A bronchoscopic lesion localization method based on reinforcement learning

Abstract

The present invention relates to the technical field of reinforcement learning, and in particular to a bronchoscopic lesion localization method based on reinforcement learning, the method comprising: obtaining a current bronchoscopic image, performing connected domain extraction on the current bronchoscopic image, screening out channel connected domains and cartilage candidate edges from the obtained connected domains; screening out real cartilage edges; constructing channel cartilage edge clusters, and determining the co-ring coefficient between each two real cartilage edges in each channel cartilage edge cluster; screening out co-ring edge clusters, and identifying interval lines; determining an incomplete factor corresponding to each interval line; determining a consistency index corresponding to each incomplete interval line; if the consistency index corresponding to the incomplete interval line is greater than a preset consistency threshold, determining the incomplete interval line as a lesion interval line. The present invention realizes the localization of the lesion area of ​​the bronchoscopic image and improves the accuracy of the localization of the lesion area of ​​the bronchoscopic image.

Description

Technical Field

[0001] The present invention relates to the technical field of reinforcement learning, and in particular to a bronchoscopic lesion localization method based on reinforcement learning. Background Art

[0002] When a patient's bronchus has a lesion, doctors can observe the airway to identify the lesion (such as a tumor, inflammation, or infection) and take appropriate measures based on the lesion's shape and location. Therefore, accurate localization of the lesion area in bronchoscopic images is a crucial step in bronchoscopy, playing a crucial role in diagnosis and treatment. Currently, the method commonly used to locate the target area is to segment the target area based on grayscale value differences.

[0003] However, when locating the lesion area in bronchoscopic images by using the difference in grayscale values, the following technical problems often occur:

[0004] Since the grayscale difference between the lesion area and the normal tissue area in the bronchoscopic image is not large, directly locating the lesion area of ​​the bronchoscopic image by the difference in grayscale values ​​may lead to misjudgment of the lesion pixels, resulting in poor accuracy in locating the lesion area of ​​the bronchoscopic image. Summary of the Invention

[0005] In order to solve the technical problem of poor accuracy in locating lesion areas in bronchoscopic images, the present invention proposes a bronchoscopic lesion localization method based on reinforcement learning.

[0006] In a first aspect, the present invention provides a bronchoscopic lesion localization method based on reinforcement learning, the method comprising:

[0007] Acquire the current bronchoscopic image, perform connected domain extraction on the current bronchoscopic image, and screen out channel connected domains and cartilage candidate edges from the obtained connected domains;

[0008] The channel connected domain closest to the cartilage candidate edge is determined as the representative connected domain corresponding to the cartilage candidate edge, and the real cartilage edge is screened out from all the cartilage candidate edges whose corresponding representative connected domains are the same connected domain;

[0009] The corresponding real cartilage edges representing the same connected domain are used to form a channel cartilage edge cluster, and the co-circularity coefficient between every two real cartilage edges in each channel cartilage edge cluster is determined;

[0010] According to the co-circularity coefficient between different real cartilage edges in each channel cartilage edge cluster, a co-circular edge cluster is screened from each channel cartilage edge cluster, and a septal line is identified from each co-circular edge cluster;

[0011] According to the distribution of each interval line, an incomplete factor corresponding to each interval line is determined; if the incomplete factor corresponding to the interval line is greater than a preset incomplete threshold, the interval line is determined as an incomplete interval line;

[0012] Determine the consistency index corresponding to each incomplete interval line based on the current bronchoscopic image and the previous bronchoscopic image;

[0013] If the consistency index corresponding to the incomplete interval line is greater than the preset consistency threshold, the incomplete interval line is determined to be the lesion interval line.

[0014] In combination with the first aspect above, in a possible implementation, screening out the channel connected domain and the cartilage candidate edge from the obtained connected domain includes:

[0015] Performing threshold segmentation on all connected domains in the current bronchoscope image, and determining a connected domain in a type of connected domain with a lower grayscale value among the two types of connected domains obtained by segmentation as a channel connected domain;

[0016] Filtering reference pixel points from a connected domain in the current bronchoscope image;

[0017] Filter out from all reference pixels a reference pixel whose corresponding first pixel is also a reference pixel as a selected pixel;

[0018] The line formed by the continuous selected pixel points is determined as the candidate cartilage edge.

[0019] In combination with the first aspect above, in a possible implementation, screening reference pixels from a connected domain in the current bronchoscopic image includes:

[0020] Determining any connected domain in the current bronchoscope image as a marked connected domain, and determining any pixel point in the marked connected domain as a marked pixel point;

[0021] Filtering a pixel point with the smallest gradient difference with the marked pixel point from a preset neighborhood corresponding to the marked pixel point as the first pixel point corresponding to the marked pixel point;

[0022] Filtering out a pixel point with the largest corresponding gradient value from a preset neighborhood corresponding to the marked pixel point as a second pixel point corresponding to the marked pixel point;

[0023] If the first pixel point and the second pixel point corresponding to the marked pixel point are the same pixel point, the marked pixel point is determined as the reference pixel point.

[0024] In combination with the first aspect above, in a possible implementation, screening out the true cartilage edge from all cartilage candidate edges whose corresponding representative connected domains are the same connected domain includes:

[0025] The representative connected domain corresponding to any cartilage candidate edge is determined as the reference connected domain;

[0026] By random sampling, a different number of cartilage candidate edges are randomly extracted from all cartilage candidate edges whose corresponding representative connected domain is the reference connected domain, and the cartilage candidate edges extracted each time are formed into a cartilage candidate edge set, thereby obtaining multiple cartilage candidate edge sets under the reference connected domain;

[0027] Performing ipsilateral division on the cartilage candidate edges in each cartilage candidate edge set to obtain respective cartilage candidate edge groups corresponding to each cartilage candidate edge set;

[0028] The information entropy of the distance between the centroids of all cartilage candidate edges and the centroids of their corresponding representative connected domains in each cartilage candidate edge group is determined as the distance information entropy corresponding to each cartilage candidate edge group;

[0029] The mean of the DTW distances between different cartilage candidate edges in each cartilage candidate edge group is determined as the target difference factor corresponding to each cartilage candidate edge group;

[0030] Determine the target loss value corresponding to each cartilage candidate edge set based on all distance information entropies and target difference factors corresponding to each cartilage candidate edge set;

[0031] A cartilage candidate edge set with the smallest corresponding target loss value is screened out from all cartilage candidate edge sets under the reference connected domain as a temporary edge set, and each cartilage candidate edge in the temporary edge set is determined as a true cartilage edge.

[0032] In combination with the first aspect above, in a possible implementation, the formula corresponding to the target loss value corresponding to the cartilage candidate edge set is:

[0033] ; Where E is the target loss value corresponding to the set of cartilage candidate edges; is a hyperbolic function; is the variance of all distance information entropies corresponding to the set of cartilage candidate edges; is the mean of all target difference factors corresponding to the set of cartilage candidate edges.

[0034] In combination with the first aspect above, in a possible implementation, determining the co-circularity coefficient between every two real cartilage edges in each channel cartilage edge cluster includes:

[0035] Any real cartilage edge is determined as the marked cartilage edge, and a ray passing through the mass center of the representative connected domain corresponding to the marked cartilage edge is taken as the endpoint, and recorded as the target ray corresponding to the marked cartilage edge;

[0036] Determine the intersection of the target ray corresponding to the marked cartilage edge and the representative connected domain as the target line segment corresponding to the marked cartilage edge;

[0037] From all real cartilage edges intersecting the target line segment corresponding to the marked cartilage edge, screen out the real cartilage edge farthest from the centroid of the marked cartilage edge as the reference edge corresponding to the marked cartilage edge;

[0038] Determine the distance between the centroid of the marked cartilage edge and the centroid of the corresponding representative connected domain as a first distance factor corresponding to the marked cartilage edge;

[0039] Determine the distance between the centroid of the reference edge corresponding to the marked cartilage edge and the centroid of the representative connected domain corresponding to the marked cartilage edge as a second distance factor corresponding to the marked cartilage edge;

[0040] Determining the ratio of the first distance factor and the second distance factor corresponding to the marked cartilage edge as the reference ratio corresponding to the marked cartilage edge;

[0041] Any channel cartilage edge cluster is determined as a marked cartilage edge cluster, and the co-circular coefficient between each two real cartilage edges in the marked cartilage edge cluster is determined based on the reference ratio corresponding to each two real cartilage edges in the marked cartilage edge cluster.

[0042] In combination with the first aspect above, in one possible implementation, determining the co-circular coefficient between each two real cartilage edges in the marked cartilage edge cluster according to the reference ratio corresponding to each two real cartilage edges in the marked cartilage edge cluster includes:

[0043] Determine any two real cartilage edges in the marked cartilage edge cluster as a first marked edge and a second marked edge respectively;

[0044] Performing ellipse fitting on the union of the first marked edge and the second marked edge, and determining a fitting loss obtained by the ellipse fitting as a non-co-loop factor between the first marked edge and the second marked edge;

[0045] The co-circular coefficient between the first marked edge and the second marked edge is determined based on the non-co-circular factor between the first marked edge and the second marked edge and the difference between the reference ratio corresponding to the first marked edge and the reference ratio corresponding to the second marked edge.

[0046] In conjunction with the first aspect above, in a possible implementation, a formula corresponding to the co-loop coefficient between the first marked edge and the second marked edge is:

[0047] ;in, is the co-loop coefficient between the first marked edge and the second marked edge; It is an exponential function with a natural constant as its base; is the non-co-loop factor between the first labeled edge and the second labeled edge; It is the absolute value function; is the reference ratio corresponding to the first marked edge; is the reference ratio corresponding to the second marked edge.

[0048] In combination with the first aspect above, in a possible implementation, screening out a co-ring edge cluster from each channel cartilage edge cluster based on the co-ring coefficients between different real cartilage edges in each channel cartilage edge cluster includes:

[0049] Determining any channel cartilage edge cluster as a marked cartilage edge cluster, and determining any real cartilage edge in the marked cartilage edge cluster as a temporary cartilage edge;

[0050] Screening out a real cartilage edge having the largest co-circular coefficient with the provisional cartilage edge from the marked cartilage edge cluster as the first edge corresponding to the provisional cartilage edge;

[0051] The real cartilage edges with the same corresponding first edges in the marked cartilage edge clusters are used to form a possible co-ring cluster;

[0052] If the real cartilage edges in the possible co-ring cluster are on different sides, the possible co-ring cluster is determined to be a co-ring edge cluster.

[0053] In conjunction with the first aspect above, in a possible implementation, determining the incomplete factor corresponding to each interval line according to the distribution of each interval line includes:

[0054] All the co-ring edge clusters in each channel cartilage edge cluster form a co-ring edge cluster set;

[0055] Determine any interval line as a marking interval line, and select the interval line closest to the marking interval line from each co-loop edge cluster in the co-loop edge cluster set to which the marking interval line belongs, to form a first reference interval line set corresponding to the marking interval line;

[0056] sorting the interval lines in the first reference interval line set in ascending order according to the lengths of the interval lines in the first reference interval line set to obtain a first reference interval line sequence;

[0057] Sorting the interval lines in the first reference interval line set in order from top to bottom and from left to right according to positions of the interval lines in the current bronchoscope image to obtain a first temporary interval line sequence;

[0058] Determine the sorting difference between the first reference spacer line sequence and the first temporary spacer line sequence as the first quantity difference corresponding to the mark spacer lines;

[0059] All the interval lines in the first reference interval line set corresponding to the mark interval line, except the mark interval line, constitute a second reference interval line set corresponding to the mark interval line;

[0060] sorting the interval lines in the second reference interval line set in ascending order according to the lengths of the interval lines in the second reference interval line set to obtain a second reference interval line sequence;

[0061] Sorting the interval lines in the second reference interval line set in order from top to bottom and from left to right according to the positions of the interval lines in the current bronchoscope image to obtain a second temporary interval line sequence;

[0062] determining the sorting difference between the second reference spacer line sequence and the second temporary spacer line sequence as the second quantity difference corresponding to the mark spacer lines;

[0063] Determine the ray from the centroid of the connected domain of the channel to which the marker interval line belongs to the centroid of the marker interval line as the reference ray corresponding to the marker interval line;

[0064] Determine the distance between the centroid of each interval line in the second reference interval line set and the reference ray as a reference distance coefficient corresponding to each interval line in the second reference interval line set;

[0065] Determine the average value of the reference distance coefficients corresponding to all the interval lines in the second reference interval line set as the average reference distance corresponding to the marked interval line;

[0066] An incomplete factor corresponding to the marked spaced lines is determined according to the first quantity difference, the second quantity difference, and the average reference distance corresponding to the marked spaced lines.

[0067] In a second aspect, the present invention provides a system for accurately locating lesion areas in bronchoscopic images, the system comprising:

[0068] An acquisition, extraction, and screening module is used to acquire the current bronchoscope image, perform connected domain extraction on the current bronchoscope image, and screen out channel connected domains and cartilage candidate edges from the obtained connected domains;

[0069] a determination and screening module, configured to determine the channel connected domain closest to the cartilage candidate edge as the representative connected domain corresponding to the cartilage candidate edge, and screen out the true cartilage edge from all cartilage candidate edges whose corresponding representative connected domains are the same connected domain;

[0070] Constructing a determination module for forming a channel cartilage edge cluster from corresponding real cartilage edges representing the same connected domain, and determining a co-circularity coefficient between every two real cartilage edges in each channel cartilage edge cluster;

[0071] a screening and identification module, configured to screen out a co-ring edge cluster from each channel cartilage edge cluster according to a co-ring coefficient between different real cartilage edges in each channel cartilage edge cluster, and identify a septal line from each co-ring edge cluster;

[0072] A dual determination module is used to determine the incomplete factor corresponding to each interval line according to the distribution of each interval line, and if the incomplete factor corresponding to the interval line is greater than a preset incomplete threshold, the interval line is determined to be an incomplete interval line;

[0073] a consistency index determination module, configured to determine a consistency index corresponding to each incomplete interval line based on a current bronchoscopic image and a previous bronchoscopic image;

[0074] The judgment and determination module is used to determine the incomplete interval line as a lesion interval line if the consistency index corresponding to the incomplete interval line is greater than a preset consistency threshold.

[0075] In a third aspect, a server is provided, comprising a memory and a processor. The memory is configured to store executable program code, and the processor is configured to call and execute the executable program code from the memory, so that the device executes the method of the first aspect or any possible implementation of the first aspect.

[0076] In a fourth aspect, a computer program product is provided, comprising: a computer program code, which, when executed on a computer, enables the computer to execute the method in the first aspect or any possible implementation of the first aspect.

[0077] In a fifth aspect, a computer-readable storage medium is provided, which stores a computer program code. When the computer program code runs on a computer, the computer executes the method in the above-mentioned first aspect or any possible implementation of the first aspect.

[0078] The present invention has the following beneficial effects:

[0079] The present invention provides a bronchoscopic lesion localization method based on reinforcement learning, which realizes the localization of lesion areas in bronchoscopic images, solves the technical problem of poor accuracy in localization of lesion areas in bronchoscopic images, and improves the accuracy of localization of lesion areas in bronchoscopic images. Compared with localizing lesion areas in bronchoscopic images by grayscale values, the present invention comprehensively considers multiple features related to the lesion conditions in bronchoscopic images, such as the co-circularity coefficient, the incompleteness factor, and the consistency index, thereby improving the accuracy of quantification of lesion interval lines, and further improving the accuracy of localization of lesion areas in bronchoscopic images. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0081] Figure 1 This is a flow chart of a bronchoscopic lesion localization method based on reinforcement learning according to the present invention;

[0082] Figure 2 A schematic diagram of the structure of a system for accurately locating lesion areas in bronchoscopic images according to the present invention;

[0083] Figure 3 The figure is a structural diagram of a computer device of the present invention. DETAILED DESCRIPTION

[0084] To further illustrate the technical means and effects employed by the present invention to achieve its intended objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementations, structures, features, and effects of the technical solutions proposed by the present invention. In the following description, references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0085] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0086] In order to facilitate doctors to identify the morphology and location of lesions in the bronchi, machine vision is used to accurately locate the lesions in bronchoscopic images. However, since there are many cartilage-like protrusions on the bronchial wall in bronchoscopic images and the changes in the cartilage-like morphology and channels may be confused with the impact of the lesion on the bronchial wall, it is more difficult to locate the lesion.

[0087] The advantage of the present invention over the prior art is that in order to accurately locate the lesion area in the bronchoscope image, the method uses a bronchoscope to collect images of the patient's bronchi, preprocesses them, and segments the bronchial connected areas, extracts the bronchial cartilage edges through the grayscale changes in the image, and detects lesions based on the integrity of the tube edge, providing good data support for subsequent lesion positioning.

[0088] refer to Figure 1 , shows the process of some embodiments of a bronchoscopic lesion localization method based on reinforcement learning according to the present invention. The bronchoscopic lesion localization method based on reinforcement learning includes the following steps:

[0089] Step S1: obtaining a current bronchoscope image, performing connected domain extraction on the current bronchoscope image, and screening out channel connected domains and cartilage candidate edges from the obtained connected domains.

[0090] The current bronchoscopic image may be a bronchoscopic image acquired at the current moment.

[0091] As an example, this step may include the following steps:

[0092] The first step is to obtain the current bronchoscopic image.

[0093] It should be noted that before a bronchoscopy, a series of imaging tests, such as chest X-rays or CT (Computed Tomography) scans, are typically performed to help doctors understand the overall condition of the airway and preliminarily locate possible lesions. After the possible lesions are initially identified, local or general anesthesia is administered to alleviate patient discomfort. Local anesthesia desensitizes the airway and reduces discomfort. After anesthesia is complete, the doctor inserts a bronchoscope through the mouth or nose, down the trachea, and into the bronchi. The camera captures images of the airway in real time, which are known as bronchoscopic images. These real-time bronchoscopic images can be recorded and stored for subsequent diagnosis and treatment.

[0094] The second step is to extract the connected domain of the current bronchoscopic image.

[0095] For example, a connected domain extraction algorithm may be used to extract connected domains from the current bronchoscope image to obtain multiple connected domains.

[0096] It should be noted that canny edge detection can also be performed on the patient's bronchial image. Since there is a large difference between the acquisition platform interface and the bronchial image, canny edge detection can well detect the edge of the patient's bronchial image, so the partially connected area of ​​the bronchus and several edges inside it can be extracted.

[0097] In the third step, threshold segmentation is performed on all connected domains in the current bronchoscope image, and the connected domain in the connected domain with the lower grayscale value among the two types of connected domains obtained by segmentation is determined as the channel connected domain.

[0098] It should be noted that the obtained bronchial connected domain can be segmented using Otsu threshold segmentation. Since the inter-class difference obtained by Otsu threshold segmentation is the largest, the bronchial connected domain can be segmented into the bronchial wall with higher grayscale values ​​and the channel part with lower grayscale values. The channel part with lower grayscale values ​​is the channel connected domain.

[0099] It should be noted that for the bronchial connectivity images of patients obtained by the above operation, since the structure of the bronchi is composed of a combination of incomplete C-shaped cartilage rings, the cartilage part can keep the trachea unobstructed, so the bronchi are rings in the bronchoscopic image; and the presence of lesions in the bronchi will cause changes in the annular structure of the bronchi, so the lesions can be identified subsequently by analyzing the annular structure of the bronchi.

[0100] The fourth step, screening reference pixels from the connected domain in the current bronchoscopic image, may include the following sub-steps:

[0101] In the first sub-step, any connected domain in the current bronchoscope image is determined as a marked connected domain, and any pixel point in the marked connected domain is determined as a marked pixel point.

[0102] In the second sub-step, a pixel point having the smallest gradient value difference with the marked pixel point is selected from a preset neighborhood corresponding to the marked pixel point as the first pixel point corresponding to the marked pixel point.

[0103] The preset neighborhood may be a pre-set neighborhood, which may be an eight-neighborhood neighborhood.

[0104] In the third sub-step, a pixel point with the largest corresponding gradient value is selected from the preset neighborhood corresponding to the marked pixel point as the second pixel point corresponding to the marked pixel point.

[0105] In a fourth sub-step, if the first pixel point and the second pixel point corresponding to the marked pixel point are the same pixel point, the marked pixel point is determined as a reference pixel point.

[0106] In the fifth step, a reference pixel point whose corresponding first pixel point is also a reference pixel point is selected from all reference pixel points as a selected pixel point.

[0107] The selected pixel point may be a reference pixel point, and the first pixel point corresponding to the selected pixel point is also a reference pixel point.

[0108] In the sixth step, the line formed by the continuous selected pixel points is determined as the candidate cartilage edge.

[0109] The number of pixels selected from the candidate cartilage edge is at least 1. The candidate cartilage edge is also called a selection line. The obtained selection line includes the cartilage edge in the bronchoscopic image and the edges of some other tissues. Therefore, the cartilage edge needs to be selected based on its circular arrangement.

[0110] It should be noted that since the C-shaped cartilage combination of the bronchus will cause a tubular structure with an annular protrusion to form inside the bronchus, the cartilage edge in a normal bronchoscopic image will appear in a C-shaped superimposed arrangement. Therefore, it is necessary to obtain the cartilage edge in the image based on the arrangement of the edges of the cartilage part.

[0111] In step S2, the channel connected domain closest to the cartilage candidate edge is determined as the representative connected domain corresponding to the cartilage candidate edge, and the real cartilage edge is screened out from all cartilage candidate edges whose corresponding representative connected domains are the same connected domain.

[0112] The channel connected domain closest to the cartilage candidate edge may be a channel connected domain whose corresponding centroid is closest to the centroid of the cartilage candidate edge.

[0113] As an example, this step may include the following steps:

[0114] In the first step, the representative connected domain corresponding to any cartilage candidate edge is determined as the reference connected domain.

[0115] In the second step, different numbers of cartilage candidate edges are randomly selected from all cartilage candidate edges whose corresponding representative connected domains are the above-mentioned reference connected domains by random sampling, and the cartilage candidate edges selected each time are formed into a cartilage candidate edge set to obtain multiple cartilage candidate edge sets under the above-mentioned reference connected domains.

[0116] For example, if the corresponding representative connected domain is the cartilage candidate edge of the reference connected domain, and the cartilage candidate edges include: the first cartilage candidate edge, the second cartilage candidate edge and the third cartilage candidate edge in sequence, and the randomly selected numbers are 2 and 3, then 4 cartilage candidate edge sets under the reference connected domain can be obtained, which can be {first cartilage candidate edge, second cartilage candidate edge}, {first cartilage candidate edge, third cartilage candidate edge}, {second cartilage candidate edge, third cartilage candidate edge} and {first cartilage candidate edge, second cartilage candidate edge, third cartilage candidate edge} in sequence.

[0117] In the third step, the cartilage candidate edges in each cartilage candidate edge set are divided into the same side to obtain respective cartilage candidate edge groups corresponding to each cartilage candidate edge set.

[0118] It should be noted that other cartilage candidate edges on the line connecting the centroid of the cartilage candidate edge and the centroid of the channel connected domain are recorded as cartilage edges on the same side.

[0119] In the fourth step, the information entropy of the distance between the centroids of all cartilage candidate edges and the centroids of their corresponding representative connected domains in each cartilage candidate edge group is determined as the distance information entropy corresponding to each cartilage candidate edge group.

[0120] In the fifth step, the mean of the DTW distances between different cartilage candidate edges in each cartilage candidate edge group is determined as the target difference factor corresponding to each cartilage candidate edge group.

[0121] Among them, DTW distance is the corresponding shortest path obtained using the DTW algorithm.

[0122] In the sixth step, the target loss value corresponding to each cartilage candidate edge set is determined based on all distance information entropies and target difference factors corresponding to each cartilage candidate edge set.

[0123] For example, the formula for determining the target loss value corresponding to the set of cartilage candidate edges can be:

[0124] ; Where E is the target loss value corresponding to the set of cartilage candidate edges. is a hyperbolic function. is the variance of all distance information entropies corresponding to the set of cartilage candidate edges. is the mean of all target difference factors corresponding to the set of cartilage candidate edges.

[0125] It should be noted that when The smaller the value, the closer the edge distance information entropy values ​​between the cartilage candidate edge groups are, and the closer the edge distance information entropy values ​​between the cartilage candidate edge groups are, the closer the edge distance information entropy values ​​between the cartilage candidate edge groups are. The smaller the value, the more likely it is that the morphologies between the edges in the cartilage edge cluster are similar.

[0126] In the seventh step, the cartilage candidate edge set with the smallest target loss value is selected from all the cartilage candidate edge sets under the above reference connected domain as the temporary edge set, and each cartilage candidate edge in the temporary edge set is determined as the true cartilage edge.

[0127] In step S3, the corresponding real cartilage edges representing the same connected domain are grouped into a channel cartilage edge cluster, and the co-circular coefficient between every two real cartilage edges in each channel cartilage edge cluster is determined.

[0128] It should be noted that for several cartilage edges in the bronchoscopic images obtained by the above operation, there may be cases where the annular edges are disconnected due to lesions on the bronchial surface. However, due to the C-shaped structure of the cartilage in the bronchus, the cartilage edge in the bronchus itself is not a closed circular structure. In addition, due to the shooting angle of the bronchoscope, the tracheal annular structure near the bottom is blocked by the annular structure above. Therefore, it is necessary to analyze the actual integrity of the cartilage edge.

[0129] As an example, this step may include the following steps:

[0130] In the first step, any real cartilage edge is determined as the marked cartilage edge, and the centroid of the connected domain corresponding to the marked cartilage edge is used as the endpoint. A ray passing through the centroid of the marked cartilage edge is recorded as the target ray corresponding to the marked cartilage edge.

[0131] In the second step, the intersection of the target ray corresponding to the marked cartilage edge and the connected domain is determined as the target line segment corresponding to the marked cartilage edge.

[0132] The third step is to select the real cartilage edge farthest from the centroid of the marked cartilage edge from all the real cartilage edges intersecting the target line segment corresponding to the marked cartilage edge as the reference edge corresponding to the marked cartilage edge.

[0133] In the fourth step, the distance between the centroid of the marked cartilage edge and the centroid of the corresponding representative connected domain is determined as the first distance factor corresponding to the marked cartilage edge.

[0134] In the fifth step, the distance between the centroid of the reference edge corresponding to the marked cartilage edge and the centroid of the representative connected domain corresponding to the marked cartilage edge is determined as the second distance factor corresponding to the marked cartilage edge.

[0135] In the sixth step, the ratio of the first distance factor and the second distance factor corresponding to the marked cartilage edge is determined as the reference ratio corresponding to the marked cartilage edge.

[0136] The seventh step is to determine any channel cartilage edge cluster as a marked cartilage edge cluster, and determine the co-circular coefficient between each two real cartilage edges in the marked cartilage edge cluster based on the reference ratio corresponding to each two real cartilage edges in the marked cartilage edge cluster, which may include the following sub-steps:

[0137] In the first sub-step, any two real cartilage edges in the marked cartilage edge cluster are respectively determined as the first marked edge and the second marked edge.

[0138] In the second sub-step, an ellipse fitting is performed on the union of the first marked edge and the second marked edge, and a fitting loss obtained by the ellipse fitting is determined as a non-co-loop factor between the first marked edge and the second marked edge.

[0139] The third sub-step is to determine the co-circular coefficient between the first marked edge and the second marked edge based on the non-co-circular factor between the first marked edge and the second marked edge, and the difference between the reference ratio corresponding to the first marked edge and the reference ratio corresponding to the second marked edge.

[0140] For example, the formula for determining the co-circular coefficient between the first marked edge and the second marked edge may be:

[0141] ;in, is the co-circularity coefficient between the first marked edge and the second marked edge. It is an exponential function with a natural constant as its base. is the non-cocyclic factor between the first labeled edge and the second labeled edge. It is the absolute value function. is the reference ratio corresponding to the first marked edge. is the reference ratio corresponding to the second marked edge.

[0142] It should be noted that when The larger it is, the more likely the edges belong to the same ellipse and are located in the same layer in the channel, so the edge co-ring coefficient is higher.

[0143] Step S4, based on the co-circular coefficients between different real cartilage edges in each channel cartilage edge cluster, screen out co-circular edge clusters from each channel cartilage edge cluster, and identify interval lines from each co-circular edge cluster.

[0144] It should be noted that, since there are regularly arranged C-shaped cricoid cartilages in normal bronchus, regularly arranged gaps will normally appear in the cartilage edge. Therefore, the gaps in the cricoid edge can be used to analyze the integrity of the cartilage edge.

[0145] As an example, this step may include the following steps:

[0146] In the first step, any channel cartilage edge cluster is determined as a marked cartilage edge cluster, and any real cartilage edge in the marked cartilage edge cluster is determined as a temporary cartilage edge.

[0147] In the second step, the real cartilage edge with the largest co-circular coefficient with the temporary cartilage edge is selected from the marked cartilage edge cluster as the first edge corresponding to the temporary cartilage edge.

[0148] In the third step, the real cartilage edges with the same corresponding first edge in the above-mentioned marked cartilage edge clusters are combined to form a possible co-ring cluster.

[0149] In the fourth step, if the real cartilage edges in the possible co-ring cluster are on different sides, the possible co-ring cluster is determined as a co-ring edge cluster.

[0150] Among them, different sides of the true cartilage edges often indicate that these true cartilage edges are not on the same side of the channel.

[0151] In the fifth step, the interval lines are identified from each co-ring edge cluster.

[0152] For example, endpoints between adjacent edges in each co-ring edge cluster can be obtained, and the optimal fitting line between the endpoints can be obtained by fitting an ellipse using the least squares method, thereby obtaining a total of several interval lines of the co-ring edge clusters.

[0153] Step S5: determining the incomplete factor corresponding to each interval line according to the distribution of each interval line; if the incomplete factor corresponding to the interval line is greater than a preset incomplete threshold, the interval line is determined as an incomplete interval line.

[0154] The preset incomplete threshold may be a preset threshold, which may be 0.7.

[0155] As an example, this step may include the following steps:

[0156] In the first step, all co-ring edge clusters in each channel cartilage edge cluster are combined to form a co-ring edge cluster set.

[0157] In the second step, any interval line is determined as the marked interval line, and the interval line closest to the marked interval line is selected from each co-loop edge cluster in the co-loop edge cluster set to which the marked interval line belongs to form a first reference interval line set corresponding to the marked interval line.

[0158] The third step is to sort the interval lines in the first reference interval line set in ascending order according to the lengths of the interval lines in the first reference interval line set to obtain a first reference interval line sequence.

[0159] The fourth step is to sort the interval lines in the first reference interval line set according to their positions in the current bronchoscope image, in order from top to bottom and from left to right, to obtain a first temporary interval line sequence.

[0160] In the fifth step, the order difference between the first reference interval line sequence and the first temporary interval line sequence is determined as the first quantity difference corresponding to the above-mentioned mark interval lines.

[0161] For example, if the first reference interval line sequence is {first interval line, second interval line, third interval line, fourth interval line, fifth interval line}, and the first temporary interval line sequence is {first interval line, third interval line, fifth interval line, fourth interval line, second interval line}, then the sorting difference between the two sequences can be 3, that is, the first quantity difference corresponding to the marked interval lines can be 3.

[0162] In the sixth step, all the interval lines in the first reference interval line set corresponding to the above-mentioned mark interval lines except the above-mentioned mark interval lines are used to form a second reference interval line set corresponding to the above-mentioned mark interval lines.

[0163] Among them, the interval lines in the second reference interval line set can be called reference interval lines.

[0164] In the seventh step, the interval lines in the second reference interval line set are sorted in ascending order according to the lengths of the interval lines in the second reference interval line set to obtain a second reference interval line sequence.

[0165] In the eighth step, the interval lines in the second reference interval line set are sorted in order from top to bottom and from left to right according to their positions in the current bronchoscope image to obtain a second temporary interval line sequence.

[0166] In the ninth step, the order difference between the second reference spacer sequence and the second temporary spacer sequence is determined as the second quantity difference corresponding to the above-mentioned mark spacers.

[0167] In the tenth step, the ray from the centroid of the connected domain of the channel to which the mark interval line belongs to the centroid of the mark interval line is determined as the reference ray corresponding to the mark interval line.

[0168] In the eleventh step, the distance between the centroid of each interval line in the second reference interval line set and the reference ray is determined as the reference distance coefficient corresponding to each interval line in the second reference interval line set.

[0169] In the twelfth step, the average value of the reference distance coefficients corresponding to all the interval lines in the second reference interval line set is determined as the average reference distance corresponding to the above-mentioned marked interval lines.

[0170] In the thirteenth step, the incomplete factor corresponding to the marked interval line is determined based on the first quantity difference, the second quantity difference and the average reference distance corresponding to the marked interval line.

[0171] For example, the formula for determining the incomplete factor corresponding to the marked interval line can be:

[0172] ;in, is the incomplete factor corresponding to the marked interval line. is a hyperbolic function. is the average reference distance corresponding to the marked interval lines. is the first quantity difference corresponding to the marked interval line. is the second quantity difference corresponding to the marked interval line.

[0173] It should be noted that when The smaller the value, the more likely it is that the interval line and its reference interval line mean extend outward from the channel centroid. The smaller it is, the closer the arrangement of the interval lines is to the normal arrangement, and therefore the smaller the incomplete factor of the interval lines is.

[0174] In step 14, if the incomplete factor corresponding to the interval line is greater than a preset incomplete threshold, the interval line is determined as an incomplete interval line.

[0175] The preset incomplete threshold may be a preset threshold, which may be 0.7.

[0176] For example, the incomplete factor of the interval line may be compared with a preset threshold of 0.7 to obtain interval lines whose incomplete factors are greater than the threshold, and mark them, thereby obtaining incomplete cartilage edges.

[0177] It should be noted that the marked interval lines in the incomplete cartilage edge may be caused by the light source on the surface of the bronchoscope forming a reflected light spot on the bronchial surface, or by the morphology of the cartilage surface tissue changed by the lesion of the bronchial wall.

[0178] Step S6: determining a consistency index corresponding to each incomplete interval line based on the current bronchoscopic image and the previous bronchoscopic image.

[0179] As an example, this step may include the following steps:

[0180] In the first step, the optical flow detection method is used to screen out the area corresponding to each incomplete interval line from the previous frame of bronchoscopic image as the historical interval line corresponding to each incomplete interval line, and the area corresponding to the co-ring edge cluster to which each incomplete interval line belongs is screened out from the previous frame of bronchoscopic image as the historical co-ring edge cluster corresponding to each incomplete interval line.

[0181] For example, corner detection can be used to obtain key points on the edges of each co-ring edge cluster, and optical flow detection can be used to obtain the optical flow vectors of the key points in the current frame image and the previous frame image; based on the optical flow vectors combined with the above operations, the position correspondence in different frame images can be achieved.

[0182] In the second step, the consistency index corresponding to each incomplete interval line is determined based on the DTW distance between each incomplete interval line and its corresponding historical interval line, as well as the area ratio between the historical co-ring edge cluster corresponding to each incomplete interval line and its corresponding co-ring edge cluster.

[0183] It should be noted that since the incomplete septal lines generated by the lesion will change in the same way as the surrounding cartilage as the lens moves, the incomplete septal lines will change in the same way as the co-annular edge clusters, thus obtaining the consistency of each incomplete septal line.

[0184] For example, the formula for determining the consistency index corresponding to the incomplete interval line can be:

[0185] ;in, It is the consistency index corresponding to the incomplete interval line. It is an exponential function with a natural constant as its base. is the DTW distance between the incomplete interval line and its corresponding historical interval line. It is the absolute value function. It is the area ratio between the historical co-ring edge cluster corresponding to the incomplete interval line and the co-ring edge cluster to which it belongs. is the ratio of the lengths of the incomplete interval line to its corresponding historical interval line.

[0186] It should be noted that when The smaller it is, the more similar the incomplete interval line is to its corresponding historical interval line. The smaller the value, the more similar the changes of the incomplete septal line are to those of the cartilage edge, and thus the greater their consistency.

[0187] Step S7: If the consistency index corresponding to the incomplete interval line is greater than the preset consistency threshold, the incomplete interval line is determined as the lesion interval line.

[0188] The preset consistency threshold may be a pre-set threshold, which may be 0.5.

[0189] It should be noted that the consistency of the incomplete septal lines obtained by the above operation can be compared with a preset threshold of 0.5 to obtain incomplete septal lines with a consistency greater than the threshold and mark them; the incomplete septal lines are subjected to regional growth, wherein adjacent incomplete septal lines grow into the same area and cannot destroy the original cartilage edge, thereby obtaining several lesion areas. Accurately locating the lesion area in the bronchoscopic image obtained by the above operation can include: marking the lesion area in the bronchoscopic image obtained by the above operation in the image to obtain coordinates, and continuously changing the coordinates of the lesion as the lens moves to locate it; after locating the lesion, lesion biopsy is a common operation, the purpose of which is to obtain tissue samples for pathological examination to determine the nature of the lesion; the pathological results are usually fed back to the patient through outpatient or resident doctors, and subsequent treatment or observation plans are discussed.

[0190] refer to Figure 2 Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a system for accurately locating lesion areas in bronchoscopic images. The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of a bronchoscopic lesion localization method based on reinforcement learning may specifically include:

[0191] An acquisition, extraction, and screening module 201 is configured to acquire a current bronchoscopic image, perform connected domain extraction on the current bronchoscopic image, and screen out channel connected domains and cartilage candidate edges from the obtained connected domains;

[0192] a determination and screening module 202 for determining the channel connected domain closest to the cartilage candidate edge as the representative connected domain corresponding to the cartilage candidate edge, and screening out the true cartilage edge from all cartilage candidate edges whose corresponding representative connected domains are the same connected domain;

[0193] A construction determination module 203 is configured to form a channel cartilage edge cluster from corresponding real cartilage edges representing the same connected domain, and determine a common loop coefficient between every two real cartilage edges in each channel cartilage edge cluster;

[0194] a screening and identification module 204 for screening out a co-circular edge cluster from each channel cartilage edge cluster based on the co-circular coefficients between different real cartilage edges in each channel cartilage edge cluster, and identifying a separation line from each co-circular edge cluster;

[0195] A dual determination module 205 is configured to determine an incomplete factor corresponding to each interval line according to the distribution of each interval line, and to determine the interval line as an incomplete interval line if the incomplete factor corresponding to the interval line is greater than a preset incomplete threshold;

[0196] a consistency index determination module 206 for determining a consistency index corresponding to each incomplete interval line based on the current bronchoscopic image and the previous bronchoscopic image;

[0197] The judgment and determination module 207 is configured to determine the incomplete interval line as a lesion interval line if the consistency index corresponding to the incomplete interval line is greater than a preset consistency threshold.

[0198] Figure 3 FIG. 1 is a schematic diagram of the structure of a computer device provided by an embodiment of the present invention. For example, Figure 3 As shown, the computer device 300 includes: a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and running on the processor 302, wherein when the processor 302 executes the computer program 303, the computer device can execute any one of the bronchoscopic lesion localization methods based on reinforcement learning introduced above.

[0199] Based on the same inventive concept as the above-described method embodiments, the present invention provides a server comprising a memory and a processor. The memory is configured to store executable program code, and the processor is configured to retrieve and execute the executable program code from the memory, thereby enabling the device to perform any of the above-described methods for bronchoscopic lesion localization based on reinforcement learning.

[0200] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a computer program product, which includes: computer program code, which, when running on a computer, enables the computer to execute any one of the above-mentioned bronchoscopic lesion localization methods based on reinforcement learning.

[0201] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a computer-readable storage medium, which stores a computer program code. When the computer program code runs on a computer, the computer executes any one of the above-mentioned bronchoscopic lesion localization methods based on reinforcement learning.

[0202] In summary, compared with locating the lesion area of ​​the bronchoscopic image by the difference in grayscale values, the present invention comprehensively considers multiple features related to the lesion conditions of the bronchoscopic image, such as the co-circular coefficient, the incompleteness factor and the consistency index, thereby improving the accuracy of the quantification of the lesion interval line, and further improving the accuracy of the locating of the lesion area of ​​the bronchoscopic image.

[0203] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A bronchoscopic lesion localization method based on reinforcement learning, characterized in that: The following steps are involved: Acquire the current bronchoscopic image, perform connected domain extraction on the current bronchoscopic image, and screen out channel connected domains and cartilage candidate edges from the obtained connected domains; The channel connected domain closest to the cartilage candidate edge is determined as the representative connected domain corresponding to the cartilage candidate edge, and the real cartilage edge is screened out from all the cartilage candidate edges whose corresponding representative connected domains are the same connected domain; The corresponding real cartilage edges representing the same connected domain are used to form a channel cartilage edge cluster, and the co-circularity coefficient between every two real cartilage edges in each channel cartilage edge cluster is determined; According to the co-circularity coefficient between different real cartilage edges in each channel cartilage edge cluster, a co-circular edge cluster is screened from each channel cartilage edge cluster, and a septal line is identified from each co-circular edge cluster; According to the distribution of each interval line, an incomplete factor corresponding to each interval line is determined; if the incomplete factor corresponding to the interval line is greater than a preset incomplete threshold, the interval line is determined as an incomplete interval line; Determine the consistency index corresponding to each incomplete interval line based on the current bronchoscopic image and the previous bronchoscopic image; If the consistency index corresponding to the incomplete interval line is greater than the preset consistency threshold, the incomplete interval line is determined to be the lesion interval line.

2. The bronchoscopic lesion localization method based on reinforcement learning according to claim 1, characterized in that: The step of screening the channel connected domain and the cartilage candidate edge from the obtained connected domain includes: Performing threshold segmentation on all connected domains in the current bronchoscope image, and determining a connected domain in a type of connected domain with a lower grayscale value among the two types of connected domains obtained by segmentation as a channel connected domain; Filtering reference pixel points from a connected domain in the current bronchoscope image; Filter out from all reference pixels a reference pixel whose corresponding first pixel is also a reference pixel as a selected pixel; The line formed by the continuous selected pixel points is determined as the candidate cartilage edge.

3. The bronchoscopic lesion localization method based on reinforcement learning according to claim 2, characterized in that: The step of selecting reference pixels from a connected domain in the current bronchoscope image includes: Determining any connected domain in the current bronchoscope image as a marked connected domain, and determining any pixel point in the marked connected domain as a marked pixel point; Filtering a pixel point with the smallest gradient difference with the marked pixel point from a preset neighborhood corresponding to the marked pixel point as the first pixel point corresponding to the marked pixel point; Filtering out a pixel point with the largest corresponding gradient value from a preset neighborhood corresponding to the marked pixel point as a second pixel point corresponding to the marked pixel point; If the first pixel point and the second pixel point corresponding to the marked pixel point are the same pixel point, the marked pixel point is determined as the reference pixel point.

4. The bronchoscopic lesion localization method based on reinforcement learning according to claim 1, characterized in that: The step of selecting the true cartilage edge from all the cartilage candidate edges whose corresponding representative connected domains are the same connected domain comprises: The representative connected domain corresponding to any cartilage candidate edge is determined as the reference connected domain; By random sampling, a different number of cartilage candidate edges are randomly extracted from all cartilage candidate edges whose corresponding representative connected domain is the reference connected domain, and the cartilage candidate edges extracted each time are formed into a cartilage candidate edge set, thereby obtaining multiple cartilage candidate edge sets under the reference connected domain; Performing ipsilateral division on the cartilage candidate edges in each cartilage candidate edge set to obtain respective cartilage candidate edge groups corresponding to each cartilage candidate edge set; The information entropy of the distance between the centroids of all cartilage candidate edges and the centroids of their corresponding representative connected domains in each cartilage candidate edge group is determined as the distance information entropy corresponding to each cartilage candidate edge group; The mean of the DTW distances between different cartilage candidate edges in each cartilage candidate edge group is determined as the target difference factor corresponding to each cartilage candidate edge group; Determine the target loss value corresponding to each cartilage candidate edge set based on all distance information entropies and target difference factors corresponding to each cartilage candidate edge set; A cartilage candidate edge set with the smallest corresponding target loss value is screened out from all cartilage candidate edge sets under the reference connected domain as a temporary edge set, and each cartilage candidate edge in the temporary edge set is determined as a true cartilage edge.

5. The bronchoscopic lesion localization method based on reinforcement learning according to claim 4, characterized in that: The formula for the target loss value corresponding to the set of cartilage candidate edges is: ; Where E is the target loss value corresponding to the set of cartilage candidate edges; is a hyperbolic function; is the variance of all distance information entropies corresponding to the set of cartilage candidate edges; is the mean of all target difference factors corresponding to the set of cartilage candidate edges.

6. The bronchoscopic lesion localization method based on reinforcement learning according to claim 1, characterized in that: Determining the co-circularity coefficient between each two real cartilage edges in each channel cartilage edge cluster comprises: Any real cartilage edge is determined as the marked cartilage edge, and a ray passing through the mass center of the representative connected domain corresponding to the marked cartilage edge is taken as the endpoint, and recorded as the target ray corresponding to the marked cartilage edge; Determine the intersection of the target ray corresponding to the marked cartilage edge and the representative connected domain as the target line segment corresponding to the marked cartilage edge; From all real cartilage edges intersecting the target line segment corresponding to the marked cartilage edge, screen out the real cartilage edge farthest from the centroid of the marked cartilage edge as the reference edge corresponding to the marked cartilage edge; Determine the distance between the centroid of the marked cartilage edge and the centroid of the corresponding representative connected domain as a first distance factor corresponding to the marked cartilage edge; Determine the distance between the centroid of the reference edge corresponding to the marked cartilage edge and the centroid of the representative connected domain corresponding to the marked cartilage edge as a second distance factor corresponding to the marked cartilage edge; Determining the ratio of the first distance factor and the second distance factor corresponding to the marked cartilage edge as the reference ratio corresponding to the marked cartilage edge; Any channel cartilage edge cluster is determined as a marked cartilage edge cluster, and the co-circular coefficient between each two real cartilage edges in the marked cartilage edge cluster is determined based on the reference ratio corresponding to each two real cartilage edges in the marked cartilage edge cluster.

7. The bronchoscopic lesion localization method based on reinforcement learning according to claim 6, characterized in that: Determining the co-circularity coefficient between each two real cartilage edges in the marked cartilage edge cluster according to the reference ratio corresponding to each two real cartilage edges in the marked cartilage edge cluster comprises: Determine any two real cartilage edges in the marked cartilage edge cluster as a first marked edge and a second marked edge respectively; Performing ellipse fitting on the union of the first marked edge and the second marked edge, and determining a fitting loss obtained by the ellipse fitting as a non-co-loop factor between the first marked edge and the second marked edge; The co-circular coefficient between the first marked edge and the second marked edge is determined based on the non-co-circular factor between the first marked edge and the second marked edge and the difference between the reference ratio corresponding to the first marked edge and the reference ratio corresponding to the second marked edge.

8. The bronchoscopic lesion localization method based on reinforcement learning according to claim 7, characterized in that: The formula for the co-circular coefficient between the first marked edge and the second marked edge is: ;in, is the co-loop coefficient between the first marked edge and the second marked edge; It is an exponential function with a natural constant as its base; is the non-co-loop factor between the first labeled edge and the second labeled edge; It is the absolute value function; is the reference ratio corresponding to the first marked edge; is the reference ratio corresponding to the second marked edge.

9. The bronchoscopic lesion localization method based on reinforcement learning according to claim 1, characterized in that: The method of screening out a co-ring edge cluster from each channel cartilage edge cluster according to the co-ring coefficients between different real cartilage edges in each channel cartilage edge cluster comprises: Determining any channel cartilage edge cluster as a marked cartilage edge cluster, and determining any real cartilage edge in the marked cartilage edge cluster as a temporary cartilage edge; Screening out a real cartilage edge having the largest co-circular coefficient with the provisional cartilage edge from the marked cartilage edge cluster as the first edge corresponding to the provisional cartilage edge; The real cartilage edges with the same corresponding first edges in the marked cartilage edge clusters are used to form a possible co-ring cluster; If the real cartilage edges in the possible co-ring cluster are on different sides, the possible co-ring cluster is determined to be a co-ring edge cluster.

10. The bronchoscopic lesion localization method based on reinforcement learning according to claim 1, characterized in that: Determining the incomplete factor corresponding to each interval line according to the distribution of each interval line includes: All the co-ring edge clusters in each channel cartilage edge cluster form a co-ring edge cluster set; Determine any interval line as a marking interval line, and select the interval line closest to the marking interval line from each co-loop edge cluster in the co-loop edge cluster set to which the marking interval line belongs, to form a first reference interval line set corresponding to the marking interval line; sorting the interval lines in the first reference interval line set in ascending order according to the lengths of the interval lines in the first reference interval line set to obtain a first reference interval line sequence; Sorting the interval lines in the first reference interval line set in order from top to bottom and from left to right according to positions of the interval lines in the current bronchoscope image to obtain a first temporary interval line sequence; Determine the sorting difference between the first reference spacer line sequence and the first temporary spacer line sequence as the first quantity difference corresponding to the mark spacer lines; All the interval lines in the first reference interval line set corresponding to the mark interval line, except the mark interval line, constitute a second reference interval line set corresponding to the mark interval line; sorting the interval lines in the second reference interval line set in ascending order according to the lengths of the interval lines in the second reference interval line set to obtain a second reference interval line sequence; Sorting the interval lines in the second reference interval line set in order from top to bottom and from left to right according to the positions of the interval lines in the current bronchoscope image to obtain a second temporary interval line sequence; determining the sorting difference between the second reference spacer line sequence and the second temporary spacer line sequence as the second quantity difference corresponding to the mark spacer lines; Determine the ray from the centroid of the connected domain of the channel to which the marker interval line belongs to the centroid of the marker interval line as the reference ray corresponding to the marker interval line; Determine the distance between the centroid of each interval line in the second reference interval line set and the reference ray as a reference distance coefficient corresponding to each interval line in the second reference interval line set; Determine the average value of the reference distance coefficients corresponding to all the interval lines in the second reference interval line set as the average reference distance corresponding to the marked interval line; An incomplete factor corresponding to the marked spaced lines is determined according to the first quantity difference, the second quantity difference, and the average reference distance corresponding to the marked spaced lines.

Images