CT retroperitoneal lymph node segmentation and quantification method

Automatic segmentation and quantification of CT retroperitoneal lymph nodes through a deep learning model based on Dense U-net, the problem of manual identification and measurement in lymph node diagnosis is solved, high-precision automated diagnosis is achieved, diagnostic efficiency and accuracy are improved, and diagnostic efficiency is reduced. subjective error.

CN118941489APending Publication Date: 2024-11-12SOUTHWEAT UNIV OF SCI & TECH
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202310525235.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-11
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

In the prior art, the diagnosis of lymph nodes depends on manual identification and measurement by doctors, and there are problems such as difficult boundary distinction, low degree of diagnosis uniformity, and difficulty in quantifying lymph nodes, resulting in subjective errors and low diagnostic efficiency.

Method used

The deep learning model based on Dense U-net is used to automatically segment and quantify the retroperitoneal lymph nodes of CT. The model is optimized through the cosine annealing attenuation learning strategy to realize automatic volume quantification of three-dimensional lymph nodes, reduce manual intervention, and improve diagnostic accuracy.

Benefits of technology

It realizes the automation, high-precision identification and segmentation of lymph nodes, reduces doctors' workload, reduces subjective errors, improves diagnostic efficiency, provides unified standards, and guides correct clinical treatment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

The invention relates to an automatic segmentation and quantification method for a CT (Computed Tomography) retroperitoneal lymph node, which is mainly used for carrying out automatic and high-precision identification segmentation and three-dimensional volume automatic quantification on a CT retroperitoneal lymph node image based on Dense U-net so as to obtain an accurate lymph region, edge and volume. The method has the advantages that the workload of doctors can be greatly reduced, subjective diagnosis errors of the doctors are reduced, the working efficiency of the doctors is improved, and correct clinical treatment is guided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a CT retroperitoneal lymph node segmentation and quantification method, which mainly performs automatic segmentation and three-dimensional volume automatic quantification on CT retroperitoneal lymph nodes based on Dense U-net. Background Art

[0002] Lymph nodes are ellipsoidal or bean-shaped, of varying sizes. They are an important part of the immune system and can effectively fight viruses and bacteria, remove harmful substances from the body, and protect the body from disease infection. The main causes of lymph node disease are sarcoidosis, lymphoma, lymphadenopathy (swelling), or fungal infection. When the body is abnormal, lymph nodes vary in size. The important measurement is the change in lymph node size over time, which can be used as an indicator of tumor activity and is a key factor in tumor staging. Therefore, it is very necessary to evaluate the size of lymph nodes. It can not only be used to determine treatment plans and evaluate the patient's response to a specific course of treatment, but also determine the prognosis of treatment.

[0003] At present, computed tomography (CT) imaging mode provides good spatial resolution and high-quality imaging for lymph nodes and is one of the most important means of lymph node diagnosis in clinical practice. The diagnosis of enlarged lymph nodes is mainly done by experienced doctors who manually identify the location of lymph nodes and manually measure the long and short axes of lymph nodes. The short axes are then used to determine whether the lymph nodes have enlarged and how much they have enlarged, to evaluate the size, shape and relationship of the lymph nodes, and thus to diagnose the disease and monitor the efficacy. However, this method has problems such as difficulty in distinguishing boundaries, low diagnostic uniformity, and difficulty in quantifying lymph nodes.

[0004] This paper studies the deep learning automatic segmentation technology of retroperitoneal lymph node CT, and builds a precise segmentation system for retroperitoneal enlarged lymph nodes based on the Dense U-net network. It performs automatic and high-precision recognition and segmentation of lymph node CT images to obtain accurate lymph node areas, edges and volumes, and unify standards. It can greatly reduce the workload of doctors, reduce the subjective errors of doctors' diagnosis, improve doctors' work efficiency, and guide correct clinical treatment. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a CT retroperitoneal lymph node automatic quantification system based on the improved D-Dense U-net, which can quickly perform automatic segmentation and quantification of CT retroperitoneal lymph nodes without manual intervention, while achieving high accuracy. The model is trained with nearly 10,000 screened retroperitoneal lymph data, and then the retroperitoneal lymph CT data set of Mianyang Central Hospital is used for model generalization testing. It is optimized through the cosine annealing attenuation learning strategy, and finally the algorithm is used to realize the automatic volume quantification of three-dimensional lymph nodes and build a segmentation and quantification system.

[0006] A specific implementation of a CT retroperitoneal lymph node segmentation and quantification method includes the following steps: Step 1: After importing the Dense U-net model, input the patient's CT retroperitoneal lymph node image, read the pixel spacing and layer thickness of the image, input the ROI cropped area into the network, extract and normalize the ROI, and obtain a lymph node image suitable for model inference.

[0007] Step 2: Use the Dense U-net network model based on deep learning to achieve pixel-level segmentation of the lymph image, extract the segmented MASK and fill the holes, restore it to the original image size, mark the segmented MASK in red, and superimpose it with the original image.

[0008] Step 3: Automatically quantify the three-dimensional volume of lymph nodes based on the pixel-level segmentation results: 3.1 Combining the slice spacing and pixel spacing of CT images to achieve automatic calculation of lymph node volume: Traverse the lymph node contours on the CT images, calculate the maximum axial diameter d of the tumor contour on each CT image, take the maximum lymph node diameter maxlenth in all CT images in mm, and use the lymph node volume calculation function in step 3.2 to obtain the volume in ml.

[0009] 3.2 Lymph node volume calculation function The calculation of lymph node volume is realized by combining the slice spacing and pixel spacing of CT images. The calculation formula is as follows:

[0010] Wherein: N: number of pixels of retroperitoneal lymph nodes, SS: pixel spacing, SD: slice spacing, i: number of abdominal lymph node image layers.

[0011] Compared with the prior art, the advantages of the present invention are: it is a method completely based on artificial intelligence and deep learning to solve the problem of detecting enlarged lymph nodes, has strong generalization performance, and three-dimensional automatic quantification improves the calculation deviation caused by subjective factors in actual clinical work, improves the work efficiency of doctors, and guides correct clinical treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is an explanatory diagram of the retroperitoneal lymph nodes.

[0013] Figure 2 This is the CT interpretation of the retroperitoneal lymph nodes (red and yellow marks indicate enlarged lymph nodes).

[0014] Figure 3 This is the main interface of the system after importing the model and data.

[0015] Figure 4This is a diagram showing the system reasoning segmentation results. DETAILED DESCRIPTION

[0016] The present invention is further described in detail below with reference to the accompanying drawings.

[0017] A specific implementation of a CT retroperitoneal lymph node segmentation and quantification method includes the following steps: Step 1: After importing the Dense U-net model, input the patient's CT retroperitoneal lymph node image, read the pixel spacing and layer thickness of the image, input the ROI cropped area into the network, extract and normalize the ROI, and obtain a lymph node image suitable for model inference.

[0018] Step 2: Use the Dense U-net network model based on deep learning to achieve pixel-level segmentation of the lymphatic image, extract the segmented MASK and fill the holes to restore it to the original image size.

[0019] Step 3: Automatically quantify the three-dimensional volume of lymph nodes based on the pixel-level segmentation results: 3.1 Combining the slice spacing and pixel spacing of CT images to achieve automatic calculation of lymph node volume: Traverse the lymph node contours on the CT images, calculate the maximum axial diameter d of the tumor contour on each CT image, take the maximum lymph node diameter maxlenth in all CT images in mm, and use the lymph node volume calculation function in step 3.2 to obtain the volume in ml.

[0020] 3.2 Lymph node volume calculation function The calculation of lymph node volume is realized by combining the slice spacing and pixel spacing of CT images. The calculation formula is as follows:

[0021] Wherein: N: number of pixels of retroperitoneal lymph nodes, SS: pixel spacing, SD: slice spacing, i: number of abdominal lymph node image layers.

[0022] Compared with the prior art, the advantages of the present invention are: a method completely based on artificial intelligence and deep learning to solve the problem of detecting enlarged lymph nodes, strong generalization performance, three-dimensional automatic quantification to improve the calculation deviation caused by subjective factors in actual clinical work, improve the work efficiency of doctors, and guide correct clinical treatment.

[0023] The excellent performance of the present invention is verified by experiments as follows: method Dice IoU Recall DenseUnet 0.7782 0.6569 0.7849 Dilated convolution + DenseUnet 0.7856 0.6720 0.8189 Attention+DenseUnet 0.7881 0.6579 0.7632 method Dice IoU Recall Unet 0.7318 0.6273 0.7867 DenseUnet 0.7782 0.6569 0.7849 D-DenseUnet 0.7961 0.6788 0.8036 Dataset ABD_RL RLN_CT Dice 0.7961 0.7424 IoU 0.6788 0.6232 Recall 0.8036 0.7530 From the above, it can be seen that the improved dense U-shaped symmetric semantic segmentation model D-DenseUnet used in the present invention can be well applied to different data sets, can achieve accurate segmentation of retroperitoneal lymph nodes, and has good generalization ability and adaptability. This shows that the present invention can assist or replace doctors in diagnosing enlarged lymph nodes to a certain extent, improve doctors' diagnostic efficiency, and promote the intelligent development of hospitals.

Claims

1. A specific implementation of a CT retroperitoneal lymph node segmentation and quantification method includes the following steps: Step 1: After importing the Dense U-net model, input the patient's CT retroperitoneal lymph node image, read the pixel spacing and layer thickness of the image, input the ROI cropped area into the network, extract and normalize the ROI, and obtain a lymph node image suitable for model inference; Step 2: Use the Dense U-net network model based on deep learning to achieve pixel-level segmentation of the lymphatic image, extract the segmented MASK and fill the holes to restore it to the original image size, mark the segmented MASK in red, and superimpose it with the original image; Step 3: Automatically quantify the three-dimensional volume of lymph nodes based on the pixel-level segmentation results: 3.1 Combining the slice spacing and pixel spacing of CT images to achieve automatic calculation of lymph node volume: Traverse the lymph node contours on the CT images, calculate the maximum axial diameter d of the tumor contour on each CT image, take the maximum lymph node diameter maxlenth in all CT images in mm, and use the lymph node volume calculation function in step 3.2 to obtain the volume Volume in ml; 3.2 Lymph node volume calculation function The calculation of lymph node volume is realized by combining the slice spacing and pixel spacing of CT images. The calculation formula is as follows: ; in: N: number of pixels of retroperitoneal lymph nodes, SS: pixel spacing, SD: slice spacing, i: number of abdominal lymph node image layers; Compared with the prior art, the advantages of the present invention are: it is a method completely based on artificial intelligence and deep learning to solve the problem of detecting enlarged lymph nodes, has strong generalization performance, and three-dimensional automatic quantification improves the calculation deviation caused by subjective factors in actual clinical work, improves the work efficiency of doctors, and guides correct clinical treatment.

Citation Information

Patent Citations

  • Lesion volume measurement method and device

    CN111047611A

  • Pulmonary nodule segmentation method and device, computing equipment and storable medium

    CN113838067A

  • Automatic hemorrhage expansion detection from head CT images

    US20220309667A1