A lung cancer IGTV automatic delineation method and system

By using the attention Unet network to automatically outline 4D CT images in the outline of IGTV in lung cancer, the problem of low outline accuracy in the prior art is solved, and higher outline accuracy and efficiency of IGTV in lung cancer are achieved.

CN118609764BActive Publication Date: 2024-11-08江西省肿瘤医院(江西省第二人民医院 江西省癌症中心)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411060946.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-08
Estimated Expiration
2044-08-05

AI Technical Summary

Technical Problem

Existing IGTV outline methods for lung cancer are low in accuracy when dealing with irregular respiratory movements, especially manual outlines are time-consuming and inefficient. Alternatives such as maximum density projection and average density projection may underestimate tumor volume.

Method used

Using an automatic outline method based on 4D CT images, GTV is outlined phase by obtaining historical images, AIP and MIP images are generated, and attention Unet network is built, and the images are trained to obtain the target neural network model, which is used to automatically outline IGTV and improve accuracy. .

Benefits of technology

It improves the automatic outline accuracy of IGTV in lung cancer, can more accurately include information in 4D CT, is suitable for irregular motion patterns, and improves the accuracy of radiotherapy plans.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118609764B_ABST
    Figure CN118609764B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for automatically delineating IGTV of lung cancer. The method obtains historical 4D CT images, delineates GTV phase by phase, obtains a first GTV, and fuses the delineation results to obtain a historical IGTV; based on the historical 4D CT images, generates corresponding AIP images and MIP images, delineates GTVs respectively, resamples the images to a preset spatial resolution, crops out corresponding regions, and normalizes the intensity; constructs an attention Unet network; trains the attention Unet network according to the historical IGTV, and then inputs the GTV of the AIP image or MIP image to be processed into the trained neural network model to obtain a target IGTV. Specifically, with the help of the MIP image or the AIP image, the information contained in the 4D CT can be used to the greatest extent to obtain higher automatic delineation accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic delineation of lung cancer IGTV, and in particular relates to a method and system for automatic delineation of lung cancer IGTV. Background Art

[0002] Lung cancer is the leading cause of cancer-related deaths worldwide. Radiotherapy is one of the main treatments for lung cancer, aiming to deliver high doses of radiation to malignant tumor cells while minimizing the impact on surrounding healthy tissues. Respiration-induced intra-fractionation motion significantly affects the precise dose delivery to the tumor, which is particularly evident in lung cancer. This variation may lead to insufficient tumor control or organ-threatening damage. To address the impact of intra-fractionation motion, researchers have developed various techniques, such as respiratory gating, tumor tracking, breath holding, and motion envelope strategies.

[0003] Among them, the concept of internal target volume (ITV) is widely accepted and applied. This method expands the target volume through four-dimensional (4D) computed tomography (CT) to include the full range of tumor movement throughout the respiratory cycle. These 4D CT scans generate a series of three-dimensional (3D) images, each capturing a different phase of the normal respiratory cycle. The internal gross tumor volume (IGTV) can be used to efficiently define the ITV. The ITV is calculated as the IGTV plus a boundary, which is usually a fixed value to cover surrounding subclinical lesions, etc. Therefore, accurate delineation of the IGTV is essential for effective radiotherapy planning. The most accurate IGTV delineation method currently is to manually outline the GTV (gross tumor volume) of all ten respiratory phases in the 4DCT images and then merge them to create a complete IGTV. However, this method is time-consuming, labor-intensive and inefficient.

[0004] Maximum intensity projection (MIP) is widely used as an alternative. MIP captures the highest density of each pixel during the respiratory cycle of 4D CT. In practice, IGTV delineation is usually performed directly on the MIP image and then adjusted according to the motion information of the respiratory phase. Multiple studies have verified its effectiveness in target delineation, but some studies have suggested that it may underestimate the size of ITV. Average intensity projection (AIP) is another post-processing method that creates a single 3D image by averaging the voxel intensity of all phases in the respiratory cycle, thus containing all the information of 4D CT. Although it has certain clinical value, it also has the problem of underestimating tumor volume.

[0005] It can be found that the above method is not applicable to all types of respiratory movements, especially for irregular movement patterns, which may affect the delineation accuracy. Summary of the invention

[0006] Based on this, an embodiment of the present invention provides a method and system for automatic delineation of lung cancer IGTV, aiming to improve the accuracy of automatic delineation of lung cancer IGTV.

[0007] A first aspect of an embodiment of the present invention provides a method for automatically delineating an IGTV for lung cancer, the method comprising:

[0008] Obtaining historical 4D CT images, outlining the GTV phase by phase on the historical 4D CT images to obtain the first GTV, and fusing the outlining results to obtain the historical IGTV;

[0009] Based on the historical 4D CT images, the corresponding AIP images and MIP images are generated, and GTVs are delineated on the AIP images and MIP images respectively to obtain the second GTV and the third GTV;

[0010] Resample the historical IGTV, AIP image, MIP image, second GTV and third GTV to a preset spatial resolution, and set a region of preset pixel size based on the center of the first GTV, crop the corresponding regions of the historical IGTV, AIP image, MIP image, second GTV and third GTV, and normalize the intensity to the range of 0 to 1;

[0011] Construct attention Unet networks for AIP images and MIP images respectively. The number of channels of each convolutional layer in the attention Unet network is 32, 64, 128, and 256 respectively. For AIP images, the input of the corresponding attention Unet network is the AIP image, and the output is the fused image of the second GTV. For MIP images, the input of the corresponding attention Unet network is the MIP image, and the output is the fused image of the third GTV.

[0012] According to the historical IGTV, the attention Unet network for the AIP image and the MIP image is trained respectively to obtain a target neural network model;

[0013] The GTV of the AIP image or MIP image to be processed is obtained and input into the corresponding target neural network model to obtain the target IGTV.

[0014] Furthermore, in the step of acquiring a historical 4D CT image, outlining the GTV phase by phase for the historical 4D CT image to obtain a first GTV, and fusing the outlining results to obtain a historical IGTV, a logical union is taken of the binary images obtained by outlining the GTV phase by phase to complete the fusion.

[0015] Furthermore, the step of taking a logical union of the binary images obtained by delineating the GTV phase by phase to complete the fusion includes:

[0016] The coordinate position of the target pixel point of the binary image obtained by delineating the GTV in each phase is obtained respectively, wherein the target pixel point represents a pixel point whose pixel value is white;

[0017] According to the coordinate positions, the outermost pixels among the target pixels of the binary image obtained by outlining the GTV in each phase are determined and marked to obtain the contours of the corresponding GTV respectively;

[0018] Determine the contour obtained by first performing phase delineation GTV as the first contour, and on the basis of the first contour, take the logical union of the contour obtained by performing next phase delineation GTV and the first contour to obtain a fused second contour, and so on, continuously take the logical union of the contour obtained by performing next phase delineation GTV and the current contour to obtain a target contour;

[0019] All the pixels within the target outline are assigned white pixel values ​​to complete the fusion.

[0020] Further, the step of determining that the contour obtained by first performing phase delineation GTV is the first contour, and based on the first contour, taking a logical union of the contour obtained by performing next phase delineation GTV and the first contour to obtain a fused second contour, and so on, continuously taking a logical union of the contour obtained by performing next phase delineation GTV and the current contour to obtain the target contour includes:

[0021] It is determined in sequence whether there are overlapping parts of the pixel points of the contours obtained by delineating the GTV in two adjacent phases;

[0022] If yes, determine the coordinate positions of the overlapping pixels of the contours obtained by delineating the GTV in two adjacent phases, and determine the outermost pixels according to the overlapping coordinate positions of the pixels, so as to finally obtain the target contour;

[0023] If not, the contour obtained by outlining the GTV in two adjacent phases is retained simultaneously to finally obtain the target contour.

[0024] Furthermore, the attention Unet network includes the attention Unet network including an encoder, a decoder, a skip connection, an attention gate and a gating signal.

[0025] Furthermore, before the step of training the attentionUnet network for the AIP image and the MIP image according to the historical IGTV to obtain the target neural network model, the normalized results are normalized using data enhancement technology, including random flipping on three axes with a probability of 0.1, random 90-degree rotation with a probability of 0.1, and random intensity offset with a probability of 0.5, with an offset of 0.1.

[0026] Furthermore, in the step of training the attentionUnet network for AIP images and MIP images respectively according to the historical IGTV to obtain the target neural network model, the optimizer adopted AdamW with weight decay of 1e-5, and used the optimal learning rate for training, the loss function was Dice loss, the maximum number of iterations was 100, and the batch size was set to 32 or 64.

[0027] A second aspect of an embodiment of the present invention provides a lung cancer IGTV automatic delineation system, the system comprising:

[0028] An acquisition module is used to acquire a historical 4D CT image, delineate the GTV phase by phase of the historical 4D CT image, obtain a first GTV, and fuse the delineation results to obtain a historical IGTV;

[0029] A delineation module is used to generate a corresponding AIP image and a MIP image based on a historical 4D CT image, and delineate the GTV on the AIP image and the MIP image respectively to obtain a second GTV and a third GTV;

[0030] A normalization module is used to resample the historical IGTV, AIP image, MIP image, second GTV and third GTV to a preset spatial resolution, and to set an area of ​​preset pixel size based on the center of the first GTV, to crop the corresponding areas of the historical IGTV, AIP image, MIP image, second GTV and third GTV, and to normalize the intensity to a range of 0 to 1;

[0031] The network construction module is used to construct the attention Unet network for AIP images and MIP images respectively. The number of channels of each convolutional layer in the attention Unet network is 32, 64, 128, and 256 respectively. For AIP images, the input of the corresponding attention Unet network is the AIP image, and the output is the fused image of the second GTV. For MIP images, the input of the corresponding attention Unet network is the MIP image, and the output is the fused image of the third GTV.

[0032] A training module, used for training the attentionUnet network for the AIP image and the MIP image respectively according to the historical IGTV to obtain a target neural network model;

[0033] The input module is used to obtain the GTV of the AIP image or MIP image to be processed and input it into the corresponding target neural network model to obtain the target IGTV.

[0034] A third aspect of an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the lung cancer IGTV automatic delineation method provided in the first aspect.

[0035] A fourth aspect of an embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the automatic delineation method for IGTV of lung cancer provided in the first aspect is implemented.

[0036] A method and system for automatically delineating IGTV of lung cancer provided in an embodiment of the present invention, the method obtains a historical 4D CT image, delineates the GTV of the historical 4D CT image phase by phase, obtains a first GTV, and fuses the delineation results to obtain a historical IGTV; based on the historical 4D CT image, generates a corresponding AIP image and a MIP image, and delineates the GTV of the AIP image and the MIP image respectively, resamples the image to a preset spatial resolution, and sets an area of ​​a preset pixel size based on the center of the first GTV, crops the corresponding area of ​​the image, and normalizes the intensity; constructs attention Unet networks for the AIP image and the MIP image respectively; according to the historical IGTV, trains the attention Unet networks for the AIP image and the MIP image respectively to obtain a target neural network model; obtains the GTV of the AIP image or the MIP image to be processed, and inputs it into the corresponding target neural network model to obtain the target IGTV. Specifically, with the help of the MIP image or the AIP image, the information contained in the 4D CT can be used to the greatest extent to obtain higher automatic delineation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flowchart of an automatic delineation method for IGTV of lung cancer provided in the first embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the attention Unet network structure;

[0039] Figure 3 The relationship diagrams of MIP image and AIP image and Dyss coefficient respectively;

[0040] Figure 4 The relationship diagrams between MIP image and AIP image and Hausdorff distance respectively;

[0041] Figure 5 is the relationship diagram of the Dyess coefficient for each patient based on the MIP image;

[0042] Figure 6 is the Hausdorff distance relationship diagram for each patient based on MIP images;

[0043] Figure 7 is the relationship diagram of the Dyess coefficient for each patient based on the AIP image;

[0044] Figure 8 is the Hausdorff distance relationship diagram of each patient based on AIP images;

[0045] Fig. 9 Comparison diagram of the evaluation of the Dyce coefficient in the automatic outlining performance of IGTV between attention Unet, Unet, and Vnet using AIP images;

[0046] Fig.10 Comparison diagram of the evaluation of Hausdorff distance in IGTV automatic outlining performance between attention Unet, Unet and Vnet using AIP images;

[0047] Fig.11 A structural block diagram of a lung cancer IGTV automatic delineation system provided in Embodiment 3 of the present invention;

[0048] Fig.12 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0049] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0050] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0052] Embodiment 1

[0053] See also Figure 1 , Figure 1 A flowchart of an implementation of a lung cancer IGTV automatic delineation method provided in Embodiment 1 of the present invention is shown. The lung cancer IGTV automatic delineation method specifically includes steps S01 to S06.

[0054] Step S01, obtaining a historical 4D CT image, outlining the GTV phase by phase of the historical 4D CT image to obtain a first GTV, and fusing the outlining results to obtain a historical IGTV.

[0055] It should be noted that the first GTV obtained after outlining the GTV phase by phase is a three-dimensional image, and the final fusion result is also a three-dimensional image. The Eclipse software (v13.5) is used to outline the GTV phase by phase. Specifically, the binary image obtained by outlining the GTV phase by phase is logically unioned to complete the fusion. It can be understood that in image processing, a binary image (BinaryImage) usually contains only two pixel values, usually 0 (black) and 1 (white), or vice versa. Taking the logical union of two binary images means that for each pair of pixels at corresponding positions, if the pixel value in any image is 1, the pixel value at that position in the result image is also 1; only when the pixel values ​​at that position in both images are 0, the pixel value at that position in the result image is 0.

[0056] In this embodiment, the coordinate position of the target pixel point of the binary image obtained by delineating the GTV in each phase is obtained respectively, wherein the target pixel point represents a pixel point with a white pixel value, that is, a pixel point with a pixel value of 1;

[0057] According to the coordinate positions, the outermost pixels among the target pixels of the binary image obtained by outlining the GTV in each phase are determined and marked to obtain the contours of the corresponding GTVs. It can be understood that the contours are three-dimensional.

[0058] Determine the contour obtained by first performing phase delineation GTV as the first contour. Based on the first contour, take the logical union of the contour obtained by the next phase delineation GTV and the first contour to obtain the fused second contour. Similarly, continuously take the logical union of the contour obtained by the next phase delineation GTV and the current contour to obtain the target contour. Specifically, determine in turn whether there is an overlap in the pixel points of the contours obtained by two adjacent phase delineation GTVs.

[0059] If it is determined that there is an overlapped portion of the pixel points of the contours obtained by the two adjacent phases outlining the GTV, the coordinate positions of the overlapped pixel points of the contours obtained by the two adjacent phases outlining the GTV are determined, and the outermost pixel points are determined according to the coordinate positions of the overlapped pixel points, so as to finally obtain the target contour;

[0060] If it is determined that there is no overlap in the pixel points of the contours obtained by delineating the GTV in two adjacent phases, the contours obtained by delineating the GTV in the two adjacent phases are retained at the same time to finally obtain the target contour;

[0061] All pixels within the target contour are assigned white pixel values ​​to complete the fusion. The advantage of this is that you only need to take the logical union of the pixel values ​​of the contour and finally assign values ​​to the pixels within the contour to quickly complete the fusion of multiple images.

[0062] Step S02: Generate corresponding AIP images and MIP images based on historical 4D CT images, and delineate GTVs on the AIP images and MIP images respectively to obtain a second GTV and a third GTV.

[0063] Similarly, the same user was asked to use Eclipse software (v13.5) to draw GTV on the AIP image and the MIP image, where the result of GTV drawing on the AIP image was the second GTV, and the result of GTV drawing on the MIP image was the third GTV.

[0064] Step S03, resample the historical IGTV, AIP image, MIP image, second GTV and third GTV to a preset spatial resolution, and set an area of ​​preset pixel size based on the center of the first GTV, crop the corresponding areas of the historical IGTV, AIP image, MIP image, second GTV and third GTV, and normalize the intensity to the range of 0~1.

[0065] Specifically, the historical IGTV, AIP image, MIP image, second GTV and third GTV were resampled to 1 mm × 1 mm × 2 mm, and with the center of the first GTV as the reference, a region of 112 × 112 × 80 pixels was set, the corresponding regions of the historical IGTV, AIP image, MIP image, second GTV and third GTV were cropped, and the intensity was normalized to the range of 0–1.

[0066] Step S04, constructing attention Unet networks for AIP images and MIP images respectively, the number of channels of each convolutional layer in the attention Unet network is 32, 64, 128, and 256 respectively, wherein, for AIP images, the input of the corresponding attention Unet network is the AIP image, and the output is the fused image of the second GTV, and for MIP images, the input of the corresponding attention Unet network is the MIP image, and the output is the fused image of the third GTV.

[0067] The images are divided into training set, validation set and test set in the ratio of 7:1:2. In addition, before model training, data augmentation technology is used to normalize the results including random flipping on three axes with a probability of 0.1, random 90-degree rotation with a probability of 0.1, and random intensity offset with a probability of 0.5, with an offset of 0.1.

[0068] The attention Unet network includes the encoder, decoder, skip connection, attention gate and gating signal. Please refer to Figure 2, is a schematic diagram of the attention Unet network structure. Specifically, input: a single-channel 3D CT image with a size of 112×112×80; encoder: the encoder part consists of multiple convolution blocks, each of which contains two consecutive 3×3×3 convolution operations, followed by a ReLU activation function. Each convolution block is followed by a max-pooling layer to reduce the spatial dimension and increase the number of feature channels; decoder: the decoder part is symmetrical with the encoder, and each decoding block contains upsampling, convolution operations, and ReLU activation. The upsampling operation is used to restore the spatial dimension, and the convolution operation is used to refine the features; jump connection: there is a jump connection from each layer of the encoder to the corresponding decoder layer, which helps to restore high-resolution features; attention gate and gating signal: the gating signal generated by the attention module is introduced at the jump connection. By calculating the attention weight between the encoder and decoder feature maps, the features of important areas are enhanced and irrelevant backgrounds are suppressed. The attention module includes query, key, and value, where the query comes from the decoder and the key and value come from the encoder. The attention weight is calculated by the query and key, and then multiplied by the value to generate a weighted feature map, which is the gating signal; Output: The last layer is a 1×1×1 convolution, which is used to map the feature map to the required number of segmentation categories and generate the final segmentation result with a size of 112×112×80.

[0069] Through this structure, Attention U-Net further improves the model's ability to focus on important areas while maintaining the U-Net's efficient feature extraction capability, making it perform better in medical image segmentation tasks.

[0070] Step S05: According to the historical IGTV, the attention Unet network for the AIP image and the MIP image is trained respectively to obtain a target neural network model.

[0071] Specifically, the attention Unet network was built using Python (v3.10.9) and PyTorch (v2.0.1). The model was trained on a server with eight NVIDIA GeForce RTX 4090 GPUs. The optimizer used AdamW with a weight decay of 1e-5 and the optimal learning rate was used for training. The loss function was Dice loss, the maximum number of iterations was 100, and the batch size was set to 32 or 64.

[0072] Step S06, obtaining the GTV of the AIP image or MIP image to be processed, and inputting it into the corresponding target neural network model to obtain the target IGTV.

[0073] In summary, the lung cancer IGTV automatic delineation method in the above embodiment of the present invention obtains a historical 4D CT image, delineates the GTV of the historical 4D CT image phase by phase, obtains a first GTV, and fuses the delineation results to obtain a historical IGTV; based on the historical 4D CT image, generates a corresponding AIP image and a MIP image, and delineates the GTV of the AIP image and the MIP image respectively, resamples the image to a preset spatial resolution, and sets an area of ​​a preset pixel size based on the center of the first GTV, crops the corresponding area of ​​the image, and normalizes the intensity; constructs attention Unet networks for the AIP image and the MIP image respectively; according to the historical IGTV, trains the attention Unet networks for the AIP image and the MIP image respectively to obtain a target neural network model; obtains the GTV of the AIP image or the MIP image to be processed, and inputs it into the corresponding target neural network model to obtain the target IGTV. Specifically, with the help of the MIP image or the AIP image, the information contained in the 4D CT can be used to the greatest extent to obtain a higher automatic delineation accuracy.

[0074] Embodiment 2

[0075] Embodiment 2 of the present invention In order to verify the feasibility of the automatic delineation method of lung cancer IGTV in Embodiment 1 of the present invention, data of 124 lung cancer patients were used for verification. Specifically, 4D CT images were acquired using Siemens SOMATOMDefinition AS CT, with a voxel spacing of 0.976mm×0.976mm×2mm or 1.269mm×1.269mm×2mm. Among them, 70% (n=87) were assigned to the training set for model training, 10% (n=12) were assigned to the validation set for hyperparameter adjustment, and the remaining 20% ​​(n=25) were assigned to the test set for model evaluation.

[0076] The evaluation indicators are the Dyss coefficient (DSC) and the Hausdorff distance (HD). The baseline is calculated based on GTV and IGTVgt. Here, GTV is obtained by outlining the GTV of the 4D CT image phase by phase, and IGTVgt is the IGTV obtained by fusing the outlining results.

[0077] Figure 3 are the relationship diagrams between MIP image and AIP image and Dyss coefficient respectively. Figure 4Figure 2 is the relationship between MIP images and AIP images and Hausdorff distance. The mean DSC (± standard deviation, SD) of the model on MIP images was 0.852±0.053, slightly lower than the mean DSC on AIP images (0.871±0.048), but the difference was not statistically significant (P value = 0.13). Similarly, the mean HD (± SD) of the model on MIP and AIP images were 3.209±2.136 and 2.958±2.266, respectively, with a P value of 0.43, indicating that the difference was not statistically significant. Figure 5 is the relationship diagram of the Diess coefficient for each patient based on the MIP image, Figure 6 is the Hausdorff distance relationship diagram for each patient based on MIP images, Figure 7 is the relationship diagram of the Dyss coefficient for each patient based on the AIP image, Figure 8 Figure 2 shows the Hausdorff distance relationship for each patient based on the AIP image. The best Diess coefficient and Hausdorff distance values ​​were 0.929 (patient 6) and 1.000 (patient 24), respectively, both achieved on the AIP image.

[0078] In addition, attention Unet is compared with Unet and Vnet.

[0079] Fig. 9 This is a comparison chart of the evaluation of the Dice coefficient in the automatic outlining performance of IGTV between attention Unet, Unet and Vnet using AIP images. Fig.10 Comparison of the evaluation of Hausdorff distance in IGTV automatic delineation performance between attention Unet, Unet, and Vnet using AIP images. The average Dyss coefficients were 0.871±0.048, 0.870±0.048, and 0.853±0.051, respectively. The average Hausdorff distances were 2.958±2.266, 2.932±2.262, and 3.263±2.319, respectively. Overall, attention Unet and Unet performed similarly and both outperformed Vnet. However, there was no statistically significant difference in either the Dyss coefficient or the Hausdorff distance, as the P values ​​were all above 0.05.

[0080] Embodiment 3

[0081] See also Fig.11 , Fig.11 1 is a structural block diagram of a lung cancer IGTV automatic delineation system 200 provided in Embodiment 3 of the present invention. The lung cancer IGTV automatic delineation system 200 comprises: an acquisition module 21, a delineation module 22, a normalization module 23, a network construction module 24, a training module 25 and an input module 26, wherein:

[0082] The acquisition module 21 is used to acquire a historical 4D CT image, delineate the GTV phase by phase of the historical 4D CT image to obtain a first GTV, and fuse the delineation results to obtain a historical IGTV, wherein a logical union is taken for the binary images obtained by delineating the GTV phase by phase to complete the fusion;

[0083] A delineation module 22 is used to generate a corresponding AIP image and a MIP image based on the historical 4D CT image, and delineate the GTV on the AIP image and the MIP image respectively to obtain a second GTV and a third GTV;

[0084] A normalization module 23 is used to resample the historical IGTV, AIP image, MIP image, second GTV and third GTV to a preset spatial resolution, and to set an area of ​​preset pixel size based on the center of the first GTV, to crop the corresponding areas of the historical IGTV, AIP image, MIP image, second GTV and third GTV, and to normalize the intensity to a range of 0 to 1;

[0085] A network construction module 24 is used to construct attention Unet networks for AIP images and MIP images respectively. The number of channels of each convolutional layer in the attention Unet network is 32, 64, 128, and 256 respectively. For AIP images, the input of the corresponding attention Unet network is the AIP image, and the output is the fused image of the second GTV. For MIP images, the input of the corresponding attention Unet network is the MIP image, and the output is the fused image of the third GTV.

[0086] The training module 25 is used to train the attentionUnet network for the AIP image and the MIP image according to the historical IGTV to obtain the target neural network model. Prior to this, the normalized results include random flipping on three axes with a probability of 0.1, random 90-degree rotation with a probability of 0.1, and random intensity offset with a probability of 0.5, and the offset is 0.1. In addition, during the training process, the optimizer uses AdamW with a weight decay of 1e-5, and uses the optimal learning rate for training. The loss function is Dice loss, the maximum number of iterations is 100, and the batch size is set to 32 or 64;

[0087] The input module 26 is used to obtain the GTV of the AIP image or MIP image to be processed, and input it into the corresponding target neural network model to obtain the target IGTV.

[0088] Furthermore, in some optional embodiments of the present invention, the acquisition module 21 includes:

[0089] The first acquisition unit is used to respectively acquire the coordinate position of the target pixel point of the binary image obtained by delineating the GTV in each phase, wherein the target pixel point represents a pixel point whose pixel value is white;

[0090] A first determining unit is used to determine the outermost pixel points among the target pixel points of the binary image obtained by outlining the GTV in each phase according to the coordinate position, and mark them to obtain the contours of the corresponding GTV respectively;

[0091] A merging unit is used to determine the contour obtained by first performing phase delineation GTV as the first contour, and based on the first contour, take a logical union of the contour obtained by performing next phase delineation GTV and the first contour to obtain a fused second contour, and so on, continuously taking a logical union of the contour obtained by performing next phase delineation GTV and the current contour to obtain a target contour;

[0092] The assigning unit is used to assign white pixel values ​​to all the pixel points within the target contour to complete the fusion.

[0093] Further, in some optional embodiments of the present invention, the merging unit includes:

[0094] A judging subunit is used to judge in turn whether there is an overlapped part in the pixel points of the contours obtained by delineating the GTV in two adjacent phases;

[0095] A determination subunit is used to determine the coordinate positions of the overlapped pixels of the contours obtained by the two adjacent phase outline GTVs when it is determined that there is an overlapped portion, and determine the outermost pixel points according to the overlapped coordinate positions of the pixel points, so as to finally obtain the target contour;

[0096] The retaining subunit is used to retain the contours obtained by the two adjacent phase outline GTVs at the same time when it is determined that there is no overlap in the pixel points of the contours obtained by the two adjacent phase outline GTVs, so as to finally obtain the target contour.

[0097] Embodiment 4

[0098] Another aspect of the present invention provides an electronic device, see Fig.12 , shown is an electronic device in Embodiment 4 of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, the automatic delineation method for IGTV of lung cancer as described above is implemented.

[0099] In some embodiments, the processor 10 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 20, such as executing access restriction programs.

[0100] Among them, the memory 20 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as a hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of an electronic device, such as a plug-in hard disk equipped on the electronic device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Further, the memory 20 can also include both an internal storage unit of the electronic device and an external storage device. The memory 20 can not only be used to store application software and various types of data of the electronic device, but also can be used to temporarily store data that has been output or is to be output.

[0101] It should be pointed out that Fig.12 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than those shown in the figure, or combine certain components, or arrange the components differently.

[0102] The embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned lung cancer IGTV automatic delineation method.

[0103] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0104] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0105] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0106] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0107] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the attached claims.

Claims

1. A method for automatically delineating IGTV of lung cancer, characterized in that: The method comprises: Acquire a historical 4D CT image, delineate a GTV phase by phase for the historical 4D CT image to obtain a first GTV, and fuse the delineation results to obtain a historical IGTV, wherein a logical union is taken for a binary image obtained by delineating the GTV phase by phase to complete the fusion; Based on the historical 4D CT images, the corresponding AIP images and MIP images are generated, and GTVs are delineated on the AIP images and MIP images respectively to obtain the second GTV and the third GTV; Resample the historical IGTV, AIP image, MIP image, second GTV and third GTV to a preset spatial resolution, and set a region of preset pixel size based on the center of the first GTV, crop the corresponding regions of the historical IGTV, AIP image, MIP image, second GTV and third GTV, and normalize the intensity to the range of 0 to 1; Construct attention Unet networks for AIP images and MIP images respectively. The number of channels of each convolutional layer in the attention Unet network is 32, 64, 128, and 256 respectively. For AIP images, the input of the corresponding attention Unet network is the AIP image, and the output is the fused image of the second GTV. For MIP images, the input of the corresponding attention Unet network is the MIP image, and the output is the fused image of the third GTV. According to the historical IGTV, the attention Unet network for the AIP image and the MIP image is trained respectively to obtain a target neural network model; Obtain the GTV of the AIP image or MIP image to be processed, and input it into the corresponding target neural network model to obtain the target IGTV; The step of taking a logical union of the binary images obtained by delineating the GTV phase by phase to complete the fusion includes: The coordinate position of the target pixel point of the binary image obtained by delineating the GTV in each phase is obtained respectively, wherein the target pixel point represents a pixel point whose pixel value is white; According to the coordinate positions, the outermost pixels among the target pixels of the binary image obtained by outlining the GTV in each phase are determined and marked to obtain the contours of the corresponding GTV respectively; Determine that the contour obtained by first performing phase delineation GTV is the first contour. On the basis of the first contour, take the logical union of the contour obtained by the next phase delineation GTV and the first contour to obtain a fused second contour. Similarly, continuously take the logical union of the contour obtained by the next phase delineation GTV and the current contour to obtain the target contour. Specifically, determine in turn whether there is an overlap in the pixel points of the contours obtained by two adjacent phase delineation GTVs. If yes, determine the coordinate positions of the overlapping pixels of the contours obtained by delineating the GTV in two adjacent phases, and determine the outermost pixels according to the overlapping coordinate positions of the pixels, so as to finally obtain the target contour; If not, the contours obtained by outlining the GTV in two adjacent phases are retained simultaneously to finally obtain the target contour; All the pixels within the target outline are assigned white pixel values ​​to complete the fusion.

2. The method for automatically delineating IGTV of lung cancer according to claim 1, characterized in that: The attention Unet network includes an encoder, a decoder, a skip connection, an attention gate, and a gating signal.

3. The method for automatically delineating IGTV of lung cancer according to claim 2, characterized in that: According to the historical IGTV, the attention Unet network for the AIP image and the MIP image is trained respectively. Before the step of obtaining the target neural network model, the data enhancement technology is used to normalize the results including random flipping on three axes with a probability of 0.1, random 90-degree rotation with a probability of 0.1, and random intensity offset with a probability of 0.5, and the offset is 0.

1.

4. The method for automatically delineating IGTV of lung cancer according to claim 3, characterized in that: In the step of training the attention Unet network for AIP images and MIP images respectively according to the historical IGTV to obtain the target neural network model, the optimizer adopts AdamW with weight decay of 1e-5, and uses the optimal learning rate for training, the loss function is Dice loss, the maximum number of iterations is 100, and the batch size is set to 32 or 64.

5. A lung cancer IGTV automatic delineation system, characterized in that: The system is used to implement the method for automatically delineating the lung cancer IGTV according to any one of claims 1 to 4, comprising: An acquisition module is used to acquire a historical 4D CT image, delineate the GTV phase by phase of the historical 4D CT image, obtain a first GTV, and fuse the delineation results to obtain a historical IGTV; A delineation module is used to generate a corresponding AIP image and a MIP image based on a historical 4D CT image, and delineate the GTV on the AIP image and the MIP image respectively to obtain a second GTV and a third GTV; A normalization module is used to resample the historical IGTV, AIP image, MIP image, second GTV and third GTV to a preset spatial resolution, and to set an area of ​​preset pixel size based on the center of the first GTV, to crop the corresponding areas of the historical IGTV, AIP image, MIP image, second GTV and third GTV, and to normalize the intensity to a range of 0 to 1; The network construction module is used to construct the attention Unet network for AIP images and MIP images respectively. The number of channels of each convolutional layer in the attention Unet network is 32, 64, 128, and 256 respectively. For AIP images, the input of the corresponding attention Unet network is the AIP image, and the output is the fused image of the second GTV. For MIP images, the input of the corresponding attention Unet network is the MIP image, and the output is the fused image of the third GTV. A training module, used for training the attention Unet network for the AIP image and the MIP image respectively according to the historical IGTV to obtain a target neural network model; The input module is used to obtain the GTV of the AIP image or MIP image to be processed and input it into the corresponding target neural network model to obtain the target IGTV.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the automatic delineation method for IGTV of lung cancer as described in any one of claims 1 to 4 is implemented.

7. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for automatically delineating the lung cancer IGTV as described in any one of claims 1 to 4 is implemented.