Pancreatic necrotic tissue volume quantification method and system based on enhanced CT images

Abstract

This invention provides a method and system for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, belonging to the field of medical image processing technology. The method involves constructing and training a deep convolutional neural network model to automatically segment enhanced abdominal venous phase CT images, outputting a pixel-level probability map of peripancreatic necrotic tissue. The probability map is then converted into a three-dimensional segmentation mask using image morphology processing algorithms. The volume of peripancreatic necrotic tissue is calculated based on the number of voxels and the volume of a single voxel in the three-dimensional segmentation mask. This invention solves the problems of high subjectivity, low accuracy, and time-consuming and laborious manual layer-by-layer delineation in traditional manual estimation. It achieves full automation of the peripancreatic necrotic tissue segmentation and volume quantification process, providing an objective, accurate, and repeatable quantitative tool for assessing the severity of acute pancreatitis, selecting intervention timing, and dynamically monitoring treatment efficacy.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, and in particular relates to a method and system for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images. Background Technology

[0002] The volume of peripancreatic necrotic tissue, as an objective and precise quantitative indicator, is significantly correlated with the persistence of organ failure, the incidence of complications, and the need for intervention. Therefore, achieving accurate and rapid quantification of peripancreatic necrosis volume has significant clinical value for the severity grading of acute pancreatitis, the selection of intervention timing, and the dynamic evaluation of treatment efficacy.

[0003] Currently, enhanced CT is the core imaging tool for the clinical diagnosis of necrotizing pancreatitis and the assessment of its local complications. However, current clinical practice has significant limitations in assessing the extent of peripancreatic necrosis. The mainstream methods mainly rely on two procedures performed by radiologists: one is semi-quantitative estimation based on experience, which is fast but has low accuracy and is highly subjective; the other is manually delineating the necrotic area layer by layer, which, although considered a reference standard, is extremely time-consuming and laborious, and the results are limited by the physician's personal experience and judgment, leading to significant differences between different observers and even within the same observer at different time points. These limitations severely restrict the accuracy, reproducibility, and efficient application of peripancreatic necrosis volume assessment in clinical work and research. Summary of the Invention

[0004] This invention provides a method and system for volume quantification of peripancreatic necrotic tissue based on enhanced CT images, which solves the problems of low accuracy, strong subjectivity and time-consuming and labor-intensive traditional manual segmentation methods, and realizes accurate segmentation and volume quantification of peripancreatic necrotic tissue.

[0005] In a first aspect, the present invention provides a method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, comprising:

[0006] Construct a deep convolutional neural network model for segmenting peripancreatic necrotic tissue; Multiple abdominal venous phase enhanced CT images labeled with peripancreatic necrotic tissue were used as a training set to train a deep convolutional neural network model, and the trained deep convolutional neural network model was used as a segmentation model for peripancreatic necrotic tissue. The first image is input into the peripancreatic necrotic tissue segmentation model to output a second image of the same size as the first image; wherein, the first image is the abdominal venous phase enhanced CT image to be detected; and the probability of each pixel belonging to peripancreatic necrotic tissue is known in the second image. The second image is converted into a 3D segmentation mask using image morphology processing algorithms; The volume of peripancreatic necrotic tissue in the first image is determined based on the number of voxels and the volume of a single voxel in the 3D segmentation mask.

[0007] Optionally, the construction of the deep convolutional neural network model for segmenting peripancreatic necrotic tissue includes: Construct N sequentially connected downsampling modules; each downsampling module includes at least two sequentially connected convolutional units; each convolutional unit includes a first convolutional layer, a batch normalization layer, and a first activation function, all sequentially connected; the output of the last convolutional unit in each downsampling module is downsampled using a pooling layer; Construct N sequentially connected upsampling modules; the first upsampling module takes the output of the last downsampling module as its input; each upsampling module uses transposed convolution for upsampling; the outputs of the N upsampling modules are skipped and connected to the output of the last convolutional unit in each downsampling module and then fused. Construct an output module; wherein the output module includes a second convolutional layer and a second activation function connected in sequence; the second convolutional layer takes the output of the last upsampling module as input.

[0008] Optionally, the construction of the deep convolutional neural network model for segmenting peripancreatic necrotic tissue includes: Construct four sequentially connected downsampling modules; each downsampling module includes at least two sequentially connected convolutional units; each convolutional unit includes a sequentially connected 3×3 convolutional layer, a batch normalization layer, and a ReLU activation function; the output of the last convolutional unit in each downsampling module is downsampled using a 2×2 pooling layer; And / or, construct four sequentially connected upsampling modules; wherein the first upsampling module takes the output of the last downsampling module as input; each upsampling module uses a 2×2 transposed convolution for upsampling; the outputs of the four upsampling modules are skip-connected to and fused with the output of the last convolutional unit in each downsampling module; And / or, construct an output module; wherein the output module comprises a 1×1 convolutional layer and a sigmoid activation function connected in sequence; the 1×1 convolutional layer takes the output of the last upsampling module as its input.

[0009] Optionally, the step of using multiple abdominal venous phase enhanced CT images labeled with peripancreatic necrosis tissue as a training set to train a deep convolutional neural network model, and using the trained deep convolutional neural network model as a segmentation model for peripancreatic necrosis tissue, includes: Calculate the Dice loss L using the following formula. Dice : ; Where, p mg represents the predicted probability of a deep convolutional neural network model that the m-th pixel belongs to peripancreatic necrotic tissue. m The true label for the m-th pixel; g m =0 indicates that the m-th pixel belongs to non-peripancreatic necrotic tissue; g m =1 indicates that the m-th pixel belongs to peripancreatic necrotic tissue; M is the total number of pixels in the abdominal venous phase enhanced CT image. Calculate the cross-entropy loss L using the following formula. CE : ; The total loss is determined based on the Dice loss and cross-entropy loss. The deep convolutional neural network model is trained based on the total loss until it converges, resulting in a well-trained deep convolutional neural network model that can be used as a segmentation model for peripancreatic necrotic tissue.

[0010] Optionally, determining the total loss based on the Dice loss and cross-entropy loss includes: Calculate the total loss L using the following formula: L=αL Dice +βL CE ; Where α is the weight of the Dice loss and β is the weight of the cross-entropy loss.

[0011] Optionally, the step of converting the second image into a three-dimensional segmentation mask using an image morphology processing algorithm includes: The probability of each pixel in the second image belonging to peripancreatic necrotic tissue is compared with the probability threshold to determine whether the voxel corresponding to each pixel is assigned a value of 0 or 1, thus obtaining the binarized initial 3D segmentation mask. The initial 3D segmentation mask is processed using image morphology processing algorithms to obtain the final 3D segmentation mask.

[0012] Optionally, the step of converting the second image into a three-dimensional segmentation mask using an image morphology processing algorithm includes: In the second image, the voxels corresponding to the pixels whose probability of belonging to peripancreatic necrotic tissue is greater than or equal to the probability threshold are assigned a value of 1. The voxels corresponding to the pixels in the second image whose probability of belonging to peripancreatic necrotic tissue is less than the probability threshold are assigned a value of 0 to obtain the binarized initial 3D segmentation mask. And / or, perform morphological opening operations on the initial 3D segmentation mask to obtain the first 3D segmentation mask; Perform morphological closing operations on the first 3D segmentation mask to obtain the second 3D segmentation mask; Connectivity analysis is performed on the second 3D segmentation mask to remove isolated regions with excessively small areas, thus obtaining the final 3D segmentation mask.

[0013] Optionally, the step of performing connected component analysis on the second 3D segmentation mask to remove isolated regions with excessively small areas to obtain the final 3D segmentation mask includes: Identify all independent connected regions in the second 3D segmentation mask; Calculate the total number of voxels in each connected region; The voxels in connected regions with a total number of voxels less than the voxel threshold are assigned a value of 0 to obtain the final 3D segmentation mask.

[0014] Optionally, determining the volume corresponding to the peripancreatic necrotic tissue in the first image based on the number of voxels in the three-dimensional segmentation mask and the volume of a single voxel includes: Obtain the number and volume of voxels with a value of 1 in the 3D segmentation mask; The product of the number of voxels with a value of 1 in the 3D segmentation mask and the voxel volume is used as the volume corresponding to the peripancreatic necrotic tissue in the first image.

[0015] Secondly, the present invention provides a system for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, characterized in that it comprises: Build modules are used to construct deep convolutional neural network models for segmenting peripancreatic necrotic tissue; The training module is used to train a deep convolutional neural network model using multiple abdominal venous phase enhanced CT images labeled with peripancreatic necrotic tissue as the training set, and to use the trained deep convolutional neural network model as a segmentation model for peripancreatic necrotic tissue. The input module does not input the first image into the peripancreatic necrotic tissue segmentation model, so as to output a second image with the same size as the first image; wherein, the first image is the abdominal venous phase enhanced CT image to be detected; and the probability of each pixel belonging to peripancreatic necrotic tissue is known in the second image. The conversion module is used to convert the second image into a 3D segmentation mask using image morphology processing algorithms. The determination module is used to determine the volume corresponding to the peripancreatic necrotic tissue in the first image based on the number of voxels and the volume of a single voxel in the 3D segmentation mask.

[0016] This invention provides a method and system for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images. The method involves constructing and training a deep convolutional neural network model for segmenting peripancreatic necrotic tissue. Addressing the issues of high subjectivity and low accuracy in manual visual estimation in the prior art, this invention transforms manual annotation experience into computable network parameters by constructing a deep learning model. The model can automatically learn the complex imaging features of peripancreatic necrotic tissue in CT images (such as density, morphology, boundaries, and relationship with surrounding tissues), thereby achieving accurate and automated identification of the necrotic region. This fundamentally avoids the subjectivity of manual visual estimation, providing a reliable foundation for subsequent accurate quantification.

[0017] This invention inputs a contrast-enhanced CT image of the abdomen in the venous phase (first image) to be detected into a trained segmentation model, and outputs a probability map (second image) showing that each pixel belongs to peripancreatic necrotic tissue, thus completing an end-to-end mapping from the input image to pixel-level probability prediction. The contrast-enhanced CT image in the venous phase itself provides the clearest view of peripancreatic necrotic tissue, offering high-quality image data. The probability map output by the segmentation model based on this data retains the model's confidence in predicting each pixel, improving the robustness and accuracy of the segmentation.

[0018] This invention utilizes image morphological processing algorithms to convert a probability map (second image) into a 3D segmentation mask. The probability map output by the model may contain imperfections such as minor noise, boundary spikes, or internal holes. By introducing morphological processing algorithms, the initial prediction results are optimized. For example, opening operations can smooth boundaries and remove isolated noise points; closing operations can fill small internal holes and connect neighboring regions. This post-processing effectively corrects minor errors that may exist in the model's predictions, resulting in a more complete, smooth, and anatomically consistent 3D segmentation mask, further improving the clinical reliability of the segmentation results.

[0019] This invention determines the volume of peripancreatic necrotic tissue in the first image based on the number of voxels and the volume of a single voxel in a 3D segmentation mask. After obtaining an accurate 3D segmentation mask, the absolute volume of the necrotic tissue can be automatically and objectively calculated using the inherent spatial resolution information of the CT image (i.e., the actual physical size corresponding to the voxels) through simple multiplication operations. Compared to the previous method of manually drawing layers and then estimating the volume, this invention achieves complete automation of volume calculation, which is not only fast but also provides objective and repeatable results, completely solving the problems of time-consuming, labor-intensive, and inconsistent results associated with traditional methods. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the dataset annotation provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the U-Net network structure provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the segmentation results provided in an embodiment of the present invention; Figure 5 This is a partial cross-sectional view of patient 1 provided in an embodiment of the present invention; Figure 6 This is a partial cross-sectional view of patient 2 provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of a system for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, provided in an embodiment of the present invention. Detailed Implementation

[0022] 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 only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1 like Figure 1 As shown, this embodiment of the invention provides a method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, including: Step 101: Construct a deep convolutional neural network model for segmenting peripancreatic necrotic tissue.

[0024] In this step, such as Figure 3 As shown, a deep convolutional neural network model is constructed based on the two-dimensional U-Net network architecture. The U-Net network adopts a symmetrical encoder-decoder structure.

[0025] For example, this step includes: Construct N (positive integer, e.g., four) sequentially connected downsampling modules; each downsampling module includes at least two sequentially connected convolutional units; each convolutional unit includes a first convolutional layer (e.g., a 3×3 convolutional layer), a batch normalization layer, and a first activation function (e.g., a ReLU activation function) sequentially connected; the output of the last convolutional unit in each downsampling module is downsampled using a pooling layer (e.g., a 2×2 pooling layer).

[0026] Construct N (e.g., four) sequentially connected upsampling modules; where the first upsampling module takes the output of the last downsampling module as input; each upsampling module uses a transposed convolution (e.g., a 2×2 transposed convolution) for upsampling; the outputs of the N upsampling modules are skipped and connected to the output of the last convolutional unit in each downsampling module and then fused.

[0027] Construct an output module; wherein the output module comprises a second convolutional layer (e.g., a 1×1 convolutional layer) and a second activation function (e.g., a sigmoid activation function) connected in sequence; the second convolutional layer takes the output of the last upsampling module as its input.

[0028] The encoding path consists of four sequentially connected downsampling modules, used to extract and compress multi-scale features from the input image I0. The input image I0 is a two-dimensional grayscale image slice, 512 pixels (height denoted as H0) × 512 pixels (width denoted as W0), with an initial number of channels C. Let the input image be... R is the real number field (i.e., the set of all real numbers); the gray value of each pixel in the input image I0 represents the CT value (an index for measuring the density of a substance, defined as the difference between the linear decay coefficient of a substance and the linear decay coefficient of water, divided by the linear decay coefficient of water, and then multiplied by the gradation factor), which is directly related to tissue density (including normal pancreas, peripancreatic necrosis, fat, and fluid accumulation).

[0029] The first downsampling module receives the input image I0. This module consists of two convolutional units connected in sequence, each designed to learn local texture features useful for identifying early, small-scale necrosis or tissue heterogeneity. Each convolutional unit performs the following operations in sequence: first, a convolution operation with a kernel size of 3×3, the output of which is processed by a batch normalization (BN) layer, and finally a nonlinear transformation is introduced through a rectified linear unit (ReLU) activation function.

[0030] Taking the first convolutional unit as an example, its operation can be formally represented as: .

[0031] in, Represents the first feature map F1,1 The feature value of the c-th channel at position (x,y); w(p,q) is the learnable weight parameter of the 3×3 convolution kernel at offset (p,q), used to detect local patterns such as the boundary of necrotic regions, cystic changes or fluid accumulation; is the feature value of the c-th channel of the input image I0 at position (x+p, y+q), corresponding to the CT value of the pixel at that position; b is the bias. This indicates a batch normalization operation; (p,q) is the offset index inside the convolution kernel.

[0032] The first feature map F 1,1 As the input to the second convolutional unit, it undergoes the same processing as the first convolutional unit to output the second feature map F. 1,2 If there is a third convolutional unit, the second feature map F is used here. 1,2 As input to the third convolutional unit, and undergoing the same processing as the first convolutional unit, the third feature map F is output. 1,3 This step is only exemplified when there are two sequentially connected convolutional units. Therefore, in the output of the second feature map F... 1,2 Then, the second feature map F 1,2 Perform a 2×2 max pooling operation to obtain the fourth feature map P1. This method preserves the most significant responses in the peripancreatic region while simultaneously downsampling; where H1=H0 / 2 and W1=W0 / 2. This operation increases the feature receptive field while preserving key information, allowing the U-Net network to perceive the relationships of peripancreatic tissues over a wider area.

[0033] The input to the second downsampling module is the fourth feature map P1, and the processing procedure is the same as that of the first downsampling module, resulting in the fifth feature map F. 2,1 and the sixth feature map F 2,2 Then, after pooling, the seventh feature map P2 is obtained. H2=H1 / 2, W2=W1 / 2; the features learned in this module correspond to higher-level semantic information, such as the overall shape of the necrotic area and its relative position to adjacent blood vessels or intestinal segments; the sixth feature map F 2,2 It is saved for subsequent skip connections.

[0034] The third and fourth downsampling modules process the data sequentially, outputting the eighth feature map P3 and the ninth feature map P4, respectively. H3 = H2 / 2, W3 = W2 / 2; H4 = H3 / 2, W4 = W3 / 2; the third and fourth downsampling modules correspond to the saved tenth feature map F. 3,1 Eleventh Feature Map F 3,2 and the twelfth feature diagram F4,1 and the thirteenth feature map F 4,2 .

[0035] The ninth feature map P4 extracted by the fourth downsampling module is a highly abstract feature representation, corresponding to a large receptive field, which can encode the global contextual information of the entire peripancreatic region. This is crucial for distinguishing between diffuse necrosis and localized effusion. After four downsampling operations, the feature map size decreased sequentially from 512×512 to 256×256, 128×128, and 64×64, finally reaching 32×32, while the number of channels doubled sequentially.

[0036] The decoding path consists of four sequentially connected upsampling modules. Through upsampling and feature fusion, it gradually recovers a high-resolution segmentation map to accurately depict the complex and irregular boundaries of the peripancreatic necrotic tissue. This path includes four upsampling stages.

[0037] The first upsampling module takes the ninth feature map P4 as input. It first performs upsampling through a 2×2 transposed convolution, which can be represented as: .

[0038] Among them, U1 is the feature map output after the ninth feature map P4 is upsampled by 2×2 transposed convolution; ; This represents the eigenvalue of U1 in the c'th channel at position (x', y'); q' represents the learnable weight parameters of the 2×2 transposed convolution kernel at offset (p',q'); c' represents the channel dimension index of the transposed convolution input to the ninth feature map P4. This indicates that P4 is in position. The eigenvalue of the c'th channel; This indicates rounding down; after obtaining U1, U1 is combined with the thirteenth feature map F, which is stored at the same scale as the encoding path and contains rich detailed information. 4,2 Concatenate along the channel dimension: Cat1 = Concatenate(U1, F 4,2 Cat1 is U1 and F 4,2 After performing a concatenation operation along the channel dimension, a fused feature map is obtained; this fuses deep semantics with shallow details, which is crucial for recovering small-scale necrosis or precise boundary locations. Cat1 is then processed by two more convolutional units (both with the same processing procedure as the first convolutional unit) to output the fourteenth feature map D1.

[0039] The subsequent three upsampling modules process the same as the first upsampling module. The second, third, and fourth upsampling modules sequentially output the fifteenth feature map D2, the sixteenth feature map D3, and the seventeenth feature map D4; among them, The original resolution of the input image I0 has been restored. During the final upsampling module operation, the decoded features are compared with the second feature map F. 1,2 The fusion effectively combines the lowest-level image texture information, which helps to correct the prediction of blurred boundary areas. The final output of the U-Net network, the seventeenth feature map D4, is processed by a 1×1 convolutional layer to map the number of channels from C to 1, resulting in a single-channel eighteenth feature map Z. The pixel value of the eighteenth feature map Z integrates the aforementioned multi-layer features and can be considered as the evidence strength that each pixel belongs to the peripancreatic necrosis region. Subsequently, the Sigmoid activation function is applied to calculate the predicted probability P of each pixel (i,j) belonging to the peripancreatic necrosis tissue. i,j : .

[0040] in, Z is the Sigmoid function; i,j Let be the Logit value of pixel (i,j) in the eighteenth feature map Z.

[0041] The final output is a probability prediction graph. This figure directly maps the probability that each pixel in the input image I0 is peripancreatic necrotic tissue, providing a basis for subsequent quantitative analysis (such as necrotic volume calculation).

[0042] Step 102: Use multiple abdominal venous phase enhanced CT images labeled with peripancreatic necrotic tissue as a training set to train a deep convolutional neural network model, and use the trained deep convolutional neural network model as a segmentation model for peripancreatic necrotic tissue.

[0043] This step requires acquiring a dataset of enhanced CT images of the abdomen in the venous phase, in DICOM format. This embodiment uses enhanced CT images of the venous phase as the data source. Compared to plain CT or enhanced arterial phase CT, the venous phase provides the clearest visualization of peripancreatic necrotic tissue. This phase selection can more accurately display the extent, morphology, and distribution characteristics of the peripancreatic necrotic tissue, providing a high-quality image foundation for subsequent automatic segmentation.

[0044] The acquired image data underwent preprocessing, including: resampling the images to a uniform voxel size (1mm×1mm×1mm) and adjusting the image matrix size of each slice to 512×512 to ensure consistency among different patients. Linear interpolation was used for resampling to maintain the spatial resolution and geometric features of the images. Based on the CT value range of the peripancreatic necrotic tissue, the window width and level were adjusted to a specific range (window width 400HU-600HU, window level 40HU-60HU) to enhance the contrast between the necrotic tissue and surrounding tissue, facilitating subsequent segmentation. The pixel values ​​of the CT images were normalized to the [0,1] interval to eliminate differences caused by different scanning equipment and parameters. A minimum-maximum normalization method was used to normalize the pixel intensity of each slice, as detailed below: Where I is the original pixel value, I min and I max These represent the minimum and maximum intensity of the image's grayscale value, respectively.

[0045] Manual annotation includes the following steps: For the segmentation of peripancreatic necrosis, a reference standard is provided manually for the training of the deep convolutional neural network model. The peripancreatic necrosis region is delineated in the CT image after resampling, window width and level adjustment, and normalization processing. The annotation is then repeatedly refined based on binarization processing to obtain the final annotation result, such as... Figure 2 As shown, only white and black pixels are retained. The white area represents peripancreatic necrotic tissue, and the black area represents non-peripancreatic necrotic tissue.

[0046] Furthermore, the preprocessed image data and manually labeled data are simultaneously subjected to the same data augmentation operations to expand the training dataset and improve the generalization ability of the deep convolutional neural network model. These operations mainly include: random rotation (-15° to 15°), random translation (-10% to 10%), random scaling (0.9 to 1.1 times), and random flipping (horizontal or vertical). Finally, each pair of image data and manually labeled data is converted into an npy format file and used as input for training the deep convolutional neural network model.

[0047] During the training phase of the deep convolutional neural network model, a dataset of enhanced CT images of peripancreatic necrosis tissue with meticulous manual annotation was used, divided into training and validation sets in an 8:2 ratio. A five-fold cross-validation strategy was employed to fully evaluate model performance and prevent overfitting. The batch size was set to 8, and the training epochs were 100. The Adam optimizer was used, with an initial learning rate of 0.001, and a learning rate decay strategy was applied, decreasing to 0.5 times the original rate every 20 epochs.

[0048] Furthermore, the loss function employs a weighted combination of Dice loss and cross-entropy loss.

[0049] Calculate the Dice loss L using the following formula. Dice : .

[0050] Where, p m g represents the predicted probability of a deep convolutional neural network model that the m-th pixel belongs to peripancreatic necrotic tissue. m The true label for the m-th pixel; g m =0 indicates that the m-th pixel belongs to non-peripancreatic necrotic tissue; g m =1 indicates that the m-th pixel belongs to peripancreatic necrotic tissue; M is the total number of pixels in the abdominal venous phase enhanced CT image.

[0051] Calculate the cross-entropy loss L using the following formula. CE : .

[0052] The total loss L is determined based on Dice loss and cross-entropy loss: L=αL Dice +βL CE .

[0053] Where α is the weight of the Dice loss and β is the weight of the cross-entropy loss.

[0054] The deep convolutional neural network model is trained based on the total loss until it converges, resulting in a well-trained deep convolutional neural network model that can be used as a segmentation model for peripancreatic necrotic tissue.

[0055] A composite loss function, combining Dice loss and cross-entropy loss, is employed. This design aims to overcome the limitations of single loss functions in specific scenarios, thereby comprehensively improving the performance of the segmentation model. Specifically, the Dice loss function directly optimizes the overlap between the predicted segmented region and the ground truth labeled region. Its global optimization characteristic makes it naturally robust to datasets with imbalanced foreground and background classes, effectively enabling the model to learn the complete and continuous morphology of necrotic tissue regions. Simultaneously, the cross-entropy loss function independently penalizes the prediction error of each pixel, providing stable and detailed pixel-wise gradient signals, which is particularly beneficial for the model to learn accurate boundary details and internal texture features. This combined strategy guides the model during training to focus not only on the overlap rate of the global segmented region but also on the prediction confidence of local pixels, thus significantly improving the segmentation accuracy and robustness of deep convolutional neural network models for peripancreatic necrotic tissue in complex medical image segmentation tasks.

[0056] Step 103: Input the first image into the peripancreatic necrotic tissue segmentation model to output a second image with the same size as the first image; wherein, the first image is the abdominal venous phase enhanced CT image to be detected; and the probability of each pixel belonging to peripancreatic necrotic tissue is known in the second image.

[0057] Step 104: Use an image morphology processing algorithm to convert the second image into a three-dimensional segmentation mask.

[0058] like Figure 4 As shown, morphological post-processing is performed on the preliminary segmentation results output by the peripancreatic necrotic tissue segmentation model.

[0059] For example, this step includes: The probability of each pixel in the second image belonging to peripancreatic necrotic tissue is compared with a probability threshold to determine whether the corresponding voxel of each pixel is assigned a value of 0 or 1, thus obtaining a binarized initial 3D segmentation mask.

[0060] The initial 3D segmentation mask is processed using image morphology processing algorithms to obtain the final 3D segmentation mask.

[0061] Furthermore, the voxels corresponding to pixels in the second image whose probability of belonging to peripancreatic necrotic tissue is greater than or equal to a probability threshold (e.g., 0.5) are assigned a value of 1.

[0062] The voxels corresponding to pixels in the second image whose probability of belonging to peripancreatic necrotic tissue is less than a probability threshold (e.g., 0.5) are assigned a value of 0 to obtain the binarized initial 3D segmentation mask.

[0063] A morphological opening operation (erosion followed by dilation) is performed on the initial 3D segmentation mask to remove small isolated regions caused by noise or uncertainty in the prediction of the deep convolutional neural network model, and the segmentation boundary is smoothed to obtain the first 3D segmentation mask.

[0064] A morphological closing operation (dilation followed by erosion) is performed on the first 3D segmentation mask to fill the small holes inside the segmented region and connect the adjacent but broken regions to obtain the second 3D segmentation mask.

[0065] Connectivity analysis is performed on the second 3D segmentation mask to remove isolated regions with excessively small areas, thus obtaining the final 3D segmentation mask.

[0066] Furthermore, an area threshold is set based on the minimum clinically treatable lesion volume and image resolution limitations. For example, in a CT image, the actual size of a single voxel is 0.5mm × 0.5mm × 1mm, and the minimum feasible lesion volume for interventional treatment is 2mm. 3(Approximately equivalent to 8 voxels), and the number of voxels corresponding to the smallest area that the imaging device can stably resolve is taken as the lower limit (e.g., 20 voxels). The larger value of the two is taken to set the area threshold (20 voxels). All connected regions with an area greater than the area threshold are retained as necrotic tissue regions, and isolated noise points with too small an area are removed.

[0067] Step 105: Determine the volume of the peripancreatic necrotic tissue in the first image based on the number of voxels and the volume of a single voxel in the 3D segmentation mask.

[0068] As shown in Table 1, the performance of the segmentation method for quantifying the volume of peripancreatic necrotic tissue provided in this embodiment was evaluated on the test set. Dice was used as the evaluation index (mean ± standard deviation, Mean ± SD) to evaluate the performance of the optimal weight for each segment and the final optimal weight.

[0069] Table 1. Comparison of U-Net's Five-Fold Segmentation Performance

[0070] Furthermore, the number of voxels with a value of 1 and the volume of voxels in the (final) 3D segmentation mask are obtained; where the volume of a single voxel is the product of the length, width and height of a single voxel.

[0071] The product of the number of voxels with a value of 1 in the 3D segmentation mask and the voxel volume is used as the volume corresponding to the peripancreatic necrotic tissue in the first image.

[0072] Convert the volume unit to the clinically common cubic centimeter, and output the final segmentation result and the corresponding volume value.

[0073] Figure 5 and Figure 6 The model outputs the segmentation results of peripancreatic necrotic tissue from two patients. Figure 5 and Figure 6 Each includes the original image, segmentation results, and volume calculation results. From... Figure 5 and Figure 6 As can be seen, the model can accurately identify and segment the peripancreatic necrotic tissue region and effectively exclude interference from surrounding normal tissue. The segmentation results clearly show the morphology, distribution, and extent of the necrotic tissue, providing intuitive imaging evidence for clinical diagnosis. Furthermore, the automatically calculated necrotic tissue volume based on the segmentation results provides an objective quantitative indicator for assessing the severity and prognosis of acute pancreatitis, validating the effectiveness and practicality of the method of this invention.

[0074] In summary, this embodiment provides a method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images. By constructing and training a deep convolutional neural network model, manual annotation experience is transformed into calculable network parameters, enabling automatic identification and pixel-level probability prediction of peripancreatic necrotic tissue in enhanced CT images of the abdominal venous phase. This fundamentally avoids the problems of high subjectivity and low accuracy associated with traditional manual visual estimation. Furthermore, image morphology processing algorithms are used to optimize the probability map output by the model. Through noise removal, boundary smoothing, and hole filling, the probability map is converted into a complete and anatomically consistent 3D segmentation mask, effectively correcting minor errors that may exist in the model prediction and significantly improving the accuracy and clinical reliability of the segmentation results. Finally, based on the number of voxels in the 3D segmentation mask and the actual physical volume of a single voxel, the absolute volume value of the peripancreatic necrotic tissue is automatically calculated, completing the fully automated quantization process from image input to volume output. This ultimately forms an objective, accurate, and repeatable quantization scheme, completely solving the problems of time-consuming, labor-intensive, and inconsistent results associated with traditional layer-by-layer manual delineation methods.

[0075] Example 2 Based on the same inventive concept as Embodiment 1, this embodiment also provides a system for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images. Since the principle of this system in solving the problem is similar to the aforementioned method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, the implementation of this system can refer to the implementation of the method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images.

[0076] like Figure 7 As shown, the system for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images includes: Module 10 is used to build a deep convolutional neural network model for segmenting peripancreatic necrotic tissue.

[0077] Training module 20 is used to train a deep convolutional neural network model using multiple abdominal venous phase enhanced CT images labeled with peripancreatic necrotic tissue as a training set, and to use the trained deep convolutional neural network model as a segmentation model for peripancreatic necrotic tissue.

[0078] The input module 30 does not input the first image into the peripancreatic necrotic tissue segmentation model, so as to output a second image with the same size as the first image; wherein, the first image is the abdominal venous phase enhanced CT image to be detected; and the probability of each pixel belonging to peripancreatic necrotic tissue is known in the second image.

[0079] The conversion module 40 is used to convert the second image into a three-dimensional segmentation mask using an image morphology processing algorithm.

[0080] The determination module 50 is used to determine the volume corresponding to the peripancreatic necrotic tissue in the first image based on the number of voxels and the volume of a single voxel in the three-dimensional segmentation mask.

[0081] For more detailed information on the working process of each of the above modules, please refer to the relevant content disclosed in Example 1, which will not be repeated here.

[0082] Example 3 This embodiment provides a computer device, including a processor and a memory; wherein, when the processor executes a computer program stored in the memory, it implements the steps of the method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images as described in Embodiment 1.

[0083] For a more detailed explanation of the above method, please refer to the relevant content disclosed in Example 1, which will not be repeated here.

[0084] Example 4 This embodiment provides a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, it implements the steps of the method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images described in Embodiment 1.

[0085] For a more detailed explanation of the above method, please refer to the relevant content disclosed in Example 1, which will not be repeated here.

[0086] Example 5 This embodiment provides a computer program product, including computer-executable instructions or a computer program. When the computer-executable instructions or the computer program are executed by a processor, they implement the steps of the method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images described in Embodiment 1.

[0087] For a more detailed explanation of the above method, please refer to the relevant content disclosed in Example 1, which will not be repeated here.

[0088] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems, devices, storage media, and computer program products disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0089] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0090] In some embodiments, computer-executable instructions may take the form of programs, software, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.

[0091] As an example, computer-executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., files that store one or more modules, subroutines, or code sections).

[0092] As an example, computer-executable instructions can be deployed to execute on a single electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed across multiple locations and interconnected via a communication network.

[0093] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this embodiment are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations.

[0094] The present invention has been described in detail above with reference to specific embodiments and exemplary examples; however, these descriptions should not be construed as limiting the present invention. Those skilled in the art will understand that various equivalent substitutions, modifications, or improvements can be made to the technical solutions and embodiments of the present invention without departing from the spirit and scope of the invention, and all such modifications and improvements fall within the scope of the present invention. The scope of protection of the present invention is defined by the appended claims.

Claims

1. A method for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, characterized in that, include: Construct a deep convolutional neural network model for segmenting peripancreatic necrotic tissue; Multiple abdominal venous phase enhanced CT images labeled with peripancreatic necrotic tissue were used as a training set to train a deep convolutional neural network model, and the trained deep convolutional neural network model was used as a segmentation model for peripancreatic necrotic tissue. The first image is input into the peripancreatic necrotic tissue segmentation model to output a second image of the same size as the first image; wherein, the first image is the abdominal venous phase enhanced CT image to be detected; and the probability of each pixel belonging to peripancreatic necrotic tissue is known in the second image. The second image is converted into a 3D segmentation mask using image morphology processing algorithms; The volume of peripancreatic necrotic tissue in the first image is determined based on the number of voxels and the volume of a single voxel in the 3D segmentation mask.

2. The method for quantifying the volume of peripancreatic necrotic tissue according to claim 1, characterized in that, The construction of the deep convolutional neural network model for segmenting peripancreatic necrotic tissue includes: Construct N sequentially connected downsampling modules; each downsampling module includes at least two sequentially connected convolutional units; each convolutional unit includes a first convolutional layer, a batch normalization layer, and a first activation function, all sequentially connected; the output of the last convolutional unit in each downsampling module is downsampled using a pooling layer; Construct N sequentially connected upsampling modules; the first upsampling module takes the output of the last downsampling module as its input; each upsampling module uses transposed convolution for upsampling; the outputs of the N upsampling modules are skipped and connected to the output of the last convolutional unit in each downsampling module and then fused. Construct an output module; wherein the output module includes a second convolutional layer and a second activation function connected in sequence; the second convolutional layer takes the output of the last upsampling module as input.

3. The method for quantifying the volume of peripancreatic necrotic tissue according to claim 2, characterized in that, The construction of the deep convolutional neural network model for segmenting peripancreatic necrotic tissue includes: Construct four sequentially connected downsampling modules; each downsampling module includes at least two sequentially connected convolutional units; each convolutional unit includes a sequentially connected 3×3 convolutional layer, a batch normalization layer, and a ReLU activation function; the output of the last convolutional unit in each downsampling module is downsampled using a 2×2 pooling layer; And / or, construct four sequentially connected upsampling modules; wherein the first upsampling module takes the output of the last downsampling module as input; each upsampling module uses a 2×2 transposed convolution for upsampling; the outputs of the four upsampling modules are skip-connected to and fused with the output of the last convolutional unit in each downsampling module; And / or, construct an output module; wherein the output module comprises a 1×1 convolutional layer and a sigmoid activation function connected in sequence; the 1×1 convolutional layer takes the output of the last upsampling module as its input.

4. The method for quantifying the volume of peripancreatic necrotic tissue according to claim 1, characterized in that, The step of training a deep convolutional neural network model using multiple abdominal venous phase enhanced CT images labeled with peripancreatic necrosis tissue as a training set, and then using the trained deep convolutional neural network model as a segmentation model for peripancreatic necrosis tissue, includes: Calculate the Dice loss L using the following formula. Dice : ； Where, p m g represents the predicted probability of a deep convolutional neural network model that the m-th pixel belongs to peripancreatic necrotic tissue. m The true label for the m-th pixel; g m =0 indicates that the m-th pixel belongs to non-peripancreatic necrotic tissue; g m =1 indicates that the m-th pixel belongs to peripancreatic necrotic tissue; M is the total number of pixels in the abdominal venous phase enhanced CT image. Calculate the cross-entropy loss L using the following formula. CE : ； The total loss is determined based on the Dice loss and cross-entropy loss. The deep convolutional neural network model is trained based on the total loss until it converges, resulting in a well-trained deep convolutional neural network model, which can be used as a segmentation model for peripancreatic necrotic tissue.

5. The method for quantifying the volume of peripancreatic necrotic tissue according to claim 4, characterized in that, The determination of the total loss based on Dice loss and cross-entropy loss includes: Calculate the total loss L using the following formula: L=αL Dice +βL CE ; Where α is the weight of the Dice loss and β is the weight of the cross-entropy loss.

6. The method for quantifying the volume of peripancreatic necrotic tissue according to claim 1, characterized in that, The process of converting the second image into a three-dimensional segmentation mask using an image morphology processing algorithm includes: The probability of each pixel in the second image belonging to peripancreatic necrotic tissue is compared with the probability threshold to determine whether the voxel corresponding to each pixel is assigned a value of 0 or 1, thus obtaining the binarized initial three-dimensional segmentation mask. The initial 3D segmentation mask is processed using image morphology processing algorithms to obtain the final 3D segmentation mask.

7. The method for quantifying the volume of peripancreatic necrotic tissue according to claim 6, characterized in that, The process of converting the second image into a three-dimensional segmentation mask using an image morphology processing algorithm includes: In the second image, the voxels corresponding to the pixels whose probability of belonging to peripancreatic necrotic tissue is greater than or equal to the probability threshold are assigned a value of 1. The voxels corresponding to the pixels in the second image whose probability of belonging to peripancreatic necrotic tissue is less than the probability threshold are assigned a value of 0 to obtain the binarized initial 3D segmentation mask. And / or, perform morphological opening operations on the initial 3D segmentation mask to obtain the first 3D segmentation mask; A morphological closing operation is performed on the first 3D segmentation mask to obtain the second 3D segmentation mask; Connectivity analysis is performed on the second 3D segmentation mask to remove isolated regions with too small an area, thus obtaining the final 3D segmentation mask.

8. The method for quantifying the volume of peripancreatic necrotic tissue according to claim 7, characterized in that, The process of performing connected component analysis on the second 3D segmentation mask to remove isolated regions with excessively small areas, thereby obtaining the final 3D segmentation mask, includes: Identify all independent connected regions in the second 3D segmentation mask; Calculate the total number of voxels in each connected region; The voxels in connected regions with a total number of voxels less than the voxel threshold are assigned a value of 0 to obtain the final 3D segmentation mask.

9. The method for quantifying the volume of peripancreatic necrotic tissue according to claim 1, characterized in that, The step of determining the volume corresponding to the peripancreatic necrotic tissue in the first image based on the number of voxels and the volume of a single voxel in the 3D segmentation mask includes: Obtain the number and volume of voxels with a value of 1 in the 3D segmentation mask; The product of the number of voxels with a value of 1 in the 3D segmentation mask and the voxel volume is used as the volume corresponding to the peripancreatic necrotic tissue in the first image.

10. A system for quantifying the volume of peripancreatic necrotic tissue based on enhanced CT images, characterized in that, include: Build modules are used to construct deep convolutional neural network models for segmenting peripancreatic necrotic tissue; The training module is used to train a deep convolutional neural network model using multiple abdominal venous phase enhanced CT images labeled with peripancreatic necrotic tissue as the training set, and to use the trained deep convolutional neural network model as a segmentation model for peripancreatic necrotic tissue. The input module does not input the first image into the peripancreatic necrotic tissue segmentation model, so as to output a second image with the same size as the first image; wherein, the first image is the abdominal venous phase enhanced CT image to be detected; and the probability of each pixel belonging to peripancreatic necrotic tissue is known in the second image. The conversion module is used to convert the second image into a 3D segmentation mask using image morphology processing algorithms. The determination module is used to determine the volume corresponding to the peripancreatic necrotic tissue in the first image based on the number of voxels and the volume of a single voxel in the 3D segmentation mask.

Images