Intelligent digital image processing system for urinary surgery

Abstract

The invention relates to an intelligent digital image processing system for urinary surgery. The system comprises an image receiving module used for receiving a plurality of original medical images; the preprocessing module is used for preprocessing the original medical image to obtain a standardized image data set; the image segmentation module is used for carrying out image segmentation on the standardized image data set according to a U-Net + + model to obtain a lesion image; and the focus classification module is used for classifying the focus image through a DenseNet classification network to obtain an auxiliary diagnosis result. The system can assist medical personnel in diagnosis, reduce the dependence of diagnosis on artificial experience, and liberate manpower.

Description

Technical Field

[0001] The present invention belongs to the field of image processing, and in particular relates to an intelligent digital image processing system. Background Art

[0002] With the development of medical diagnostic technology, medical image processing technology has emerged. Traditional medical image processing technology uses a simple threshold method to confirm lesions in urology, and requires manual feature extraction when classifying lesions. It has the disadvantages of being extremely dependent on manual experience, wasting a lot of manpower, and having a low judgment rate for complex lesions. Summary of the Invention

[0003] Based on this, it is necessary to address the above technical issues and provide an intelligent digital image processing system for urology that can effectively reduce the diagnosis's dependence on manual experience and liberate manpower.

[0004] In a first aspect, the present application provides an intelligent digital image processing system for urology, comprising:

[0005] An image receiving module, used for receiving a variety of original medical images;

[0006] A preprocessing module is used to preprocess the original medical images to obtain a standardized image dataset;

[0007] The image segmentation module is used to perform image segmentation on the standardized image dataset based on the U-Net++ model to obtain lesion images;

[0008] The lesion classification module is used to classify lesion images through the DenseNet classification network to obtain auxiliary diagnosis results.

[0009] Furthermore, the original medical images are preprocessed to obtain a standardized image dataset, including:

[0010] The contrast of the original medical image is enhanced by using a histogram equalization algorithm to obtain a contrast-enhanced medical image;

[0011] The contrast-enhanced medical image is subjected to noise suppression using the non-local mean denoising algorithm to obtain a denoised grayscale image.

[0012] Perform histogram matching and affine transformation registration on a set of denoised grayscale images to obtain a fused image with consistent space and grayscale.

[0013] The fused images are elastically deformed and randomly rotated to obtain a set of deformed fused images; the deformed fused images are determined as a standardized image dataset.

[0014] Furthermore, the U-Net++ model performs image segmentation on the standardized image dataset to obtain lesion images using the following method:

[0015] Local features of the standardized image dataset are gradually extracted through multiple encoders;

[0016] The local features are further processed by setting an additional convolutional layer at the end of the encoder to obtain more compact terminal features;

[0017] The terminal features are input into the decoder; each decoder receives the upsampling results of the upper decoder, the features of the encoder at the same level, and the features of the previous node in the same path; after processing by the convolution block, the features are mapped back to the original image size and the segmentation result is generated to obtain the lesion image;

[0018] Among them, let the encoder layer i input be x i,0 , the input of the decoder layer j is confirmed by the following formula:

[0019] x i,j =H(Concat([x i,0 ,Upsample(x i+1,j-1 )])(j>0);

[0020] Among them, H() is the convolution and activation function operation;

[0021] Among them, an auxiliary loss function is added to each decoder layer, and the auxiliary loss function is used to optimize the model.

[0022] Furthermore, the DenseNet classification network classifies lesion images through the following methods to obtain auxiliary diagnosis results:

[0023] Extracting multimodal non-image features from standardized image datasets; multimodal non-image features include imaging features and clinical parameters;

[0024] Input the lesion image into the densely connected block to obtain the corresponding visual feature vector;

[0025] The input of each layer of the densely connected block is the concatenation of the outputs of all previous layers. Let the input of the lth layer be x0, and the output of the lth layer is determined by the following formula:

[0026] x l =H l ([x0,x1,……x l-1 ])

[0027] Among them, H l () is a composite operation, which compresses the channel through 1×1 convolution and extracts features through 3×3 convolution;

[0028] The multimodal non-image features are input into the fully connected layer to obtain a multimodal non-image feature vector aligned with the visual feature vector;

[0029] Concatenate the visual feature vector and the multimodal non-image feature vector to obtain a fusion vector;

[0030] Input the fusion vector into the classification layer to obtain the classification result and the corresponding probability distribution;

[0031] Based on the probability distribution of the classification results, the disease with the highest probability is determined as the disease entity;

[0032] Determine the auxiliary diagnosis results based on the disease entity.

[0033] Furthermore, the DenseNet classification network is updated by the following method:

[0034] Perform cross entropy calculation on the classification results and the corresponding probability distribution to obtain the loss function;

[0035] Backpropagate the loss function and update the DenseNet classification network.

[0036] Furthermore, the system also includes a three-dimensional modeling module for modeling the lesion based on the original medical image and the lesion image:

[0037] Perform interlayer interpolation on the original medical image to obtain slices of uniform thickness;

[0038] Surface reconstruction is performed on slices and lesion images to obtain a lesion model.

[0039] Furthermore, after surface reconstruction of the slice and lesion images is performed to obtain a lesion model, the following steps are further included:

[0040] Based on the auxiliary diagnosis results, the historical treatment methods of the corresponding diseases are matched in the case database and the historical treatment methods are determined as the treatment plan;

[0041] Calculate the size and shape of the lesion based on the lesion model and adjust the treatment plan.

[0042] In a second aspect, the present application further provides an intelligent digital image processing method for urology, comprising:

[0043] Receive multiple original medical images;

[0044] Preprocess the original medical images to obtain a standardized image dataset;

[0045] Perform image segmentation on the standardized image dataset using the U-Net++ model to obtain lesion images;

[0046] The lesion images are classified through the DenseNet classification network to obtain auxiliary diagnosis results.

[0047] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the system provided in the first aspect of the present application when executing the computer program.

[0048] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the system provided in the first aspect of the present application.

[0049] The intelligent digital image processing system for urology receives multiple raw medical images through an image receiving module; preprocesses the raw medical images through a preprocessing module to obtain a standardized image dataset; uses an image segmentation module to segment the standardized image dataset using a U-Net++ model to obtain lesion images; and uses a lesion classification module to classify lesion images using a DenseNet classification network to obtain auxiliary diagnosis results. This effectively solves the problem of existing traditional medical image processing technology in urology, which relies on manual experience and wastes a lot of manpower, reducing reliance on manual experience and freeing up manpower. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 This is a diagram of an intelligent digital image processing system for urology surgery according to the present invention;

[0052] Figure 2 This is a flow chart of an intelligent digital image processing method for urology surgery according to the present invention; DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0054] An intelligent digital image processing system for urology provided in an embodiment of the present application can be applied in urology to assist in disease diagnosis and screening by processing digital medical images.

[0055] In one embodiment, Figure 1 As shown, an intelligent digital image processing system 100 for urology surgery is provided. This embodiment uses the system applied to a terminal as an example. It is understood that the system can also be applied to a server, or to an environment including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the system includes the following modules:

[0056] The image receiving module 101 is used to receive a variety of original medical images.

[0057] Specifically, the multiple original medical images can be a single image or a group of images, including two-dimensional images and three-dimensional images, including CT (Computed Tomography) images, MRI (Magnetic Resonance Imaging) images, DR (Digital Radiography) images, images captured during endoscopy, etc.; and also include a variety of non-image data such as patient age, weight or image feature data.

[0058] Exemplarily, the CT image includes the HU value of each pixel.

[0059] The preprocessing module 102 is used to preprocess the original medical images to obtain a standardized image dataset.

[0060] Specifically, the preprocessing module removes noise and performs image enhancement on a set of raw medical images. This standardizes image variations caused by different equipment and unifies image size, brightness, and color space to facilitate subsequent neural network model decision-making and training. The standardized image dataset includes a set of relevant preprocessed images and corresponding non-image feature data.

[0061] The image segmentation module 103 is used to perform image segmentation on the standardized image dataset according to the U-Net++ model to obtain a lesion image.

[0062] Specifically, a lesion image is an image of a lesion area on the human body. Different diseases produce different lesion images, and the lesion areas also vary. The U-Net++ model consists of an equal number of encoders and decoders. The encoder performs downsampling, while the decoder performs upsampling. The encoder is connected to the decoder via convolutional layers at the bottom, forming a U-shape. The decoder uses nested jumps to connect the multi-scale features of each encoder layer. The encoder continuously extracts high-level semantic features, while the decoder gradually restores the image resolution. Feature maps at different levels are fused through concatenation or addition to output the lesion image.

[0063] The lesion classification module 104 is used to classify the lesion image through the DenseNet classification network to obtain auxiliary diagnosis results.

[0064] Specifically, auxiliary diagnosis results are only auxiliary to disease diagnosis and may deviate from the actual disease condition to a certain extent. The final specific diagnosis results need to be confirmed by medical staff. The DenseNet classification network processes lesion images through densely connected blocks. The input of each layer in the densely connected block is the concatenation of the outputs of all previous layers. By splicing features from different layers, gradient flow and information transfer are enhanced, alleviating the gradient vanishing problem. The final classification head of the DenseNet classification network includes a global average pooling layer and a fully connected layer, outputting a probability distribution.

[0065] The embodiment of the present application provides an intelligent digital image processing system for urology, comprising an image receiving module for receiving a variety of original medical images; a preprocessing module for preprocessing the original medical images to obtain a standardized image dataset; an image segmentation module for performing image segmentation on the standardized image dataset according to the U-Net++ model to obtain lesion images; and a lesion classification module for classifying lesion images using the DenseNet classification network to obtain auxiliary diagnosis results. The above modules can effectively alleviate the problems of relying on manual experience and wasting human resources in traditional urology digital image processing systems, reduce the workload of medical staff, and free up manpower.

[0066] In one embodiment, preprocessing the original medical images to obtain a standardized image dataset includes:

[0067] Step 201 : performing contrast enhancement on the original medical image using a histogram equalization algorithm to obtain a contrast-enhanced medical image.

[0068] Specifically, histogram equalization is used to enhance the contrast of an image, making the distribution of pixels at various brightness levels in the image more even, highlighting details, and better distinguishing the texture of various organs in the medical field.

[0069] Step 202 : noise reduction is performed on the contrast-enhanced medical image using a non-local means denoising algorithm to obtain a denoised grayscale image.

[0070] Specifically, non-local means denoising is to search for similar blocks in a certain area around each pixel, calculate the weight of the block affecting the pixel based on the distance between the block and the pixel, normalize the weight and take the weighted average to obtain the denoised value of the pixel.

[0071] For example, if the image is a color image, weighted averaging is performed on the image to obtain a denoised grayscale image.

[0072] Step 203 : performing histogram matching and affine transformation registration on a set of denoised grayscale images to obtain a fused image with consistent space and grayscale.

[0073] Specifically, histogram matching adjusts images captured under different lighting conditions to have similar brightness distributions. Using the target histogram to perform grayscale mapping on the original image can standardize brightness differences in medical images caused by different camera equipment. Affine transformation registration spatially fuses these denoised grayscale images to address translational and rotational differences between images caused by varying camera equipment, camera angles, and patient positions.

[0074] In step 204 , the fused image is elastically deformed and randomly rotated to obtain a set of deformed fused images; the deformed fused images are determined as a standardized image dataset.

[0075] Specifically, elastic deformation simulates nonlinear deformations in the real world, improving the generalization ability of deep learning models to deformations. Random rotation generates new images by rotating an image, reducing the deep learning model's reliance on specific angles. Using both together can increase the number of samples and address sample imbalance.

[0076] For example, random rotation solves the distribution deviation caused by acquisition direction in urology, such as left and right X-rays.

[0077] This embodiment preprocesses the original medical images to generate a standardized image dataset, which can improve data quality, reduce the difficulty of the next step of model training decision-making, standardize the output format, and enhance the applicability of the system.

[0078] In one embodiment, the U-Net++ model performs image segmentation on a standardized image dataset to obtain a lesion image by:

[0079] Step 301 : Local features of a standardized image dataset are gradually extracted through multiple encoders.

[0080] Specifically, each encoder has two convolutional layers and one pooling layer. The encoder downsamples the image to gradually extract high-level semantic features, sacrificing spatial resolution to capture global information. The number of convolutional output channels gradually doubles, and after the pooling operation, the spatial resolution of the feature map decreases, but the number of channels increases.

[0081] In step 302, the local features are further processed by an additional convolutional layer provided at the end of the encoder to obtain more compact terminal features.

[0082] Specifically, the final output of the encoder is convolved twice without pooling to maintain the same output size.

[0083] In step 303, the terminal features are input into the decoder; each decoder receives the upsampling result of the upper decoder, the features of the encoder at the same level, and the features of the previous node in the same path; after processing by the convolution block, the features are mapped back to the original image size, and the segmentation result is generated to obtain the lesion image.

[0084] Specifically, the decoder gradually restores spatial resolution through upsampling and combines low-level features from skip connections to achieve pixel-level localization. Each decoder doubles the size of the feature map through deconvolution or interpolation. The feature map from the corresponding encoder stage is concatenated with the upsampling result channel by channel through a multi-stage skip path. The concatenated feature map undergoes two convolutions, gradually halving the number of channels. The final output is a binary mask of the lesion.

[0085] Exemplarily, the multi-stage skip path is the path from the encoder layer 3 to the decoder layer 3 as follows: the encoder layer 3 is connected to the decoder layer 3 through a 1st convolution block, the encoder layer 2 is connected to the encoder layer 3 through upsampling and a 1st convolution block, and the encoder layer 1 is connected to the encoder layer 3 through upsampling and a 2nd convolution block.

[0086] Among them, let the encoder layer i input be x i,0 , the input of the decoder layer j is confirmed by the following formula:

[0087] x i,j =H(Concat([x i,0 ,Upsample(x i+1,j-1 )])(j>0);

[0088] Among them, H() is the convolution and activation function operation;

[0089] Among them, an auxiliary loss function is added to each decoder layer, and the auxiliary loss function is used to optimize the model.

[0090] This embodiment introduces the U-Net++ model to segment images, solving the problem of manual division of lesion areas and reliance on human experience in traditional image processing. The lesion areas are divided by machine learning, reducing the workload of medical staff and freeing up manpower.

[0091] In one embodiment, the DenseNet classification network classifies lesion images to obtain auxiliary diagnosis results using the following method:

[0092] Step 401 : extracting multimodal non-image features from a standardized image dataset; the multimodal non-image features include imaging features and clinical parameters.

[0093] Specifically, image features include the grayscale mean, texture features, shape parameters, etc. of the lesion area, and clinical parameters include patient structured data such as age, gender, blood pressure, and medical history.

[0094] Step 402: Input the lesion image into the densely connected block to obtain the corresponding visual feature vector.

[0095] The input of each layer of the densely connected block is the concatenation of the outputs of all previous layers. Let the input of the lth layer be x0, and the output of the lth layer is determined by the following formula:

[0096] x l =H l ([x0,x1,……x l-1 ])

[0097] Among them, H l () is a composite operation, in which 1×1 convolution compresses the channel and 3×3 convolution extracts features.

[0098] Specifically, the composite operation H l This includes channel compression, using 1×1 convolution to reduce the number of channels and lower computational effort; feature extraction, using 3×3 convolution to extract spatial features while maintaining resolution; and finally, RuLU nonlinear activation to output a feature map. Global average pooling is performed on the last layer of feature maps to generate a fixed-length visual feature vector.

[0099] Step 403: Input the multimodal non-image features into the fully connected layer to obtain a multimodal non-image feature vector aligned with the visual feature vector.

[0100] Step 404 : Concatenate the visual feature vector and the multimodal non-image feature vector to obtain a fusion vector.

[0101] Specifically, the visual feature vector and the multimodal non-image feature vector are spliced ​​according to the channel dimension, and the dimension of the fusion vector is the sum of the dimensions of the visual feature vector and the multimodal non-image feature vector.

[0102] Step 405: Input the fusion vector into the classification layer to obtain the classification result and the corresponding probability distribution.

[0103] Specifically, the probability distribution is generated through the fully connected layer and the Softmax activation function.

[0104] Step 406: Based on the probability distribution of the classification results, determine the disease with the highest probability as the disease entity.

[0105] Step 407: Determine the auxiliary diagnosis result based on the disease entity.

[0106] This embodiment introduces a neural network to assist in the diagnosis of lesions, which reduces the workload of medical staff, increases the reliability of the judgment of complex lesions, reduces the dependence on manual experience, and liberates manpower.

[0107] In one embodiment, the DenseNet classification network is updated by:

[0108] The cross entropy of the classification results and the corresponding probability distribution is calculated to obtain the loss function.

[0109] Backpropagate the loss function and update the DenseNet classification network.

[0110] This embodiment introduces a loss function to maintain the growth of the DenseNet classification network, reduce the possibility of errors in the auxiliary diagnosis results of this system, and improve the accuracy of the system.

[0111] In one embodiment, the system further includes a three-dimensional modeling module for modeling the lesion based on the original medical image and the lesion image:

[0112] Perform interlayer interpolation on the original medical image to obtain slices of uniform thickness.

[0113] Specifically, slices of different thicknesses in the original medical image from different devices are interpolated to generate intermediate layers, which can make all slices have uniform thickness, solve the problem of uneven layer spacing in the original image, and provide the possibility for subsequent reconstruction.

[0114] Surface reconstruction is performed on slices and lesion images to obtain a lesion model.

[0115] This embodiment provides a lesion modeling process that helps patients more intuitively understand their condition through the model, easing communication difficulties between doctors and patients. The lesion model provides an intuitive solution for subsequent preoperative planning and quantitative analysis, facilitating medical staff's analysis of the condition.

[0116] In one embodiment, after performing surface reconstruction on the slice and lesion images to obtain the lesion model, the method further includes:

[0117] Based on the auxiliary diagnosis results, the historical treatment methods of the corresponding diseases are matched in the case database, and the historical treatment methods are determined as the treatment plan.

[0118] Specifically, the case database contains data such as the routine treatment process, efficacy, and patient's physiological condition of the corresponding disease.

[0119] Calculate the size and shape of the lesion based on the lesion model and adjust the treatment plan.

[0120] Specifically, in general, different lesion sizes require different treatment plans, and different lesion shapes also require different treatment plans. The specific situation will be analyzed and judged by medical staff. This treatment plan is for reference only.

[0121] This embodiment provides a treatment plan that can be used for reference by matching historical treatment methods, providing certain suggestions for medical staff for subsequent treatment, reducing the workload of medical staff and freeing up manpower.

[0122] The intelligent digital image processing system for urology surgery comprises an image receiving module for receiving a variety of raw medical images; a preprocessing module for preprocessing the raw medical images to obtain a standardized image dataset; an image segmentation module for performing image segmentation on the standardized image dataset using a U-Net++ model to obtain lesion images; and a lesion classification module for classifying lesion images using a DenseNet classification network to obtain auxiliary diagnosis results. These modules can effectively alleviate the reliance on manual experience and waste of human resources in traditional urology digital image processing systems, reducing the workload of medical staff and freeing up manpower.

[0123] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0124] Based on the same inventive concept, the present application also provides an intelligent digital image processing method for urology surgery, which is used to implement the aforementioned intelligent digital image processing system for urology surgery. The solution provided by this method is similar to the solution described in the aforementioned system. Therefore, the specific limitations of one or more embodiments of the intelligent digital image processing method for urology surgery provided below can be found in the above-mentioned limitations of the intelligent digital image processing system for urology surgery, and will not be repeated here.

[0125] In an exemplary embodiment, Figure 2 As shown, an intelligent digital image processing method for urology is provided, comprising:

[0126] Step 501, receiving a plurality of original medical images;

[0127] Step 502: pre-processing the original medical image to obtain a standardized image dataset;

[0128] Step 503: performing image segmentation on the standardized image dataset according to the U-Net++ model to obtain a lesion image;

[0129] Step 504: classify the lesion image through the DenseNet classification network to obtain an auxiliary diagnosis result.

[0130] Furthermore, the original medical images are preprocessed to obtain a standardized image dataset, including:

[0131] The contrast of the original medical image is enhanced by using a histogram equalization algorithm to obtain a contrast-enhanced medical image;

[0132] The contrast-enhanced medical image is subjected to noise suppression using the non-local mean denoising algorithm to obtain a denoised grayscale image.

[0133] Perform histogram matching and affine transformation registration on a set of denoised grayscale images to obtain a fused image with consistent space and grayscale.

[0134] The fused images are elastically deformed and randomly rotated to obtain a set of deformed fused images; the deformed fused images are determined as a standardized image dataset.

[0135] Furthermore, the U-Net++ model performs image segmentation on the standardized image dataset to obtain lesion images using the following method:

[0136] Local features of the standardized image dataset are gradually extracted through multiple encoders;

[0137] The local features are further processed by setting an additional convolutional layer at the end of the encoder to obtain more compact terminal features;

[0138] The terminal features are input into the decoder; each decoder receives the upsampling results of the upper decoder, the features of the encoder at the same level, and the features of the previous node in the same path; after processing by the convolution block, the features are mapped back to the original image size and the segmentation result is generated to obtain the lesion image;

[0139] Among them, let the encoder layer i input be x i,0 , the input of the decoder layer j is confirmed by the following formula:

[0140] x i,j =H(Concat([x i,0 ,Upsample(x i+1,j-1)])(j>0);

[0141] Among them, H() is the convolution and activation function operation;

[0142] Among them, an auxiliary loss function is added to each decoder layer, and the auxiliary loss function is used to optimize the model.

[0143] Furthermore, the DenseNet classification network classifies lesion images through the following methods to obtain auxiliary diagnosis results:

[0144] Extracting multimodal non-image features from standardized image datasets; multimodal non-image features include imaging features and clinical parameters;

[0145] Input the lesion image into the densely connected block to obtain the corresponding visual feature vector;

[0146] The input of each layer of the densely connected block is the concatenation of the outputs of all previous layers. Let the input of the lth layer be x0, and the output of the lth layer is determined by the following formula:

[0147] x l =H l ([x0,x1,……x l-1 ])

[0148] Among them, H l () is a composite operation, which compresses the channel through 1×1 convolution and extracts features through 3×3 convolution;

[0149] The multimodal non-image features are input into the fully connected layer to obtain a multimodal non-image feature vector aligned with the visual feature vector;

[0150] Concatenate the visual feature vector and the multimodal non-image feature vector to obtain a fusion vector;

[0151] Input the fusion vector into the classification layer to obtain the classification result and the corresponding probability distribution;

[0152] Based on the probability distribution of the classification results, the disease with the highest probability is determined as the disease entity;

[0153] Determine the auxiliary diagnosis results based on the disease entity.

[0154] Furthermore, the DenseNet classification network is updated by the following method:

[0155] Perform cross entropy calculation on the classification results and the corresponding probability distribution to obtain the loss function;

[0156] Backpropagate the loss function and update the DenseNet classification network.

[0157] Furthermore, the method further comprises:

[0158] Modeling the lesion based on the original medical image and the lesion image;

[0159] Perform interlayer interpolation on the original medical image to obtain slices of uniform thickness;

[0160] Surface reconstruction is performed on slices and lesion images to obtain a lesion model.

[0161] Furthermore, after surface reconstruction of the slice and lesion images is performed to obtain a lesion model, the following steps are further included:

[0162] Based on the auxiliary diagnosis results, the historical treatment methods of the corresponding diseases are matched in the case database and the historical treatment methods are determined as the treatment plan;

[0163] Calculate the size and shape of the lesion based on the lesion model and adjust the treatment plan.

[0164] In one embodiment, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the aforementioned intelligent digital image processing system for urology when executing the computer program.

[0165] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned system embodiments are implemented.

[0166] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0167] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. An intelligent digital image processing system for urology, characterized in that: The system comprises: An image receiving module, used for receiving a variety of original medical images; A preprocessing module, configured to preprocess the original medical images to obtain a standardized image dataset; An image segmentation module is used to perform image segmentation on the standardized image dataset according to a U-Net++ model to obtain a lesion image; The lesion classification module is used to classify the lesion image through the DenseNet classification network to obtain auxiliary diagnosis results.

2. The intelligent digital image processing system for urology according to claim 1, characterized in that: The preprocessing of the original medical images to obtain a standardized image dataset includes: performing contrast enhancement on the original medical image using a histogram equalization algorithm to obtain a contrast-enhanced medical image; performing noise suppression on the contrast-enhanced medical image using a non-local means denoising algorithm to obtain a denoised grayscale image; Performing histogram matching and affine transformation registration on a set of the denoised grayscale images to obtain a fused image with consistent space and grayscale; The fused images are elastically deformed and randomly rotated to obtain a set of deformed fused images; and the deformed fused images are determined as the standardized image data set.

3. The intelligent digital image processing system for urology according to claim 1, characterized in that: The U-Net++ model performs image segmentation on the standardized image dataset to obtain a lesion image by the following method: gradually extracting local features of the standardized image dataset through multiple encoders; The local features are further processed by an additional convolutional layer set at the end of the encoder to obtain more compact terminal features; The terminal features are input into the decoder; each decoder receives the upsampling results of the upper decoder, the features of the encoder at the same level, and the features of the previous node in the same path; after processing by the convolution block, the features are mapped back to the original image size and a segmentation result is generated to obtain a lesion image; Among them, let the encoder layer i input be x i,0 , the input of the decoder layer j is confirmed by the following formula: x i,j =H(Concat([x i,0 ,Upsample(x i+1,j-1 )])(j>0); Among them, H() is the convolution and activation function operation; Wherein, an auxiliary loss function is added to each level of the decoder, and the auxiliary loss function is used to optimize the model.

4. The intelligent digital image processing system for urology according to claim 1, characterized in that: The DenseNet classification network classifies the lesion image by the following method to obtain auxiliary diagnosis results: Extracting multimodal non-image features from the standardized image dataset; the multimodal non-image features include imaging features and clinical parameters; Inputting the lesion image into a densely connected block to obtain a corresponding visual feature vector; The input of each layer of the densely connected block is the concatenation of the outputs of all previous layers. Let the input of the lth layer be x0, and the output of the lth layer be determined by the following formula: x l =H l ([x0,x1,……x l-1 ]) Among them, H l () is a composite operation, which compresses the channel through 1×1 convolution and extracts features through 3×3 convolution; Inputting the multimodal non-image features into a fully connected layer to obtain a multimodal non-image feature vector aligned with the visual feature vector; splicing the visual feature vector and the multimodal non-image feature vector to obtain a fusion vector; Input the fusion vector into the classification layer to obtain the classification result and the corresponding probability distribution; Based on the probability distribution of the classification results, determining the disease with the highest probability as the disease entity; The auxiliary diagnosis result is determined based on the disease entity.

5. The intelligent digital image processing system for urology according to claim 4, characterized in that: The DenseNet classification network is updated by the following method: Performing cross entropy calculation on the classification results and the corresponding probability distribution to obtain a loss function; Back-propagate the loss function and update the DenseNet classification network.

6. An intelligent digital image processing system for urology according to any one of claims 1 to 5, characterized in that: The system further includes a three-dimensional modeling module for modeling the lesion based on the original medical image and the lesion image: performing inter-layer interpolation on the original medical image to obtain slices of uniform thickness; Surface reconstruction is performed on the slice and the lesion image to obtain a lesion model.

7. The intelligent digital image processing system for urology according to claim 6, characterized in that: After performing surface reconstruction on the slice and the lesion image to obtain a lesion model, the method further includes: Based on the auxiliary diagnosis results, matching the historical treatment methods of the corresponding disease in the case database, and determining the historical treatment methods as the treatment plan; The size and shape of the lesion are calculated based on the lesion model, and the treatment plan is adjusted.

8. An intelligent digital image processing method for urology, characterized in that: The method comprises: Receive multiple original medical images; Preprocessing the original medical images to obtain a standardized image dataset; Performing image segmentation on the standardized image dataset according to the U-Net++ model to obtain a lesion image; The lesion image is classified by the DenseNet classification network to obtain an auxiliary diagnosis result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the system according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the system according to any one of claims 1 to 7 are implemented.

Images