CN-116883426-B - Lung region segmentation method, lung disease assessment method, lung region segmentation device, lung disease assessment device, electronic equipment and storage medium

CN116883426BCN 116883426 BCN116883426 BCN 116883426BCN-116883426-B

Abstract

The disclosure relates to a lung region segmentation and lung disease assessment method and device, electronic equipment and storage medium, and relates to the technical field of DR lung image segmentation. The segmentation method comprises the steps of obtaining a segmentation model of a preset convolutional neural network, training a DR (digital radiography) lung area label image of the segmentation model, and a plurality of DR lung images to be segmented at multiple moments in the breathing process or in the breath-hold state, wherein the method for determining the DR lung area label image for training the segmentation model comprises the steps of respectively detecting rib edge boundaries, lung tip boundaries, mediastinum and transverse septum edges of left side chest images and right side chest images of the plurality of DR lung area images to obtain the DR lung area label image, training the segmentation model by utilizing the DR lung area label image, and completing left lung and/or right lung segmentation of the plurality of DR lung images to be segmented based on the trained segmentation model. The lung region segmentation and lung disease assessment of DR lung images can be achieved.

Inventors

YANG YINGJIAN
Hua Xianguo
OuYang Zhanglei
WU TIANQI
GUO PENG
LI YONG
ZHENG SHENGZHI
ZHENG JIE
CHEN JING

Assignees

深圳蓝影医学科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20230703

Claims (20)

1. A method of lung segmentation, comprising: Acquiring a segmentation model of a preset convolutional neural network, and a plurality of DR lung images to be segmented at a plurality of moments in the breathing process or in a breath-hold state, wherein the DR lung region label image is used for training the segmentation model; The method comprises the steps of determining a DR (digital radiography) region label image for training the segmentation model, calculating a first left lung sharp edge point and a left lung mediastinum edge point corresponding to the shortest distance between a lung sharp boundary and a mediastinum edge in a left chest image of a plurality of DR lung images, calculating a second left lung sharp edge point and a first left lung laterally sharp edge point corresponding to the shortest distance between the lung sharp boundary and the mediastinum edge in the left chest image, calculating a second left lung sharp edge point and a laterally sharp edge point corresponding to the shortest distance between the rib boundary and the laterally sharp edge in the left chest image, calculating a first left lung sharp edge point and a second left lung sharp edge point corresponding to the shortest distance between the rib boundary and the laterally sharp edge in the left chest image, calculating a second lung sharp edge point and a laterally sharp edge point corresponding to the laterally sharp edge point in the right chest image, and calculating a third point corresponding to the first right sharp edge point, a third point corresponding to the third sharp edge point, and a third point corresponding to the laterally sharp edge point; training the segmentation model by utilizing DR (digital radiography) lung region label images corresponding to the DR lung region images; And based on the trained segmentation model, completing left lung and/or right lung segmentation of the DR lung images to be segmented.
2. The lung segmentation method according to claim 1, wherein rib border detection is performed on the left chest image, comprising: constructing a direction derivative template of the left chest image by using the direction derivative, and setting the weighted depth of the direction derivative template; performing template traversal of direction derivative on the left chest image by using the direction derivative template corresponding to the weighted depth, and overlapping the template traversal result into the left chest image to obtain a left chest overlapping image; performing binarization processing on the left chest superimposed image to obtain a left rib edge binary image; obtaining a left rib edge angle diagram to be screened according to the left rib edge binary diagram and the left chest superposition image; obtaining a filtered left rib angle diagram based on the rib angle diagram of the left side to be filtered and the first rib angle; and selecting the communication domain from the left rib edge angle diagram to obtain a rib edge boundary corresponding to the maximum communication domain.
3. The method for distinguishing lung regions according to claim 2, wherein obtaining a left rib angle map to be screened according to the left rib binary map and the left chest superimposed image comprises: performing morphological opening and closing operation and refinement treatment on the left rib edge binary image to obtain a left rib edge binary image after morphological treatment; Performing AND operation on the morphologically processed left rib edge binary image and the gradient direction angle of each pixel in the left chest superimposed image to obtain a rib edge angle image on the left to be screened.
4. A method of lung segmentation according to any of claims 2 or 3, wherein prior to said constructing a directional derivative template of the left chest image using directional derivatives, the left chest image is scaled gaussian blurred to obtain a corresponding left chest gaussian blurred image; Constructing a directional derivative template of the left chest Gaussian blur image by using directional derivatives; In the rib edge boundary detection process of the left chest image, performing template traversal of the direction derivative of the left chest Gaussian blur image by using a direction derivative template corresponding to the weighted depth, and overlapping the template traversal result into the left chest image to obtain a left chest Gaussian blur overlapping image; Performing binarization processing on the left chest Gaussian blur superimposed image to obtain a left rib edge binary image; and obtaining a left rib edge angle diagram to be screened according to the left rib edge binary diagram and the left chest Gaussian blur superimposed image.
5. The lung segmentation method according to claim 1, wherein rib border detection of the right chest image comprises: constructing a direction derivative template of the right chest image by using the direction derivative, and setting the weighted depth of the direction derivative template; performing template traversal of direction derivative on the right chest image by using the direction derivative template corresponding to the weighted depth, and overlapping the template traversal result into the right chest image to obtain a right chest overlapping image; Performing binarization processing on the right chest superimposed image to obtain a right rib edge binary image; Obtaining a rib edge angle diagram of the right side to be screened according to the right rib edge binary diagram and the right chest superposition image; Obtaining a screened right rib angle diagram based on the rib angle diagram of the right side to be screened and the second rib angle; and selecting the connected domain from the right rib edge angle diagram to obtain a rib edge boundary corresponding to the maximum connected domain.
6. The method for distinguishing lung regions according to claim 5, wherein obtaining a right-side rib angle map to be screened from the right-side rib edge binary map and the right-side chest superimposed image comprises: Performing morphological opening and closing operation and refinement treatment on the right rib edge binary image to obtain a right rib edge binary image after morphological treatment; Performing AND operation on the morphologically processed right rib edge binary image and the gradient direction angle of each pixel in the right chest superposition image to obtain a rib edge angle image on the right side to be screened.
7. The pulmonary region segmentation method according to any one of claims 5 or 6, wherein prior to the constructing a direction derivative template of the right chest image using direction derivatives, performing a scale gaussian blur on the right chest image to obtain a corresponding right chest gaussian blur image; constructing a direction derivative template of the right chest Gaussian blur image by using the direction derivative; In the rib edge boundary detection process of the right chest image, performing template traversal of the direction derivative of the right chest Gaussian blur image by using a direction derivative template corresponding to the weighted depth, and overlapping the template traversal result into the right chest image to obtain a right chest Gaussian blur overlapping image; performing binarization processing on the Gaussian blur superimposed image of the right chest to obtain a right rib edge binary image; and obtaining a rib edge angle diagram of the right side to be screened according to the right rib edge binary diagram and the right chest Gaussian blur superimposed image.
8. The lung segmentation method according to claim 1, wherein performing left apex boundary detection on the left chest image comprises: determining a left lung apex detection area according to the left chest image; Determining a left lung apex edge binary image according to the left lung apex detection area; and fitting by adopting a quadratic function according to the left lung apex edge binary image to obtain a fitted left lung apex boundary.
9. The lung segmentation method according to claim 8, wherein the determining a left lung apex detection region from the left chest image comprises: detecting a first coordinate corresponding to the uppermost coordinate point of the rib edge of the left chest image; the region formed by the hypotenuse formed by the first coordinate and the coordinate point of the rightmost upper corner of the left chest image is configured as a left lung apex detection region.
10. The lung segmentation method according to claim 1, wherein performing right lung tip boundary detection on the right chest image comprises: determining a right lung apex detection area according to the right chest image; determining a right lung apex edge binary image according to the right lung apex detection area; and fitting by adopting a quadratic function according to the right lung apex edge binary image to obtain a fitted right lung apex boundary.
11. The lung segmentation method according to claim 9, wherein the determining a right lung apex detection region from the right chest image comprises: Detecting a second coordinate corresponding to the uppermost coordinate point of the rib edge of the right chest image; And the region formed by the hypotenuse formed by the second coordinate and the coordinate point of the leftmost upper corner of the left chest image is configured as a right lung apex detection region.
12. The lung segmentation method according to claim 1, wherein the left lung mediastinum and transverse septum edge detection is performed on the left chest image, respectively, comprising: binarizing the left chest image to obtain a left chest binary image; Performing edge detection on the left chest binary image to obtain a left chest edge binary image; obtaining a left chest edge angle diagram according to the left chest binary image and the gradient direction angle of each pixel in the left chest edge binary image; Obtaining a left side diaphragm and mediastinum edge angle diagram after selection according to the obtained left side chest edge angle diagram and the edge angle ranges of the diaphragm and the mediastinum; And carrying out connected domain selection processing according to the selected left transverse diaphragm and mediastinum edge angle diagram to obtain the left lung mediastinum and transverse diaphragm edge corresponding to the maximum connected domain.
13. The method of claim 12, wherein binarizing the left chest image to obtain a left chest binary image comprises: and performing contrast enhancement processing and maximum inter-class variance processing on the left chest Gaussian blur image corresponding to the left chest image to obtain a left chest binary image.
14. The pulmonary area differentiation method according to any one of claims 12 or 13, wherein determining a gradient direction angle for each pixel in the left chest edge binary map includes: performing transverse gradient and longitudinal gradient calculation on the left chest Gaussian blur image corresponding to the left chest image to obtain a left chest transverse gradient map and a left chest longitudinal gradient map; And obtaining the gradient direction angle of each pixel in the Gaussian blur image of the left chest based on the transverse gradient image and the longitudinal gradient image of the left chest.
15. The lung segmentation method according to claim 1, wherein performing right lung mediastinum and transverse septum edge detection on the right chest image, respectively, comprises: Binarizing the right chest image to obtain a right chest binary image; Performing edge detection on the right chest binary image to obtain a right chest edge binary image; obtaining a right chest edge angle diagram according to the right chest binary image and the gradient direction angle of each pixel in the right chest edge binary image; obtaining a right side diaphragm and mediastinum edge angle diagram after selection according to the obtained right side chest edge angle diagram and the edge angle ranges of the diaphragm and the mediastinum; And carrying out connected domain selection processing according to the selected right transverse diaphragm and mediastinum edge angle diagram to obtain right lung mediastinum and transverse diaphragm edges corresponding to the maximum connected domain.
16. The method of claim 14, wherein binarizing the right chest image to obtain a right chest binary image comprises: and performing contrast enhancement processing and maximum inter-class variance processing on the right chest Gaussian blur image corresponding to the right chest image to obtain a right chest binary image.
17. The pulmonary area differentiation method according to any one of claims 15 or 16, wherein determining a gradient direction angle for each pixel in the right chest edge binary map comprises: Performing transverse gradient and longitudinal gradient calculation on the right chest Gaussian blur image corresponding to the right chest image to obtain a transverse gradient map and a longitudinal gradient map of the right chest; and obtaining the gradient direction angle of each pixel in the Gaussian blurred image of the right chest based on the transverse gradient image and the longitudinal gradient image of the right chest.
18. The method of any one of claims 1-3, 5, 6, 8-13, 15, 16, wherein before performing rib border, tip border, and mediastinum and diaphragm border detection on the left side chest image and the right side chest image of the plurality of DR lung region images, respectively, obtaining DR lung region tag images corresponding to the plurality of DR lung region images, comprising: respectively carrying out chest detection on the DR lung region images to obtain a plurality of chest images corresponding to the DR lung region images; Dividing the plurality of chest images into the left side chest image and the right side chest image; and respectively detecting the rib margin boundary, the lung tip boundary, the mediastinum and the transverse margin of the left chest image and the right chest image.
19. The method according to claim 4, wherein before performing rib border, tip border, mediastinum and diaphragm border detection on the left chest image and the right chest image of the plurality of DR lung region images, respectively, obtaining DR lung region tag images corresponding to the plurality of DR lung region images, the method comprises: respectively carrying out chest detection on the DR lung region images to obtain a plurality of chest images corresponding to the DR lung region images; Dividing the plurality of chest images into the left side chest image and the right side chest image; and respectively detecting the rib margin boundary, the lung tip boundary, the mediastinum and the transverse margin of the left chest image and the right chest image.
20. The method according to claim 7, wherein before performing rib border, tip border, mediastinum and diaphragm border detection on the left side chest image and the right side chest image of the plurality of DR lung region images, respectively, obtaining DR lung region tag images corresponding to the plurality of DR lung region images, the method comprises: respectively carrying out chest detection on the DR lung region images to obtain a plurality of chest images corresponding to the DR lung region images; Dividing the plurality of chest images into the left side chest image and the right side chest image; and respectively detecting the rib margin boundary, the lung tip boundary, the mediastinum and the transverse margin of the left chest image and the right chest image.

Description

Lung region segmentation method, lung disease assessment method, lung region segmentation device, lung disease assessment device, electronic equipment and storage medium Technical Field The disclosure relates to the technical field of DR lung image segmentation, in particular to a lung region segmentation and lung disease assessment method and device, electronic equipment and storage medium. Background Digital X-Ray (DR) images can provide high-resolution and real-time X-Ray images, and have been widely used in bone system, chest, dental and other examinations, such as fracture diagnosis, lung disease screening, dental imaging, etc. In particular, DR imaging devices have been widely used for chest imaging to diagnose intrathoracic pulmonary disease or to quantitatively evaluate and analyze the lungs. One of the keys of accurate intrathoracic pulmonary disease diagnosis or quantitative pulmonary assessment and analysis is to accurately segment a plurality of DR lung images to be segmented at multiple times during respiration or in a breath-hold state. In particular, the quality of the DR lung image is difficult to ensure in the DR lung image acquisition process in the breathing process, and the technical problem to be solved is still needed to be solved at present for a plurality of DR lung images to be segmented at multiple moments in the breathing process. Based on the above, a method for performing lung segmentation on multiple DR lung images at multiple times in the respiratory process or in the breath-hold state needs to be provided, so as to provide a basis for diagnosis or quantitative evaluation and analysis of chest lung diseases, so as to solve the problem that the prior art lacks of performing lung segmentation on multiple DR lung images at multiple times in the respiratory process or in the breath-hold state, and then influences diagnosis or quantitative evaluation and analysis of chest lung diseases, and improve the intelligent auxiliary diagnosis or evaluation level based on the DR lung images. Disclosure of Invention The disclosure provides a lung region segmentation and lung disease assessment method and device, electronic equipment and a storage medium technical scheme. According to one aspect of the disclosure, a lung region segmentation method is provided, which comprises the steps of obtaining a segmentation model of a preset convolutional neural network, DR lung region label images for training the segmentation model, and a plurality of DR lung region label images to be segmented at multiple moments in the breathing process or in a breath-hold state, wherein the method for determining the DR lung region label images for training the segmentation model comprises the steps of respectively carrying out rib boundary, tip boundary, mediastinum and diaphragm edge detection on left side chest images and right side chest images of the plurality of DR lung region images to obtain DR lung region label images corresponding to the plurality of DR lung region images; training the segmentation model by utilizing DR (digital radiography) lung region label images corresponding to the DR lung region images; And based on the trained segmentation model, completing left lung and/or right lung segmentation of the DR lung images to be segmented. Preferably, the method for detecting the rib edge boundary, the lung tip boundary, the mediastinum and the transverse septum edge of the left chest image and the right chest image of the plurality of DR lung region images respectively to obtain DR lung region label images corresponding to the plurality of DR lung region images comprises the steps of obtaining left lung segmentation images corresponding to the DR lung region label images according to the rib edge boundary, the lung tip boundary, the mediastinum and the transverse septum edge corresponding to the left chest image; obtaining a right lung segmentation image corresponding to the DR lung region label image according to the rib edge boundary, the lung tip boundary, the mediastinum and the transverse diaphragm edge corresponding to the right chest image, and/or The method for rib border of the left chest image comprises the following steps: constructing a direction derivative template of the left chest image by using the direction derivative, and setting a set weighting depth of the direction derivative template; Performing template traversal of direction derivative on the left chest image by using a direction derivative template corresponding to the set weighted depth, and overlapping the template traversal result into the left chest image to obtain a left chest overlapping image; performing binarization processing on the left chest superimposed image to obtain a left rib edge binary image; obtaining a left rib edge angle diagram to be screened according to the left rib edge binary diagram and the left chest superposition image; obtaining a filtered left rib angle diagram based on the rib angle diagram of th