image features through probabilistic language transformation, thereby improving the accuracy of low-quality medical image classification. This involves: constructing a term set where each pixel is represented by a linguistic colloquialism to indicate the image's brightness; using a triangular membership function to map each pixel value to [0,1], depicting the probability that a pixel value belongs to each linguistic term, and obtaining the final multi-channel feature map, which is then used as input to the network model.

[0005] Patent application number 202510696068.6 discloses a method, system, device, and medium for classifying low-quality medical images. The method includes: acquiring sample images, probabilistic language conversion, constructing an nmODE network model, training the nmODE network model, and real-time image classification. During probabilistic language conversion, a probabilistic language terminology set is constructed, using different probabilistic terms to characterize different levels of brightness in the image. The acquired sample images are then converted using this terminology set to obtain single-channel probabilistic representations of the sample images. Based on these single-channel probabilistic representations, a multi-channel feature map that can be input into the model is obtained. Probabilistic language conversion enhances the linguistic expression of low-quality medical image features, improving the accuracy of low-quality medical image classification.

[0006] Similar to the aforementioned invention patents, existing technologies primarily employ fixed triangular membership functions to calculate the membership degrees of triangular fuzzy sets in low-quality medical sample images and convert them into probabilistic language representations. This approach is simplistic and its parameters are preset, failing to effectively address the diversity and variability of noise, artifacts, and blur types in low-quality medical images. The analysis and abstraction of image information are static and unchanging. Furthermore, the description of uncertainty using triangular history functions is rather coarse, making it difficult to clearly reveal the model's decision-making basis when processing fuzzy regions, resulting in insufficient interpretability. Consequently, existing technologies are not well-suited to complex and ever-changing fuzzy scenarios, exhibiting low classification accuracy for noisy, blurred, and low-contrast low-quality images. Summary of the Invention

[0007] The purpose of this invention is to address the technical problem of low classification accuracy of low-quality images containing noise, blur, and insufficient contrast in existing technologies, and to provide a low-quality medical image classification method, system, and device based on granular computing.

[0008] To achieve the above objectives, the present invention specifically adopts the following technical solution: A low-quality medical image classification method based on granular computation includes the following steps: Step 1: Obtain sample images and label data; Medical sample images are acquired, blurred to varying degrees, and the benign or malignant lesions in the medical sample images are labeled to obtain label data. Step 2, Probabilistic Language Conversion; Construct a probabilistic language terminology set to describe the brightness of an image, and use different probabilistic language terms in the terminology set to describe the different brightness of pixels in the image; The sample images after blurring in step S1 are subjected to probabilistic language conversion using a probabilistic language terminology set to obtain single-channel probabilistic representation results corresponding to each probabilistic language term; based on the multiple single-channel probabilistic representation results of the sample images, the final multi-channel feature map is obtained; the specific steps are as follows: Step S2-1: Construct a probabilistic language terminology set Probabilistic Language Terminology Set Five probabilistic languages ​​are used to describe the different levels of brightness in an image; Step S2-2: For each pixel value in the single-channel grayscale image of the low-quality medical sample image, map it to the membership degree in the range [0, 1] through the trapezoidal membership function, and convert it into a probabilistic language representation to obtain the multi-channel feature map of each sample image; The formula for calculating the membership degree of pixels in a single-channel grayscale image is as follows: ; in, Indicates particle size, Represents pixel value, This represents the index of a term in the set of linguistic terms S. This represents the membership degree corresponding to the pixel value; Step 3: Construct a classification network model; Construct a classification network model; Step 4: Train the classification network model; The classification network model is trained using the multi-channel feature maps of each sample image obtained in step 2 and the corresponding label data in step 1. Step 5: Real-time image classification; The medical images to be classified are acquired and probabilistic language conversion is performed to obtain multi-channel feature maps. The multi-channel feature maps are then input into the classification network model trained in step 4, and the classification network model outputs the classification results of benign and malignant lesions.

[0009] Furthermore, in step S2-1, the probabilistic language terminology set is constructed. Represented as: ; Among them, the probabilistic language terminology set The pixel range corresponding to each probabilistic language is as follows: .

[0010] Furthermore, in step 3, the constructed classification network model includes a G-PL input layer, a G-PL convolutional layer, a G-PL pooling layer, a G-PL convolutional layer, a G-PL pooling layer, a G-PL flattening layer, and a G-PL fully connected layer.

[0011] Furthermore, in G-PL convolutional layers, assuming Let the input block containing probabilistic linguistic features be the first convolutional layer. Layer The output feature vector corresponding to each feature map is represented as follows: The formula for calculating the G-PL convolutional layer is: ; in, Indicates the first The feature map obtained after convolutional processing. This represents the activation function. Indicates the first Each feature map This represents the selection set of the input feature maps. Indicates the first The output of the c-th feature map in layer -1 is used as the input of layer l. The convolutional layer represents the first... layer, Representing language terms, Representing language terms The probability, This represents the convolution operator. Indicates the connection of the first Layer The first feature map and the previous layer's first feature map Convolution kernels for each feature map, Indicates the first Layer Additive bias of each feature map.

[0012] Furthermore, in the G-PL pooling layer, the calculation formula for the G-PL pooling layer is: ; in, Indicates the first In each characteristic channel, Output values ​​of the pooling region This indicates a max pooling operation. Indicates the first The first feature channel contains probabilistic linguistic features. Layer input features, The convolutional layer represents the first... layer; This represents the traversal variable, used to traverse the positions of all elements within the "window" currently being pooled; Indicates the first Each feature channel Representing language terms, Representing language terms The probability, Indicates the size of the pooling window. Indicates the scope of pooling.

[0013] Furthermore, in the G-PL fully connected layer, the calculation formula for the G-PL fully connected layer is: ; in, Indicates the current number The first in the layer The output of each neuron This represents the activation function. This indicates the number of neurons in the previous layer. Indicates the first level of the previous level One neuron, Indicates the current layer's first... One neuron, Indicates the first The output of the g-th neuron in layer -1 The convolutional layer represents the first... layer, Representing language terms, Representing language terms The probability, Indicates the first The first in the layer Input features of each neuron This represents the convolution operator. Indicates the connection of the first Layer The first neuron and the previous layer The convolutional kernel of a neuron, Indicates the first Layer Additive bias of each neuron.

[0014] A low-quality medical image classification system based on granular computing, comprising: The sample image and label data acquisition module is used to acquire medical sample images, perform different degrees of blurring on the medical sample images, and label the benign and malignant lesions in the medical sample images to obtain label data. The probabilistic language conversion module is used to construct a probabilistic language terminology set for describing the brightness of an image, and to use different probabilistic language terms in the probabilistic language terminology set to describe the different brightness of pixels in the image. The sample images and the blurred sample images obtained from the label data acquisition module are subjected to probabilistic language conversion using a probabilistic language terminology set to obtain single-channel probabilistic representation results corresponding to each probabilistic language term; based on the multiple single-channel probabilistic representation results of the sample images, the final multi-channel feature map is obtained; the specific steps are as follows: Step S2-1: Construct a probabilistic language terminology set Probabilistic Language Terminology Set Five probabilistic languages ​​are used to describe the different levels of brightness in an image; Step S2-2: For each pixel value in the single-channel grayscale image of the low-quality medical sample image, map it to the membership degree in the range of [0, 1] through the trapezoidal membership function, and convert it into a probabilistic language representation to obtain the multi-channel feature map of each sample image; The formula for calculating the membership degree of pixels in a single-channel grayscale image is as follows: ; in, Indicates particle size, Represents pixel value, This represents the index of a term in the set of linguistic terms S. This represents the membership degree corresponding to the pixel value; The classification network model building module is used to build classification network models; The classification network model training module is used to train the classification network model using the multi-channel feature maps of each sample image obtained by the probabilistic language conversion module and the corresponding label data in the sample image and label data acquisition module. The real-time image classification module is used to acquire medical images to be classified and perform probabilistic language conversion to obtain multi-channel feature maps. The multi-channel feature maps are then input into the classification network model training module to train the classification network model, which outputs the classification results of benign and malignant lesions.

[0015] A computer device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method described above.

[0016] The beneficial effects of this invention are as follows: 1. In this invention, a probabilistic language terminology set based on granular computation is introduced during probabilistic language conversion. By combining granular computation with the probabilistic language terminology set, image features, especially multi-scale and multi-granular lesion features in medical images, are innovatively transformed into more refined and diverse language terms. By innovatively introducing a trapezoidal membership function, the parameters of this function are continuously optimized during model training, enabling fuzzy expressions to adaptively adjust according to the actual features of the input data. This allows for more flexible and accurate modeling of fuzzy information, strong adaptability to complex and ever-changing fuzzy scenes, and improved classification accuracy for low-quality images with noise, blur, and insufficient contrast. This provides a new probabilistic language image processing method and tool for research and application in the field of image classification.

[0017] 2. In this invention, the parameters of the membership function are incorporated into the trainable parameters of the neural network, and the process of "how to define fuzziness" is also made into a part of the learning process, realizing end-to-end overall optimization. The model can backpropagate and autonomously learn the optimal fuzzy feature extraction method for the task based on the loss function of the final classification task.

[0018] 3. In this invention, the classification network model adopts the novel G-PL-CNN. Through the carefully designed image processing layer, G-PL-CNN core and granularity optimization module, it can capture image details and multi-scale features more sensitively. In addition, the algorithms for controlling its forward propagation, backward propagation and parameter update are derived in detail to ensure that the model can effectively learn and utilize granular probabilistic language information. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the present invention; Figure 2 This is a schematic diagram of the classification network model in this invention; Figure 3 This is a schematic diagram of a language dataset describing the brightness and darkness of an image in this invention; Figure 4 This is a schematic diagram of the granular probabilistic language representation transformation of grayscale images in this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0021] Therefore, all other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] Example 1 This embodiment provides a low-quality medical image classification method based on granular computing, which is used to classify low-quality medical images. Taking a breast cancer dataset as an example, this embodiment includes classifying the benign and malignant lesion regions in breast images; the benign and malignant lesion regions in the images are divided into three categories: benign, neutral, and malignant.

[0023] When classifying lesions as benign or malignant in low-quality medical images (breast MRI images), such as Figure 1 As shown, it specifically includes the following steps: Step 1: Obtain sample images and label data; Medical sample images are acquired, blurred to varying degrees, and the benign or malignant lesions in the medical sample images are labeled to obtain label data.

[0024] For the acquisition and preprocessing of sample images, please refer to the description in another invention patent application previously filed by the applicant (2025106960686). In this embodiment, the medical sample images and their corresponding labels are all from the management systems of various hospitals. In the hospital management system, each breast cancer classification dataset has a set of images. The names of these images represent a characteristic character, and they consist of an associated JSON document. This JSON document contains detailed information about three attribute values: image, category, and annotation. The image attribute consists of the image's height, width, ID, and name. The category attribute is used to determine the category to which the image belongs (i.e., label data) based on the image ID. The annotation attribute contains detailed information about each image in the dataset.

[0025] Blurring medical sample images to varying degrees can be applied. This can involve applying multiple degrees of blurring to the same image or different images, allowing those skilled in the art to choose the appropriate method. This simulates low-quality medical sample images. Therefore, using the blurred images as input to a network model enables the model to learn the ability to segment and classify lesions in low-quality medical images and output the benign or malignant classification of the lesions.

[0026] Because the sample size of low-quality breast medical images in each category is imbalanced after blurring, oversampling is necessary. Specifically: First, load the breast cancer classification dataset from the hospital system, obtain the number of images for each category, and blur them; second, take the largest number of images, ensuring that the oversampling value for each category is at least close to the maximum value. Then, divide the maximum value by the number of images in each category. This yields the oversampled values ​​for each category. Finally, the training set is dynamically expanded by a preset oversampling value, so that the sum of the original samples and the duplicate samples equals the number of images multiplied by their oversampling value plus 1.

[0027] oversampling value The calculation formula is: ; in, This represents the number of images in the i-th category.

[0028] Step 2, Probabilistic Language Conversion; Construct a probabilistic language terminology set to describe the brightness of an image, and use different probabilistic language terms in the terminology set to describe the different brightness of pixels in the image; The blurred sample images from step S1 are subjected to probabilistic language conversion using a probabilistic language terminology set, yielding single-channel probabilistic representations corresponding to each probabilistic language term. Based on the multiple single-channel probabilistic representations of the sample images, the final multi-channel feature map is obtained. Specifically: Step S2-1: Construct a probabilistic language terminology set Probabilistic Language Terminology Set Five probabilistic terms are used to describe the different levels of brightness in an image. Specifically: First, a probabilistic linguistic terminology set needs to be defined beforehand to describe the brightness and darkness of an image. This terminology set defines five piecewise linear transformation parameters that map pixel values ​​to different intervals to extract texture or density information related to pathological features, and constructs an antisymmetric probability function for probabilistic linguistic classification. Specifically, the probabilistic linguistic terminology set... Terminology elements in ~ These describe different levels of brightness or darkness of pixels. The pixel value for "dark" ranges from 0 to 63.75, represented in probabilistic language. The pixel range corresponding to "slightly darker" is 0 to 127.5, which is represented in probabilistic language. The pixel range corresponding to "general" is 63.75-191.25, which is represented in probabilistic language. The pixel range corresponding to "slightly brighter" is 127.5 to 255, which is represented in probabilistic language. The pixel range corresponding to "bright" is 191.25 to 255, which can be represented in probabilistic language. Therefore, this probabilistic language terminology set It can be represented as: .

[0029] A diagram illustrating a language dataset used to describe the brightness and darkness of an image, such as... Figure 3 As shown; a set of probabilistic language terms The pixel ranges corresponding to each probabilistic language are shown in Table 1.

[0030] Table 1 Probabilistic Language - Pixel Correspondence Table

[0031] Among them, the five probabilistic languages ​​are characterized by five trapezoidal membership functions of the same shape (specifically, the function representing intermediate language terms (such as "slightly dark", "normal", "slightly bright") is a symmetric trapezoidal fuzzy set, while the fuzzy set representing the two end terms ("dark" and "bright") is a shoulder trapezoidal fuzzy set), and the sum of the membership degrees of each pixel in each fuzzy set is 1.

[0032] Step S2-2: For each pixel value in the single-channel grayscale image of the low-quality medical sample image, map it to a membership degree in the range [0, 1] using a trapezoidal membership function, and convert it into a probabilistic language representation to obtain the multi-channel feature map of each sample image.

[0033] For low-quality medical sample images, extract each single-channel grayscale image of each sample image (the sample image is an RGB image, so each image has three single-channel grayscale images); map each pixel value in each single-channel grayscale image to a membership degree in the range [0,1] through a trapezoidal membership function, calculate the membership degree, and regard the membership degree corresponding to the pixel value as the probability of the probabilistic language term element.

[0034] That is, the probabilistic representation of a pixel in a single-channel grayscale image is as follows: ; in, - These correspond to the probability of that pixel in the corresponding probability language.

[0035] The formula for calculating the membership degree of a pixel in a single-channel grayscale image is as follows: ; in, Indicates particle size, Represents pixel value, This represents the index of a term in the set of linguistic terms S. This represents the membership degree corresponding to the pixel value.

[0036] According to a specific rule, the probabilistic linguistic representation of each single-channel grayscale image is transformed into 5 independent channels (that is, the 5 probabilities are merged into five independent channels, and the probability in a certain independent channel is used as the value of that independent channel), resulting in a multi-channel feature map of each single-channel grayscale image. The multi-channel feature map corresponding to each single-channel grayscale image of each sample image will be used for the training of subsequent models.

[0037] Taking pixel 90 with a granularity of 0.1 as an example, its membership degrees to the five linguistic terms mentioned above are 0, 0.59, 0.41, 0, and 0, respectively. If we consider the membership degree as the probability of a probabilistic linguistic term element, then pixel 90 can be converted into a probabilistic linguistic representation. .

[0038] Using the aforementioned pixel-level probabilistic language description method, the overall medical image is converted into a probabilistic language representation. For a single-channel grayscale image, the five probability values ​​are treated as five independent channels. For a three-channel RGB image, the converted probabilistic language representation contains 15 channels. The specific implementation of the grayscale image is as follows... Figure 4 As shown.

[0039] Step 3: Construct a classification network model; Construct a classification network model, which is a granular probabilistic language convolutional neural network, including a G-PL input layer, a G-PL convolutional layer, a G-PL pooling layer, a G-PL flattening layer, and a G-PL fully connected layer, such as... Figure 2 As shown ( Figure 2 (The input layer is not shown in the image).

[0040] The G-PL input layer is a probabilistic language input layer based on granular computation, fundamentally different from the standard input layer in traditional convolutional neural network architectures. In traditional convolutional neural networks, the input layer directly receives the original or preprocessed pixel intensity matrix of the image, typically represented by grayscale or RGB values. In contrast, the input to the G-PL input layer is a multi-channel feature map generated through the probabilistic language transformation of pixel values. Each channel in this tensor corresponds to a specific linguistic term, representing different brightness levels—such as "dark," "slightly dark," "moderate," "slightly bright," or "bright." The numerical values ​​within each channel encode the membership degree, i.e., the probability of a specific pixel term corresponding to a brightness level. The G-PL input layer encapsulates refined brightness semantics, providing a more expressive and uncertainty-aware representation of image features.

[0041] The G-PL convolutional layer performs convolution operations between the input tensor and a set of trainable kernels, taking the feature tensor output by the G-PL input layer as input to generate local feature representations. Unlike convolutional layers that operate directly on the raw pixel intensities, the G-PL convolutional layer is designed to process probabilistic linguistic information. Therefore, semantic features—i.e., the degree of belonging to linguistic terms—are preserved throughout the convolution process. This adaptation transforms standard convolutional layers into granular computation-based probabilistic linguistic convolutional layers.

[0042] Assumption Let the input block containing probabilistic linguistic features be the first convolutional layer. Layer The output feature vector corresponding to each feature map is represented as follows: Therefore, the calculation formula for the G-PL convolutional layer is: ; in, Indicates the first The feature map obtained after convolutional processing. This represents the activation function. Indicates the first Each feature map This represents the selection set of the input feature maps. Indicates the first The output of the c-th feature map in layer c is used as Layer input; The convolutional layer represents the first... layer, Representing language terms, Representing language terms The probability, This represents the convolution operator. Indicates the connection of the first Layer The first feature map and the previous layer's first feature map Convolution kernels for each feature map, Indicates the first Layer Additive bias of each feature map.

[0043] The G-PL pooling layer, following each G-PL convolutional layer, acts as a downsampling or subsampling component. Its input is the local feature representation output by the G-PL convolutional layer, which is then spatially sampled to generate a multi-dimensional, multi-channel global feature vector. By aggregating local responses within each linguistic feature channel, this layer achieves two main objectives: refining the features extracted by the previous convolutional layer and reducing the spatial dimensionality of the feature representation.

[0044] This G-PL pooling layer not only improves computational efficiency but also preserves the key probabilistic linguistic semantics embedded in the input, thus enabling the model to maintain interpretability while enhancing generalization performance. The calculation formula for this G-PL pooling layer is as follows: ; in, Indicates the first In each characteristic channel, Output values ​​of the pooling region This indicates a max pooling operation. Indicates the first Layer On each feature channel located at Input features of the pooling region, The convolutional layer represents the first... layer; This represents the traversal variable, used to traverse the positions of all elements within the "window" currently being pooled; Indicates the first Each feature channel Representing language terms, Representing language terms The probability, Indicates the size of the pooling window. Indicates the scope of pooling.

[0045] G-PL flattening layers are a common operation in convolutional neural networks used to "flatten" multidimensional feature tensors into one-dimensional vectors before they enter fully connected layers or undergo subsequent linear transformations. Typically, if the output of the previous layer is a tensor (…), then… )(in It is the number of samples. It is the number of channels. and (Height and width), the flattening operation transforms it into ( In this way, subsequent fully connected layers can perform linear mapping and nonlinear transformation on the flattened vector of each sample, thereby achieving higher-level feature combination.

[0046] The G-PL fully connected layer has each neuron fully connected to all activations in the previous layer, enabling it to perform high-level inference on aggregated features; neurons within the same layer are not interconnected, maintaining a standard feedforward architecture.

[0047] The calculation formula for the G-PL fully connected layer is: ; in, Indicates the current number The first in the layer The output of each neuron This represents the activation function. This indicates the number of neurons in the previous layer. Indicates the first level of the previous level One neuron, Indicates the current layer's first... One neuron, Indicates the first The output of the g-th neuron in layer g is used as... Layer input; The convolutional layer represents the first... layer, Representing language terms, Representing language terms The probability, Indicates the first The first in the layer Input features of each neuron This represents the convolution operator. Indicates the connection of the first Layer The first neuron and the previous layer The convolutional kernel of a neuron, Indicates the first Layer Additive bias of each neuron.

[0048] Step 4: Train the classification network model; The classification network model is trained using the multi-channel feature maps of each sample image obtained in step 2 and the corresponding label data from step 1.

[0049] When training a classification network model, existing training methods can be used. Those skilled in the art can choose the appropriate method according to their needs without having to make any creative effort.

[0050] Step 5: Real-time image classification; The medical images to be classified are acquired and probabilistic language conversion is performed to obtain multi-channel feature maps. The multi-channel feature maps are then input into the classification network model trained in step 4, and the classification network model outputs the classification results of benign and malignant lesions.

[0051] Example 2 This embodiment provides a low-quality medical image classification system based on granular computing, which includes: The sample image and label data acquisition module is used to acquire medical sample images, perform different degrees of blurring on the medical sample images, and label the benign or malignant lesions in the medical sample images to obtain label data.

[0052] For the acquisition and preprocessing of sample images, please refer to the description in another invention patent application previously filed by the applicant (2025106960686). In this embodiment, the medical sample images and their corresponding labels are all from the management systems of various hospitals. In the hospital management system, each breast cancer classification dataset has a set of images. The names of these images represent a characteristic character, and they consist of an associated JSON document. This JSON document contains detailed information about three attribute values: image, category, and annotation. The image attribute consists of the image's height, width, ID, and name. The category attribute is used to determine the category to which the image belongs (i.e., label data) based on the image ID. The annotation attribute contains detailed information about each image in the dataset.

[0053] Blurring medical sample images to varying degrees can be applied. This can involve applying multiple degrees of blurring to the same image or different images, allowing those skilled in the art to choose the appropriate method. This simulates low-quality medical sample images. Therefore, using the blurred images as input to a network model enables the model to learn the ability to segment and classify lesions in low-quality medical images and output the benign or malignant classification of the lesions.

[0054] Because the sample size of low-quality breast medical images in each category is imbalanced after blurring, oversampling is necessary. Specifically: First, load the breast cancer classification dataset from the hospital system, obtain the number of images for each category, and blur them; second, take the largest number of images, ensuring that the oversampling value for each category is at least close to the maximum value. Then, divide the maximum value by the number of images in each category. This yields the oversampled values ​​for each category. Finally, the training set is dynamically expanded by a preset oversampling value, so that the sum of the original samples and the duplicate samples equals the number of images multiplied by their oversampling value plus 1.

[0055] oversampling value The calculation formula is: ; in, This represents the number of images in the i-th category.

[0056] The probabilistic language conversion module is used to construct a probabilistic language terminology set for describing the brightness of an image, and to use different probabilistic language terms in the probabilistic language terminology set to describe the different brightness of pixels in the image. The blurred sample images from step S1 are subjected to probabilistic language conversion using a probabilistic language terminology set, yielding single-channel probabilistic representations corresponding to each probabilistic language term. Based on the multiple single-channel probabilistic representations of the sample images, the final multi-channel feature map is obtained. Specifically: Step S2-1: Construct a probabilistic language terminology set Probabilistic Language Terminology Set Five probabilistic terms are used to describe the different levels of brightness in an image. Specifically: First, a probabilistic linguistic terminology set needs to be defined beforehand to describe the brightness and darkness of an image. This terminology set defines five piecewise linear transformation parameters that map pixel values ​​to different intervals to extract texture or density information related to pathological features, and constructs an antisymmetric probability function for probabilistic linguistic classification. Specifically, the probabilistic linguistic terminology set... Terminology elements in ~ These describe different levels of brightness or darkness of pixels. The pixel value for "dark" ranges from 0 to 63.75, represented in probabilistic language. The pixel range corresponding to "slightly darker" is 0 to 127.5, which is represented in probabilistic language. The pixel range corresponding to "general" is 63.75-191.25, which is represented in probabilistic language. The pixel range corresponding to "slightly brighter" is 127.5 to 255, which is represented in probabilistic language. The pixel range corresponding to "bright" is 191.25 to 255, which can be represented in probabilistic language. Therefore, this probabilistic language terminology set It can be represented as: .

[0057] A diagram illustrating a language dataset used to describe the brightness and darkness of an image, such as... Figure 3 As shown; a set of probabilistic language terms The pixel ranges corresponding to each probabilistic language are shown in Table 1.

[0058] Table 1 Probabilistic Language - Pixel Correspondence Table

[0059] Among them, the five probabilistic languages ​​are characterized by five trapezoidal membership functions of the same shape (specifically, the function representing intermediate language terms (such as "slightly dark", "normal", "slightly bright") is a symmetric trapezoidal fuzzy set, while the fuzzy set representing the two end terms ("dark" and "bright") is a shoulder trapezoidal fuzzy set), and the sum of the membership degrees of each pixel in each fuzzy set is 1.

[0060] Step S2-2: For each pixel value in the single-channel grayscale image of the low-quality medical sample image, map it to a membership degree in the range [0, 1] using a trapezoidal membership function, and convert it into a probabilistic language representation to obtain the multi-channel feature map of each sample image.

[0061] For low-quality medical sample images, extract each single-channel grayscale image of each sample image (the sample image is an RGB image, so each image has three single-channel grayscale images); map each pixel value in each single-channel grayscale image to a membership degree in the range [0,1] through a trapezoidal membership function, calculate the membership degree, and regard the membership degree corresponding to the pixel value as the probability of the probabilistic language term element.

[0062] That is, the probabilistic representation of a pixel in a single-channel grayscale image is as follows: ; in, - These correspond to the probability of that pixel in the corresponding probability language.

[0063] The formula for calculating the membership degree of a pixel in a single-channel grayscale image is as follows: ; in, Indicates particle size, Represents pixel value, This represents the index of a term in the set of linguistic terms S. This represents the membership degree corresponding to the pixel value.

[0064] According to a specific rule, the probabilistic linguistic representation of each single-channel grayscale image is transformed into 5 independent channels (that is, the 5 probabilities are merged into five independent channels, and the probability in a certain independent channel is used as the value of that independent channel), resulting in a multi-channel feature map of each single-channel grayscale image. The multi-channel feature map corresponding to each single-channel grayscale image of each sample image will be used for the training of subsequent models.

[0065] Taking pixel 90 with a granularity of 0.1 as an example, its membership degrees to the five linguistic terms mentioned above are 0, 0.59, 0.41, 0, and 0, respectively. If we consider the membership degree as the probability of a probabilistic linguistic term element, then pixel 90 can be converted into a probabilistic linguistic representation. .

[0066] Using the aforementioned pixel-level probabilistic language description method, the overall medical image is converted into a probabilistic language representation. For a single-channel grayscale image, the five probability values ​​are treated as five independent channels. For a three-channel RGB image, the converted probabilistic language representation contains 15 channels. The specific implementation of the grayscale image is as follows... Figure 4 As shown.

[0067] The classification network model building module is used to construct classification network models. These models are granular probabilistic language convolutional neural networks, including G-PL input layers, G-PL convolutional layers, G-PL pooling layers, G-PL flattening layers, and G-PL fully connected layers. Figure 2 As shown ( Figure 2 (The input layer is not shown in the image).

[0068] The G-PL input layer is a probabilistic language input layer based on granular computation, fundamentally different from the standard input layer in traditional convolutional neural network architectures. In traditional convolutional neural networks, the input layer directly receives the original or preprocessed pixel intensity matrix of the image, typically represented by grayscale or RGB values. In contrast, the input to the G-PL input layer is a multi-channel feature map generated through the probabilistic language transformation of pixel values. Each channel in this tensor corresponds to a specific linguistic term, representing different brightness levels—such as "dark," "slightly dark," "moderate," "slightly bright," or "bright." The numerical values ​​within each channel encode the membership degree, i.e., the probability of a specific pixel term corresponding to a brightness level. The G-PL input layer encapsulates refined brightness semantics, providing a more expressive and uncertainty-aware representation of image features.

[0069] The G-PL convolutional layer performs convolution operations between the input tensor and a set of trainable kernels, taking the feature tensor output by the G-PL input layer as input to generate local feature representations. Unlike convolutional layers that operate directly on the raw pixel intensities, the G-PL convolutional layer is designed to process probabilistic linguistic information. Therefore, semantic features—i.e., the degree of belonging to linguistic terms—are preserved throughout the convolution process. This adaptation transforms standard convolutional layers into granular computation-based probabilistic linguistic convolutional layers.

[0070] Assumption Let the input block containing probabilistic linguistic features be the first convolutional layer. Layer The output feature vector corresponding to each feature map is represented as follows: Therefore, the calculation formula for the G-PL convolutional layer is: ; in, Indicates the first The feature map obtained after convolutional processing. This represents the activation function. Indicates the first Each feature map This represents the selection set of the input feature maps. Indicates the first Layer The output of each feature map, as Layer input; The convolutional layer represents the first... layer, Representing language terms, Representing language terms The probability, This represents the convolution operator. Indicates the connection of the first Layer The first feature map and the previous layer's first feature map Convolution kernels for each feature map, Indicates the first Layer Additive bias of each feature map.

[0071] The G-PL pooling layer, following each G-PL convolutional layer, acts as a downsampling or subsampling component. Its input is the local feature representation output by the G-PL convolutional layer, which is then spatially sampled to generate a multi-dimensional, multi-channel global feature vector. By aggregating local responses within each linguistic feature channel, this layer achieves two main objectives: refining the features extracted by the previous convolutional layer and reducing the spatial dimensionality of the feature representation.

[0072] This G-PL pooling layer not only improves computational efficiency but also preserves the key probabilistic linguistic semantics embedded in the input, thus enabling the model to maintain interpretability while enhancing generalization performance. The calculation formula for this G-PL pooling layer is as follows: ; in, Indicates the first In each characteristic channel, Output values ​​of the pooling region This indicates a max pooling operation. Indicates the first Layer On each feature channel located at Input features of the pooling region, The convolutional layer represents the first... layer; This represents the traversal variable, used to traverse the positions of all elements within the "window" currently being pooled; Indicates the first Each feature channel Representing language terms, Representing language terms The probability, Indicates the size of the pooling window. Indicates the scope of pooling.

[0073] G-PL flattening layers are a common operation in convolutional neural networks used to "flatten" multidimensional feature tensors into one-dimensional vectors before they enter fully connected layers or undergo subsequent linear transformations. Typically, if the output of the previous layer is a tensor ( )(in It is the number of samples. It is the number of channels. and (Height and width), the flattening operation transforms it into ( In this way, subsequent fully connected layers can perform linear mapping and nonlinear transformation on the flattened vector of each sample, thereby achieving higher-level feature combination.

[0074] The G-PL fully connected layer has each neuron fully connected to all activations in the previous layer, enabling it to perform high-level inference on aggregated features; neurons within the same layer are not interconnected, maintaining a standard feedforward architecture.

[0075] The calculation formula for the G-PL fully connected layer is: ; in, Indicates the current number The first in the layer The output of each neuron This represents the activation function. This indicates the number of neurons in the previous layer. Indicates the first level of the previous level One neuron, Indicates the current layer's first... One neuron, Indicates the first The output of the g-th neuron in layer g is used as... Layer input; The convolutional layer represents the first... layer, Representing language terms, Representing language terms The probability, Indicates the first The first in the layer Input features of each neuron This represents the convolution operator. Indicates the connection of the first Layer The first neuron and the previous layer The convolutional kernel of a neuron, Indicates the first Layer Additive bias of neurons.

[0076] The classification network model training module is used to train the classification network model using the multi-channel feature maps of each sample image obtained from the probabilistic language conversion module and the corresponding label data from the sample image and label data acquisition module.

[0077] When training a classification network model, existing training methods can be used. Those skilled in the art can choose the appropriate method according to their needs without having to make any creative effort.

[0078] The real-time image classification module is used to acquire medical images to be classified and perform probabilistic language conversion to obtain multi-channel feature maps. The multi-channel feature maps are then input into the classification network model training module to train the classification network model, which outputs the classification results of benign and malignant lesions.

[0079] Example 3 A computer device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform steps of a granular computing-based low-quality medical image classification method.

[0080] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0081] The memory includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or 3D interface display memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory may be an internal storage unit of the computer device, such as the hard disk or memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device. Of course, the memory may include both internal storage units and external storage devices of the computer device. In this embodiment, the memory is often used to store the operating system and various application software installed on the computer device, such as the program code of the low-quality medical image classification method based on granular computing. In addition, the memory can also be used to temporarily store various types of data that have been output or will be output.

[0082] In some embodiments, the processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor is typically used to control the overall operation of the computer device. In this embodiment, the processor is used to run program code stored in the memory or process data, for example, to run the program code for the granular computing-based low-quality medical image classification method.

Claims

1. A low-quality medical image classification method based on granular computing, characterized in that, Includes the following steps: Step 1: Obtain sample images and label data; Medical sample images are acquired, blurred to varying degrees, and the benign or malignant lesions in the medical sample images are labeled to obtain label data. Step 2, Probabilistic Language Conversion; Construct a probabilistic language terminology set to describe the brightness of an image, and use different probabilistic language terms in the terminology set to describe the different brightness of pixels in the image; The sample images after blurring in step S1 are converted using a probabilistic language terminology set to obtain single-channel probabilistic representation results corresponding to each probabilistic language term. Based on the multiple single-channel probabilistic representation results of the sample images, the final multi-channel feature map is obtained. The specific steps are as follows: Step S2-1: Construct a probabilistic language terminology set Probabilistic Language Terminology Set Five probabilistic languages ​​are used to describe the different levels of brightness in an image; Step S2-2: For each pixel value in the single-channel grayscale image of the low-quality medical sample image, map it to the membership degree in the range of [0, 1] through the trapezoidal membership function, and convert it into a probabilistic language representation to obtain the multi-channel feature map of each sample image; The formula for calculating the membership degree of pixels in a single-channel grayscale image is as follows: ； in, Indicates particle size, Represents pixel value, Represents the index of a term in the probabilistic language term set S. This represents the membership degree corresponding to the pixel value; Step 3: Construct a classification network model; The classification network model includes a G-PL input layer, a G-PL convolutional layer, a G-PL pooling layer, a G-PL convolutional layer, a G-PL pooling layer, a G-PL flattening layer, and a G-PL fully connected layer. In the G-PL pooling layer, the calculation formula for the G-PL pooling layer is: ； in, Indicates the first In each characteristic channel, Output values ​​of the pooling region This indicates a max pooling operation. Indicates the first -1st floor Located on each feature channel Input features of the pooling region, The convolutional layer represents the first... layer; This represents the traversal variable, used to traverse the positions of all elements within the currently pooling window; Indicates the first Each feature channel Representing language terms, Representing language terms The probability, Indicates the size of the pooling window. Indicates the scope of pooling; Step 4: Train the classification network model; The classification network model is trained using the multi-channel feature maps of each sample image obtained in step 2 and the corresponding label data in step 1. Step 5: Real-time image classification; The medical images to be classified are acquired and probabilistic language conversion is performed to obtain multi-channel feature maps. The multi-channel feature maps are then input into the classification network model trained in step 4, and the classification network model outputs the classification results of benign and malignant lesions.

2. The low-quality medical image classification method based on granular computing as described in claim 1, characterized in that, In step S2-1, the probabilistic language terminology set is constructed. Represented as: ； Among them, the probabilistic language terminology set The pixel ranges corresponding to each probabilistic language are: 。 3. The low-quality medical image classification method based on granular computing as described in claim 1, characterized in that, In G-PL convolutional layers, assuming This represents an input block containing probabilistic linguistic features; the formula for calculating the G-PL convolutional layer is: ； in, Indicates the first The feature map obtained after convolutional processing. This represents the activation function. Indicates the first Each feature map This represents the selection set of the input feature maps. Indicates the first The output of the c-th feature map in layer -1 The convolutional layer represents the first... layer, Representing language terms, Representing language terms The probability, This represents the convolution operator. Indicates the connection of the first Layer The first feature map and the previous layer's first feature map Convolution kernels for each feature map, Indicates the first Layer Additive bias of each feature map.

4. The low-quality medical image classification method based on granular computing as described in claim 1, characterized in that, In the G-PL fully connected layer, the calculation formula for the G-PL fully connected layer is: ； in, Indicates the current number The first in the layer The output of each neuron This represents the activation function. This indicates the number of neurons in the previous layer. Indicates the first level of the previous level One neuron, Indicates the current layer's first... One neuron, Indicates the first The output of the g-th neuron in layer -1 The convolutional layer represents the first... layer, Representing language terms, Representing language terms The probability, Indicates the first The first in the layer Input features of each neuron This represents the convolution operator. Indicates the connection of the first Layer The first neuron and the previous layer The convolutional kernel of a neuron, Indicates the first Layer Additive bias of neurons.

5. A low-quality medical image classification system based on granular computing, characterized in that, include: The sample image and label data acquisition module is used to acquire medical sample images, perform different degrees of blurring on the medical sample images, and label the benign and malignant lesions in the medical sample images to obtain label data. The probabilistic language conversion module is used to construct a probabilistic language terminology set for describing the brightness of an image, and to use different probabilistic language terms in the probabilistic language terminology set to describe the different brightness of pixels in the image. The sample images, after being blurred in the label data acquisition module, are converted using a probabilistic language terminology set to obtain a single-channel probabilistic representation result corresponding to each probabilistic language term. Based on the multiple single-channel probabilistic representation results of the sample images, the final multi-channel feature map is obtained. The specific steps are as follows: Step S2-1: Construct a probabilistic language terminology set Probabilistic Language Terminology Set Five probabilistic languages ​​are used to describe the different levels of brightness in an image; Step S2-2: For each pixel value in the single-channel grayscale image of the low-quality medical sample image, map it to the membership degree in the range of [0, 1] through the trapezoidal membership function, and convert it into a probabilistic language representation to obtain the multi-channel feature map of each sample image; The formula for calculating the membership degree of pixels in a single-channel grayscale image is as follows: ； in, Indicates particle size, Represents pixel value, Represents the index of a term in the probabilistic language term set S. This represents the membership degree corresponding to the pixel value; The classification network model building module is used to build classification network models, which include G-PL input layer, G-PL convolutional layer, G-PL pooling layer, G-PL convolutional layer, G-PL pooling layer, G-PL flattening layer, and G-PL fully connected layer. In the G-PL pooling layer, the calculation formula for the G-PL pooling layer is: ； in, Indicates the first In each characteristic channel, Output values ​​of the pooling region This indicates a max pooling operation. Indicates the first -1st floor On each feature channel located at Input features of the pooling region, The convolutional layer represents the first... layer; This represents the traversal variable, used to traverse the positions of all elements within the currently pooling window; Indicates the first Each feature channel Representing language terms, Representing language terms The probability, Indicates the size of the pooling window. Indicates the scope of pooling; The classification network model training module is used to train the classification network model using the multi-channel feature maps of each sample image obtained by the probabilistic language conversion module and the corresponding label data in the sample image and label data acquisition module. The real-time image classification module is used to acquire medical images to be classified and perform probabilistic language conversion to obtain multi-channel feature maps. The multi-channel feature maps are then input into the classification network model training module to train the classification network model, which outputs the classification results of benign and malignant lesions.

6. A computer device, characterized in that: It includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 4.

Images