A method for evaluating the imaging quality of pathological images

Abstract

The present invention discloses a method for evaluating the imaging quality of pathological images, belonging to the field of medical image quality control. Using the pathological images captured by an optical microscope at multiple defocus planes, a defocus dataset of pathological images is constructed; the defocus pathological image dataset is preprocessed; a pathological image quality evaluation network based on frequency domain features is established; using a combined loss function, an optimized deep neural network model for evaluating the quality of pathological images based on frequency domain features is obtained through optimization training; using the optimized pathological image quality evaluation network, the quality level of the pathological images is output. The present invention adopts an end-to-end deep learning model, proposes a frequency convolution module, extracts the frequency features of pathological images, improves the performance of pathological image quality evaluation, and has good robustness and strong adaptability.

Description

Technical Field

[0001] The present invention relates to the field of medical image quality control, and particularly relates to a method for evaluating the imaging quality of pathological images. Background Art

[0002] Optical microscopy imaging provides tissue visualization at the cellular level and can intuitively display information such as tissue structure and physiological state. Therefore, it has gradually become an important means for biomedical research, clinical diagnosis, and prognosis of diseases such as cancer. Although optical microscopy has made significant contributions to pathology, a basic prerequisite is to capture high-quality focused images. However, due to the usually high magnification of cytopathological imaging microscopes, the depth of field of the microscope is usually very small (less than 1 micrometer). Out-of-focus blur is the main reason for the low imaging quality of microscopes, and this blur may lead to errors in doctors' judgments of lesions, affecting the accuracy of diagnosis and the effectiveness of treatment. In addition, computer-aided diagnosis (CAD) systems used for microscope imaging are also affected by low-quality samples. Therefore, the evaluation of the imaging quality of pathological images is of great significance for biomedical research, clinical diagnosis, and prognosis judgment of diseases such as cancer. Summary of the Invention

[0003] The main object of the present invention is to provide a method for evaluating the imaging quality of pathological images by combining frequency domain features and feature distillation, so as to avoid the problem that the image quality affects the auxiliary diagnosis result in the computer-aided diagnosis system of medical pathological images.

[0004] A method for evaluating the imaging quality of pathological images includes the following steps:

[0005] S1. Construct a pathological image dataset;

[0006] S2. Preprocess the pathological image dataset;

[0007] S3. Construct a deep neural network model for evaluating the imaging quality of pathological images based on frequency domain features;

[0008] S4. Train and optimize the constructed deep neural network model with the preprocessed pathological image dataset;

[0009] S5. Input the pathological image to be determined into the optimized deep neural network model for evaluating the imaging quality of pathological images to obtain the imaging quality evaluation level of the pathological image.

[0010] The method for constructing the pathological image dataset in the step S1 is as follows:

[0011] Obtain the pathological images of the sample taken by the optical microscope at multiple defocus planes, obtain the focus stack of the sample, and screen out the clear pathological images and blurred pathological images from it as training sample pairs. All the training samples constitute the pathological image dataset.

[0012] The training sample pairs are randomly sampled in pairs. When sampling, a clear pathological image and another image in the focus stack of the same sample are selected to form a training sample pair.

[0013] The screening method for the clear pathological images and blurred pathological images is as follows:

[0014] Select the image with a defocus distance of 0 as the clear pathological image, and select the image with a defocus distance greater than 4 microns as the blurred pathological image.

[0015] The steps for preprocessing the pathological image dataset in S2 are as follows:

[0016] S21. Divide the pathological image dataset into a training set and a test set according to a ratio;

[0017] S22. Cut the training sample pairs in the training set into regions of size 512 with a step size of 256;

[0018] S23. Randomly select 256 regions of 256 from each region as training samples;

[0019] S24. Construct a training set from all the training samples, and the obtained training set is:

[0020]

[0021] Among them, is the input blurred pathological image and the clear pathological image y, represents the number of channels of the image, represents the height of the image, represents the width of the image, represents the number of blurred pathological images, represents the number of clear pathological images.

[0022] The training and optimization method of the deep neural network model in S4 is as follows:

[0023] Input the training sample pairs of the training set into the deep neural network model. The backbone network of the deep neural network model extracts features from the pathological image and the pathological image The distillation network of the deep neural network model uses one of the pathological images As the input, after feature extraction and fusing the intermediate layer features extracted by the backbone network, an image with another defocus distance is generated. , so as to simulate the generation process of defocused images;

[0024] Then, by establishing another input image and the generated image , the loss function is used to optimize the parameters of the deep neural network model.

[0025] The backbone network of the deep neural network model includes four layers of Reslock. Suppose the size of the feature map extracted by the third ResBlock of the backbone network is , where C represents the number of channels, and H and W represent the height and width of the feature map. The third layer feature maps of image and image are spliced together to obtain a feature map with a size of . Through two layers of 1*1 convolutions, the defocus kernels of paired images in the same sample are estimated, the mutual defocus distance features between the two images are extracted, and the feature map is passed to the distillation network. The distillation network takes one of the images as the input. First, features are extracted through multiple encoders of the distillation network. Each time an encoder is passed through, the height and width of the feature map are halved through max pooling. The size of the feature map obtained after passing through the last encoder is also . Then, this feature map and the features Figure 1 extracted by the backbone network are input into the Adaptive Instance Normalization (AdaIn) module for feature fusion. The features of image are adjusted using the mutual defocus distance features to make it have the statistical attributes of image . The calculation formula is as follows:

[0026]

[0027] where, represents the feature map of the input image , represents the average value of , represents 's standard deviation, represents the average value of the input mutual defocus distance feature , represents the input mutual defocus distance feature 's standard deviation.

[0028] Utilize the complementary information of the features extracted by the distillation network encoder and the features obtained from the ResNet of the backbone network, aiming to learn the information of the implicit defocus distance;

[0029] In the decoder stage, the feature map is gradually restored to the original image resolution size, and finally added to the original image to obtain the final output image .

[0030] The method for optimizing the parameters of the deep neural network model is as follows:

[0031] Construct a combined loss function through a probability distribution loss function and a distillation loss function , where the probability distribution loss function uses the KL divergence loss function , and the distillation loss function includes a pixel loss function and a Fourier loss function , and the specific formulas are as follows:

[0032]

[0033] where are weight coefficients respectively.

[0034] The calculation method of the KL divergence loss function is as follows:

[0035] By calculating the difference between the true probability distribution and the predicted probability distribution and summing the differences in all dimensions, the KL divergence loss function is obtained , and the specific calculation formula is as follows:

[0036]

[0037] where is the defocus distance, is the predicted defocus distance value, M is the vector dimension, is the probability value, and the KL divergence loss function makes the probability distributions of the two closer;

[0038] The calculation method of the distillation loss function is as follows:

[0039] Let the output of image A after passing through the distillation network be , and the expected value is the difference between the paired images at each pixel, and after taking the absolute value of the differences of all pixels and summing them, and then dividing by the total number of pixels, the pixel loss function is obtained, and the specific calculation formula is as follows:

[0040]

[0041] where represents the total number of pixels;

[0042] After transforming and to the frequency domain through the fast Fourier transform, calculate and For the difference of each pixel in the frequency domain between and , after taking the absolute value of the differences of all pixels and summing them up, then dividing by the total number of pixels, the Fourier loss is obtained. The loss calculation formula in the frequency domain is as follows:

[0043]

[0044] wherein, represents the fast Fourier transform.

[0045] In the step S5, the pathological image to be determined is input into the optimized deep neural network model for evaluating the imaging quality of pathological images, and the defocus distance is obtained. Then, the image quality evaluation level is obtained through the label distribution function.

[0046] The method for establishing the label distribution function is as follows:

[0047] With a preset scanning step as the interval, the scanning range is evenly divided into M intervals, obtaining M + 1 anchor points. For any defocus distance, the following label probability distribution is obtained:

[0048]

[0049] wherein, is the defocus distance, is the Euclidean distance metric, is the probability value. The closer the true label is to this anchor point, the greater the probability value, and vice versa. The sum of all probabilities of the vector is 1;

[0050] After the input pathological image calculates the label probability distribution through the model, then each dimension of the inferred probability distribution vector is multiplied by the corresponding preset anchor point value and summed, so as to be converted into the defocus distance value inferred by the model.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] By using the pathological images taken by the optical microscope at multiple defocus planes, the clear and blurred image pairs are screened out, and a dataset dedicated to pathological image deblurring is constructed, ensuring the pertinence and practicability of the dataset, and providing a high-quality data basis for the subsequent training of the model.

[0053] Combined with the frequency domain features, a deep neural network model for evaluating the imaging quality of pathological images based on the frequency domain features is established. The frequency domain features can more deeply reflect the essential information of the image. Therefore, this method can more accurately identify and process the problem of grading the imaging quality of pathological images, and improve the effect of evaluating the imaging quality of pathological images.

[0054] Optimize the training process through the Adam algorithm, construct a combined loss function, and enable the model to continuously optimize and improve the performance of pathological image imaging quality assessment during the training process. This method can effectively improve the accuracy and stability of the model.

[0055] The present invention adopts an end-to-end deep learning model, extracts the frequency features of the image, improves the determination of the image imaging quality, has strong robustness and good adaptability, and is helpful for subsequent analysis and diagnosis of pathological images. Brief Description of the Drawings

[0056] Figure 1 is the flowchart of the method of the embodiment of the present invention;

[0057] Figure 2 is the fuzzy pathological image of the embodiment of the present invention;

[0058] Figure 3 is the clear pathological image of the embodiment of the present invention;

[0059] Figure 4 is the comparison of the test result indicators of the embodiment of the present invention;

[0060] Figure 5 is the structural schematic diagram of the deep neural network model of the embodiment of the present invention;

[0061] Figure 6 is the structural schematic diagram of the Conditional UNet network of the embodiment of the present invention. Detailed Embodiments

[0062] The following combines the drawings to describe in detail a specific embodiment of the present invention, but it should be understood that the protection scope of the present invention is not limited by the specific embodiment.

[0063] As Figures 1 to 6 shown, a method for evaluating the imaging quality of pathological images provided by an embodiment of the present invention includes the following steps:

[0064] S1. Construct a pathological image data set;

[0065] S2. Preprocess the pathological image data set;

[0066] S3. Construct a deep neural network model for evaluating the imaging quality of pathological images based on frequency domain features;

[0067] S4. Train and optimize the constructed deep neural network model through the preprocessed pathological image data set;

[0068] S5. Input the pathological image to be determined into the optimized deep neural network model for evaluating the imaging quality of pathological images to obtain the imaging quality evaluation level of the pathological image.

[0069] The method for constructing a pathological image dataset in S1 is as follows:

[0070] The following operations are performed on multiple samples: Obtain the pathological images of the samples taken by an optical microscope at multiple defocus planes to obtain a focus stack of the sample, and select clear pathological images and blurred pathological images from it as a training sample pair. All the training samples constitute a pathological image dataset.

[0071] The training sample pairs are randomly sampled in pairs. When sampling, a clear pathological image and another image in the focus stack of the same sample are selected to form a training sample pair. The selection is random during paired sampling, and it is not fixed to select clear pathological images and blurred pathological images as training sample pairs. When these two samples are simultaneously input into the deep neural network model for training, it can avoid the problems of model overfitting and poor generalization ability;

[0072] Among them, the screening methods for clear pathological images and blurred pathological images are as follows:

[0073] Select the image with a defocus distance of 0 as the clear pathological image, and select the image with a defocus distance greater than 4 microns as the blurred pathological image. The blurred pathological images move from a defocus distance of -28 microns to a defocus distance of +28 microns in steps of 4 microns to classify blurred pathological images of different degrees;

[0074] Then the following preprocessing is performed:

[0075] S21. Divide the pathological image dataset into a training set and a test set according to a ratio;

[0076] S22. Cut the training sample pairs in the training set into regions of size 512 with a step size of 256;

[0077] S23. Randomly select 256 regions of size 256 from each region as training samples;

[0078] S24. All the training samples are used to construct a training set, and the obtained training set is:

[0079]

[0080] Among them, is the input blurred pathological image and the clear pathological image y, represents the number of channels of the image, represents the height of the image, represents the width of the image, represents the number of blurred pathological images, represents the number of clear pathological images.

[0081] The division ratio of this embodiment is 4:1;

[0082] The acquisition of the pathological image dataset in this embodiment is specifically as follows:

[0083] A Nikon Eclipse motorized microscope with 0.75 NA and a 20X objective lens was used. The training samples were 35 research-grade human pathological sections, stained with hematoxylin and eosin (Omano OMSK-HP50). The images were acquired using a 5-megapixel color camera (Pointgrey BFS-U3-5155C-C) with a pixel size of 3.45 µm. During the autofocus process, three different illumination conditions were tested: (1) incoherent Kolner illumination condition, (2) two-plane wave partial coherence illumination, and (3) single-plane wave partial coherence illumination. During the acquisition process, a z-stack was obtained by moving the sample to 41 different defocus positions (from -10 µm to +10 µm, with a step size of 0.5 µm).

[0084] Each z-stack image with a size of 2448×2048 was downsampled by a ratio of 4 to 1 and segmented into regions with a size of 224×224, finally generating 128,699 regions.

[0085] The test set was divided into two groups. The first group included 697 images, which were processed using the same staining protocol as the training set and were called "same protocol". 1312 images were processed with different staining protocols at different positions and were called "different protocol". For all cases of "same protocol" and "different protocol", each case contained 41 images taken at different focal lengths.

[0086] In S3, the deep neural network model for evaluating the imaging quality of pathological images based on frequency domain features includes a backbone network and a distillation network. The backbone network is retained during both training and testing, and the other is the distillation network, which only functions during training;

[0087] The training method of the deep neural network model in S4 is as follows:

[0088] Input the training samples of the training set into the deep neural network model. The backbone network (ResNet) of the deep neural network model extracts features from the pathological image and the pathological image The distillation network (Conditional UNet) takes one of the pathological images as input and extracts features. After fusing the intermediate layer features extracted by the backbone network, it generates an image at another defocus distance to simulate the generation process of defocused images;

[0089] Specifically: The backbone network ResNet includes four ResBlocks. Assume that the size of the feature map extracted by the third ResBlock of the backbone network is , where C represents the number of channels, H and W represent the height and width of the feature map. Concatenate the third-layer feature maps of the image and the image to obtain a feature map with a size of . Estimate the defocus kernel of paired images in the same sample through two layers of 1×1 convolutions, extract the mutual defocus distance features between the two images, and pass the feature map to the distillation network (Conditional UNet), which only accepts one of the images or as input (in this embodiment, is taken as an example). First, extract features through the encoder Embedding. Each time passing through an encoder, the height and width of the feature map are halved through max-pooling once. The size of the feature map obtained after passing through the last encoder is also . After passing through the last encoder, max-pooling is no longer performed, so the sizes of these two parts of the feature maps are aligned. Then, input the features extracted by the encoder of the distillation network (Conditional UNet) and the third layer of the backbone network ResNet into the decoder layer of the distillation network (Conditional UNet) for feature fusion. By utilizing the complementary information of the features extracted by the encoder of the distillation network (Conditional UNet) and the features obtained from the backbone network ResNet, it aims to learn the information of the implicit defocus distance;

[0090] The specific fusion method is as follows:

[0091] Input this feature map and the features extracted by the backbone network Figure 1 into the Adaptive Instance Normalization (AdaIn) module for feature fusion, and use the mutual defocus distance features to adjust the features to make them have the statistical attributes. The calculation formula is as follows:

[0092]

[0093] Among them, represents the feature map of the input image , represents 's average value, represents 's standard deviation, represents the input mutual defocus distance feature 's average value, represents the input mutual defocus distance feature Standard deviation

[0094] In the decoder stage, the feature map is gradually restored to the size of the original image resolution and finally added to the original image to obtain the final output image , and the parameters of the deep neural network model are optimized by the pixel loss between the supervised input of another image and the generated image and the Fourier loss converted to the frequency domain, as follows

[0095] Construct a combined loss function. The loss function is divided into two parts, the probability distribution loss and the distillation loss. The probability distribution loss uses the KL divergence loss function. By calculating the difference between the true probability distribution and the predicted probability distribution and summing the differences in all dimensions, the KL divergence loss is obtained , and the specific calculation formula is as follows

[0096]

[0097] where is the defocus distance is the predicted defocus distance value, M is the vector dimension is the probability value, and the KL divergence loss makes the probability distributions of the two closer

[0098] The distillation loss is divided into two parts. One is the pixel loss. Assume that the output of image A after passing through the conditional UNet network is , and the expected value is the difference between the paired image at each pixel. After taking the absolute value of the differences of all pixels and summing them, and then dividing by the total number of pixels, the pixel loss is obtained. The specific calculation formula is as follows

[0099]

[0100] where represents the total number of pixels. The other part is the Fourier loss. We transform and to the frequency domain through the fast Fourier transform, calculate the difference between each pixel in their frequency domains, take the absolute value of the differences of all pixels and sum them, and then divide by the total number of pixels to obtain the Fourier loss. The loss calculation formula in the frequency domain is as follows

[0101]

[0102] where represents the fast Fourier transform

[0103] During the training process, the parameters of the backbone network are optimized by supervising the losses between the synthesized images and the real comparison images in the spatial domain and the frequency domain. The total loss function for pathological image focal length estimation includes probability distribution loss, pixel loss, and Fourier loss. The formula for the total loss function is as follows:

[0104]

[0105] Where, are the weight coefficients respectively;

[0106] The network model code is implemented based on Pytorch 1.12. Model training is performed using 1 NVIDIA GeForce RTX 3090 GPU. First, it is pre-trained on the Incoherent dataset for 30 epochs with a batch size of 8 and an initial learning rate of 1e-5. Then, using the self-owned dataset, it is trained for 40 epochs. For model optimization, the Adam optimizer is used. If the loss does not decrease for 10 consecutive epochs during training, it is reduced to 0.1 times the original. The data augmentation strategy of random rotation is used, and the input images are cropped to a size of 224*224 and fed into the model for training.

[0107] In S5, the images in the test set are input into the optimized deep neural network model in step S4 to obtain the defocus distance, and the image quality evaluation level is obtained through the label distribution function;

[0108] Among them, the method for establishing the label distribution function is as follows:

[0109] With a preset scanning step as the spacing, the scanning range is evenly divided into M intervals to obtain M + 1 anchor points. For any defocus distance, the following label probability distribution is obtained:

[0110]

[0111] Where, is the defocus distance, is the Euclidean distance metric, is the probability value. The closer the true label is to this anchor point, the greater the probability value, and vice versa. The sum of all probabilities of the vector is 1. After the input image calculates this probability distribution through the model, each dimension of the inferred probability distribution vector is multiplied by the corresponding preset anchor point value and summed, which is converted into the defocus distance value inferred by the model.

[0112] The present invention purifies the features learned by the backbone network through feature distillation. By transmitting the defocus distance features learned by the backbone network to the distillation network, after the distillation network fuses the image features and the defocus distance features, it converts the input image into a paired image and calculates the pixel loss in the spatial domain and the frequency domain, providing more accurate and refined defocus distance features. The mean absolute error is reduced from 6.86 microns to 6.41 microns. In addition, through model pre-training, the feature extraction ability of the model for sparse distribution datasets is enhanced, significantly improving the performance of the model. The mean absolute error of defocus distance determination is reduced from 6.41 microns to 1.45 microns.

[0113] Compared with the SOTA model KDAF network, this method has a smaller mean absolute error and better performance.

[0114] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit and basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0115] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for evaluating the imaging quality of a pathological image, characterized in that: The steps include: S1, build pathological image dataset; S2, preprocessing the pathological image dataset; S3. Construct a deep neural network model for pathological image quality assessment based on frequency domain features; S4, training and optimizing the constructed deep neural network model through the preprocessed pathological image dataset; S5, inputting the pathological image to be judged into the optimized pathological image imaging quality assessment deep neural network model to obtain the imaging quality assessment grade of the pathological image; The deep neural network model training and optimization method in S4 is as follows: The training sample pairs of the training set are input into the deep neural network model, and the backbone network of the deep neural network model is used to train the pathological images. and pathological images For feature extraction, the distillation network of the deep neural network model is based on one of the pathological images As input, feature extraction is performed, and after fusing the intermediate layer features extracted by the backbone network, an image of another defocus distance is generated. , in order to simulate the generation process of defocused images; Then by creating another input image And generate images The loss function is used to optimize the parameters of the deep neural network model; The method to optimize the parameters of the deep neural network model is as follows: Construct a combined loss function by using the probability distribution loss function and the distillation loss function , where the probability distribution loss function uses the KL divergence loss function , the distillation loss function includes the pixel loss function And the Fourier loss function , the specific formula is as follows: in, are weight coefficients respectively.

2. A pathological image quality assessment method as claimed in claim 1, characterized in that: The method for constructing the pathological image dataset in S1 is as follows: Obtaining a sample pathological image taken by an optical microscope at multiple defocus planes, obtaining a focus stack of the sample, and screening out a clear pathological image and a blurred pathological image as a training sample pair, and all training samples constitute a pathological image dataset; The training sample pairs are randomly sampled in pairs. During sampling, a clear pathological image and another image in the same sample focus stack are selected to form a training sample pair.

3. A pathological image quality assessment method as claimed in claim 2, characterized in that: The screening method of the clear pathological image and the fuzzy pathological image is as follows: The image with a defocus distance of 0 was selected as a clear pathological image, and the image with a defocus distance greater than 4 μm was selected as a blurred pathological image.

4. A pathological image quality assessment method as claimed in claim 2, characterized in that: The steps of preprocessing the pathological image dataset in S2 are as follows: S21, dividing the pathological image dataset into a training set and a test set according to a ratio; S22, cut the training sample pairs in the training set into 512 512-size region with a step size of 256; S23, randomly select 256 from each area The 256 regions are used as training samples; S24. All training samples construct a training set, and the obtained training set is: in, The input fuzzy pathological image and clear pathological image y, Indicates the number of channels of the image, Indicates the height of the image. Indicates the width of the image. represents the number of fuzzy pathological images, Indicates the number of clear pathology images.

5. A pathological image quality assessment method as claimed in claim 1, characterized in that: The backbone network of the deep neural network model includes four layers of Reslock. Assume that the feature map size extracted by the third layer ResBlock of the backbone network is , where C represents the number of channels, H and W represent the height and width of the feature map, and the image and images The third layer feature maps of are spliced ​​together to obtain a feature map size of , after two layers of 1*1 convolution, the defocus kernel of the paired images in the same sample is estimated, the mutual defocus distance features between the two images are extracted, and the feature map is passed to the distillation network. The distillation network takes one of the images as input and first extracts features through the multi-layer encoder of the distillation network. After each layer of encoder, the height and width of the feature map are halved through a maximum pooling. The size of the feature map obtained after the last layer of encoder is also Then the feature map and the feature map extracted by the backbone network are input into the adaptive instance normalization AdaIn module for feature fusion, and the image is adjusted using the mutual defocus distance feature. features, making it have an image The statistical properties of are calculated as follows: in, Represents the input image The feature map of express The average value of express The standard deviation of Represents the mutual defocus distance characteristics of the input The average value of Represents the mutual defocus distance characteristics of the input The standard deviation of The complementary information of the features extracted by the distillation network encoder and the features obtained from the backbone network ResNet is used to learn the implicit defocus distance information; In the decoder stage, the feature map is gradually restored to the original image resolution and finally added to the original image to obtain the final output image. .

6. A method for evaluating the imaging quality of a pathological image as claimed in claim 5, characterized in that: The KL divergence loss function The calculation method is as follows: The KL divergence loss function is obtained by calculating the difference between the true probability distribution and the predicted probability distribution and summing the differences in all dimensions. , the specific calculation formula is as follows: in, is the defocus distance, To predict the defocus distance value, M is the vector dimension, is the probability value, and the KL divergence loss function is used to make the probability distributions of the two closer; The calculation method of the distillation loss function is as follows: Suppose the output of image A after the distillation network is , the expected value is the paired image The difference at each pixel, and the absolute value of the difference of all pixels is summed up, and then divided by the total number of pixels to get the pixel loss function , the specific calculation formula is as follows: in, Indicates the total number of pixels; Will and After fast Fourier transform to frequency domain, calculate and The difference of each pixel in the frequency domain is calculated, and the absolute value of the difference of all pixels is taken and summed, and then divided by the total number of pixels to obtain the Fourier loss. The loss calculation formula in the frequency domain is as follows: in, stands for Fast Fourier Transform.

7. A pathological image quality assessment method as claimed in claim 1, characterized in that: In S5, the pathological image to be judged is input into the optimized pathological image imaging quality assessment deep neural network model to obtain the defocus distance, and the image quality assessment grade is obtained through the label distribution function.

8. A pathological image quality assessment method as claimed in claim 7, characterized in that: The label distribution function is established as follows: The scanning range is evenly divided into M intervals with the preset scanning step length as the interval, and M+1 anchor points are obtained. For any defocus distance, the following label probability distribution is obtained: in, is the defocus distance, is the Euclidean distance metric, is the probability value. The closer the true label is to this anchor point, the greater the probability value is, and vice versa. The sum of all probabilities of the vector is 1. After the input pathological image is labeled and distributed through the model, each dimension of the inferred probability distribution vector is multiplied by the corresponding preset anchor point value and summed up to convert it into the defocus distance value inferred by the model.

Images