CN118691806A - Diseased lymph node image segmentation method and lymphoma auxiliary diagnosis method - Google Patents

Abstract

The present application relates to the technical field of image data processing, and in particular to a diseased lymph node image segmentation method and a lymphoma auxiliary diagnosis method. The diseased lymph node image segmentation method includes: obtaining a PET image sequence of the part to be diagnosed; obtaining a grayscale image of each PET image in the PET image sequence; obtaining the image area of the shadow part of each PET grayscale image to obtain a PET pre-processed image sequence; inputting the PET pre-processed image sequence into a contour change pattern recognition model to identify the image area belonging to the contour change pattern of the lymph node; aligning the PET image and the CT image scanned at the same time; the image area belonging to the contour change pattern of the lymph node corresponds to the image part in the CT image and is set as a region of interest; segmenting the diseased lymph node image according to the region of interest, the CT image and the preset image segmentation model. The technical solution provided by the present application can more accurately segment the diseased lymph node image.

Description

Diseased lymph node image segmentation method and lymphoma auxiliary diagnosis method

Technical Field

The present application relates to the technical field of image data processing, and in particular to a diseased lymph node image segmentation method and a lymphoma auxiliary diagnosis method.

Background Art

In modern medical imaging technology, positron emission tomography (PET) and computed tomography (CT) images play a vital role in the diagnosis and evaluation of various diseases, especially cancer. PET images can detect diseases by measuring the activity of glucose metabolism in the body, because cancer cells usually show a higher glucose absorption rate than normal cells. Therefore, diseased lymph nodes appear particularly prominent in PET images due to their higher glucose metabolism absorption coefficient, which can give accurate hints for the segmentation of diseased lymph nodes.

However, the existing technology still has some problems, which makes it impossible to accurately segment the diseased lymph node images directly based on PET and CT images.

Summary of the invention

In order to solve the above technical problems or at least partially solve the above technical problems, the present application provides a diseased lymph node image segmentation method and a lymphoma auxiliary diagnosis method, which can more accurately segment the diseased lymph node image.

In a first aspect, the present application provides a diseased lymph node image segmentation method, the diseased lymph node image segmentation method comprising the following steps:

Acquire a PET image sequence of the part to be diagnosed of the subject, wherein the PET image sequence includes at least two or more PET images of the part to be diagnosed taken consecutively;

Acquire a grayscale image of each PET image in the PET image sequence;

Using an edge detection algorithm to obtain the image area of the shadow part of each PET grayscale image, and obtain a PET preprocessing image sequence;

Inputting the PET preprocessed image sequence into a contour change pattern recognition model to identify image regions belonging to the contour change pattern of lymph nodes;

The PET image is registered with the simultaneously scanned CT image;

The image area of the contour change pattern belonging to the lymph node in the PET image corresponds to the image part in the CT image and is set as the region of interest;

The diseased lymph node image is segmented and obtained according to the region of interest, the CT image and a preset image segmentation model.

Optionally, the contour change pattern recognition model is trained by the following steps:

Acquire multiple PET image sequences with diseased lymph nodes, and perform grayscale processing on each PET image to obtain a PET grayscale image sequence composed of PET grayscale images;

Using an edge detection algorithm to process the PET grayscale image in each PET grayscale image sequence to obtain a shadow area contained in each PET grayscale image;

Calculate the center point of the shadow area image in each PET grayscale image;

The shadow area images with the same center point position in the PET grayscale image sequence are formed into an image set, and the image set is used as a separate shadow area image sequence;

Each shadow area image sequence is manually labeled to determine whether it is a lymph node;

Using a convolution module to extract features from the shadow area image of each shadow area sequence to obtain a first feature matrix, wherein the convolution module includes at least one convolution layer and one pooling layer;

Calculate the central moment of the shadow area image in each shadow area sequence to obtain a second feature matrix;

The first feature matrix and the second feature matrix are combined to obtain a feature representation of each shadow area image in the shadow area sequence;

The feature representation of all shadow area images in the shadow area sequence is taken as input, the type of shadow area change sequence is used as the label, and a recurrent neural network is used as the neural network framework to train a contour change pattern recognition model for determining whether it belongs to a lymph node.

Optionally, the type of the shaded area change sequence is set according to the normal organs that may appear in the high metabolic area to distinguish the diseased lymph nodes by excluding the normal organs.

Optionally, segmenting the CT image according to the region of interest and a preset image segmentation model to obtain a diseased lymph node image comprises the following steps:

Obtain the center point of the region of interest in the CT image, and obtain the farthest point of the region of interest from the center point as the first feature point;

Taking the center point as the origin, extending a straight line to both ends simultaneously until one end of the straight line coincides with the first characteristic point, and setting the obtained straight line as the first straight line axis;

Creating an elliptic curve, using the first linear axis as the major axis of the elliptic curve, and gradually increasing the minor axis of the elliptic curve until the range included by the elliptic curve encompasses all the areas of interest;

The image range included in the elliptic curve is the first CT local image;

Create a first rectangle, where the center point of the first rectangle coincides with the origin, the length of the first rectangle is equal to the length of the major axis and is parallel to the major axis, and the width of the first rectangle is equal to the length of the minor axis and is parallel to the minor axis;

Taking the center point of the first rectangle as the center, create a second rectangle that includes the first rectangle, and the size of the second rectangle is such that the area occupied by the first rectangle in the second rectangle is equal to a preset ratio;

The second rectangular area image in the CT image is input into a preset image segmentation model to obtain an area image where the diseased lymph node in the CT image is located.

Optionally, the preset image segmentation model is trained by the following steps:

Acquire multiple CT images of lymph nodes of different sizes in the body, and divide the boundaries of the lymph nodes in the CT images by an expert to obtain multiple CT training images of lymph nodes;

Creating a third rectangle for each lymph node CT training image, so that the area of the third rectangle is minimal and includes the image area of the entire lymph node;

Creating a fourth rectangle, wherein the center point of the fourth rectangle is the same as that of the third rectangle, and the size of the fourth rectangle is such that the area occupied by the third rectangle in the fourth rectangle is equal to a preset ratio;

Cutting out the image included in the fourth rectangle to obtain a plurality of lymph node CT preprocessing images;

The U-net++ neural network framework is trained using the multiple lymph node CT preprocessed images to obtain the preset image segmentation model.

In a second aspect, the present application provides a method for auxiliary diagnosis of lymphoma, the method for auxiliary diagnosis of lymphoma comprising the following steps:

After segmenting the CT image of the diseased lymph node using any of the diseased lymph node image segmentation methods described in the first aspect;

The image areas of the CT image and the PET image of the diseased lymph node belonging to the contour change pattern of the lymph node correspond to the image areas in the CT image, and are marked with different colors in the CT image to form an auxiliary judgment image;

The auxiliary judgment image is transmitted to the client to be displayed to the doctor.

Compared with the prior art, the technical solution provided by this application has the following advantages:

PET images can detect diseases by measuring the activity of glucose metabolism in the body, because cancer cells usually show higher glucose absorption rates than normal cells. However, in PET images, high metabolic organs such as the heart will also show high glucose absorption rates, making it difficult to distinguish which high absorption areas belong to lymphoma cells through PET images alone.

Therefore, the existing technology can only determine the location of lymph nodes through CT images first, and then pair them with PET images to identify the high glucose absorption areas in the PET images, and correspond these areas with the lymph nodes segmented in the CT images, so as to determine the image area of the diseased lymph nodes. However, since lymph nodes are distributed throughout the body and vary in size, and the entire CT image also contains other organs, it is a challenge to accurately distinguish and segment the location of lymph nodes directly from the CT image.

PET-CT scanning can obtain a series of tomographic image sequences by shooting along the direction of the body. In the direction distribution from the head to the feet, the heart and other high metabolic organs have a relatively fixed contour change pattern, and the center point changes less. In contrast, the distribution of lymphoma cells is uneven, and the regularity of their tissue contour changes is low. Therefore, it is sufficient to learn the contour change pattern of the heart and other high metabolic organs by analyzing a small number of samples, so that by excluding these patterns, it is possible to directly identify whether it belongs to the contour change pattern of the diseased lymph node from the contour change of the shadow area of the PET image. Therefore, the present application first accurately determines the shadow area belonging to the diseased lymph node tissue through the PET image sequence, and after matching with the CT image, the area where the entire lymph node exists is obtained by adaptively determining the shadow part of the diseased lymph node in the PET image, excluding the image area that does not belong to the lymph node, and reducing the possibility of misidentification. The present application can use the information of the PET image to provide assistance for the lymph node segmentation of the CT image, so that the accuracy of the lymph node image segmentation in the CT image can be improved.

At the same time, based on the above steps, this method adaptively segments the areas where diseased lymph nodes may exist, so that lymph nodes of different sizes are trained at the same scale, achieving scale normalization. At the same scale, the shape and texture features of lymph nodes of different sizes are more consistent in the model, making the feature extraction process more accurate, thereby better identifying and segmenting lymph nodes of different sizes.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a diseased lymph node image segmentation method provided in an embodiment of the present application.

FIG. 2 is a schematic diagram of dividing a local image area where a diseased lymph node is located, provided in an embodiment of the present application.

DETAILED DESCRIPTION

The technical solution in this application will be described below in conjunction with the accompanying drawings.

In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only part of the embodiments of the present application, not all of the embodiments. It should be noted that the embodiments of the present application and the features in the embodiments may be combined with each other without conflict.

PET images can detect diseases by measuring the activity of glucose metabolism in the body, because cancer cells usually show higher glucose absorption rates than normal cells. However, in PET images, high metabolic organs such as the heart will also show high glucose absorption rates, making it difficult to distinguish which high absorption areas belong to lymphoma cells through PET images alone.

Therefore, the existing technology can only determine the location of lymph nodes through CT images first, and then align them with PET images to identify the high glucose absorption areas in the PET images, and correspond these areas with the lymph nodes segmented in the CT images, so as to determine the image area of the diseased lymph nodes. However, since lymph nodes are distributed throughout the body and vary in size, and the entire CT image also contains other organs, it is a challenge to accurately distinguish and segment the location of lymph nodes directly from the CT image.

In a first aspect, the present application provides a diseased lymph node image segmentation method, as shown in FIG1 , the diseased lymph node image segmentation method comprises the following steps:

S101: Acquire a PET image sequence of a part to be diagnosed of a subject, wherein the PET image sequence includes at least two or more PET images of the part to be diagnosed taken consecutively;

Specifically, in the embodiment of the present application, the PET image sequence includes 5 PET images obtained by continuously moving and scanning along the body direction. This value can be adjusted according to the number of images obtained by scanning, for example, it can be increased to 10 or reduced to 3. However, it should be noted that at least two PET images are required to obtain the contour changes of different body tissues along the body direction.

S102: Acquire a grayscale image of each PET image in the PET image sequence.

S103: using an edge detection algorithm to obtain an image region of a shadow portion of each PET grayscale image, and obtaining a PET preprocessing image sequence;

Specifically, in the embodiment of the present application, the Canny edge detection algorithm is used to perform image segmentation on each shadow area in each PET grayscale image, thereby obtaining multiple shadow area images in each PET grayscale image;

Traverse each PET grayscale image and calculate the center point of each shadow area image, that is, calculate the average value of all pixel coordinate points in the shadow area.

When the difference between the center point of the shadow area image in the current PET grayscale image and the center point of the shadow area in the previous PET grayscale image is within a preset threshold, for example, the center point difference is within 10 pixels, it is determined that the two shadow area images have the same center point position.

The shadow area images with the same center point position in the PET grayscale image sequence are formed into an image set, and the image set is used as a separate PET preprocessing image sequence.

S104: Inputting the PET preprocessed image sequence into a contour change pattern recognition model to identify image regions belonging to the contour change pattern of lymph nodes.

Specifically, all PET preprocessed images in each PET preprocessed image sequence are input into the contour change pattern recognition model to obtain whether each PET preprocessed image sequence belongs to a lymph node.

Specifically, the contour change pattern recognition model is trained by the following steps:

Acquire multiple PET image sequences with diseased lymph nodes, and perform grayscale processing on each PET image to obtain a PET grayscale image sequence composed of PET grayscale images;

Using an edge detection algorithm to process the PET grayscale image in each PET grayscale image sequence to obtain a shadow area contained in each PET grayscale image;

Calculate the center point of the shadow area image in each PET grayscale image;

The shadow area images with the same center point position in the PET grayscale image sequence are formed into an image set, and the image set is used as a separate shadow area image sequence;

Each shadow area image sequence is manually labeled to determine whether it is a lymph node;

A convolution module is used to extract features of the shadow area image of each shadow area sequence to obtain a first feature matrix, wherein the convolution module includes at least one convolution layer and one pooling layer.

The central moment of the shadow area image in each shadow area sequence is calculated to obtain a second feature matrix.

The first feature matrix and the second feature matrix are combined to obtain a feature representation of each shadow area image in the shadow area sequence.

The feature representation of all shadow area images in the shadow area sequence is taken as input, the type of shadow area change sequence is used as the label, and a recurrent neural network is used as the neural network framework to train a contour change pattern recognition model for determining whether it belongs to a lymph node.

In an embodiment of the present application, the type of sequence of changes in the shadow area is set according to the normal organs that may appear in the high metabolic area to distinguish the diseased lymph nodes by excluding normal organs, aiming to reduce the amount of data required for learning. The actual approach in the embodiment of the present application is to achieve the method of distinguishing lymph nodes by excluding organs in the high metabolic area that do not belong to the diseased lymph nodes. For example, in the application of diseased lymph node identification in the chest area, we only need to distinguish whether the change in the shadow area matches the pattern of change of the heart contour. If it does not belong to the change pattern of the heart, it can be directly determined as the contour change pattern of the lymph node. Since the changes in the heart contour are regular, the artificial intelligence model can easily capture these features, so that only a small amount of data is needed to distinguish whether it is a high metabolic area contour change pattern of the diseased lymph node.

S105: registering the PET image with the simultaneously scanned CT image;

The image registration technology belongs to the existing technology and will not be described in detail here.

S106: the image region of the PET image belonging to the contour change pattern of the lymph node corresponds to the image portion in the CT image and is set as a region of interest;

S107: Segment the diseased lymph node image according to the region of interest, the CT image and a preset image segmentation model.

Referring to FIG. 2 , specifically, segmenting the CT image according to the region of interest and the preset image segmentation model to obtain the diseased lymph node image includes the following steps:

S201: For example, it is assumed that the shape of the region of interest, ie, the diseased lymph node, is a rectangle.

S202: Acquire the center point of the region of interest in the CT image.

S203: Obtain the farthest point of the region of interest away from the center point as the first feature point, take the center point as the origin, extend a straight line to both ends simultaneously, until one end of the straight line coincides with the first feature point, and set the obtained straight line as the first straight line axis. In the embodiment of the present application, it is any corner point.

S204: creating an elliptic curve, taking the first linear axis as the major axis of the elliptic curve, and gradually increasing the minor axis of the elliptic curve until the range included by the elliptic curve encompasses all the areas of interest;

The image range included in the elliptic curve is the first CT partial image.

S205: Create a first rectangle, wherein the center point of the first rectangle coincides with the origin, the length of the first rectangle is equal to the length of the major axis and is parallel to the major axis, and the width of the first rectangle is equal to the length of the minor axis and is parallel to the minor axis.

S206: Create a second rectangle including the first rectangle with the center point of the first rectangle as the center, and the size of the second rectangle is such that the area occupied by the first rectangle is equal to a preset ratio. In the embodiment of the present application, the preset ratio is generally 50%-90%;

The second rectangular area image in the CT image is input into a preset image segmentation model to obtain an area image where the diseased lymph node in the CT image is located.

Specifically, the preset image segmentation model is trained by the following steps:

Acquire multiple CT images of lymph nodes of different sizes in the body, and divide the boundaries of the lymph nodes in the CT images by an expert to obtain multiple CT training images of lymph nodes;

Creating a third rectangle for each lymph node CT training image, so that the area of the third rectangle is minimal and includes the image area of the entire lymph node;

Creating a fourth rectangle, wherein the center point of the fourth rectangle is the same as that of the third rectangle, and the size of the fourth rectangle is such that the area occupied by the third rectangle in the fourth rectangle is equal to a preset ratio;

Cutting out the image included in the fourth rectangle to obtain a plurality of lymph node CT preprocessing images;

The U-net++ neural network framework is trained using the multiple lymph node CT preprocessed images to obtain the preset image segmentation model.

In summary, the technical solution provided by the embodiments of the present application has the following advantages compared with the prior art:

PET-CT scanning can obtain a series of tomographic image sequences by shooting along the direction of the body. In the direction distribution from the head to the feet, the heart and other high metabolic organs have a relatively fixed contour change pattern, and the center point changes less. In contrast, the distribution of lymphoma cells is uneven, and the regularity of their tissue contour changes is low. Therefore, it is sufficient to learn the contour change pattern of the heart and other high metabolic organs by analyzing a small number of samples, so that by excluding these patterns, it is possible to directly identify whether it belongs to the contour change pattern of the diseased lymph node from the contour change of the shadow area of the PET image. Therefore, the present application first accurately determines the shadow area belonging to the diseased lymph node tissue through the PET image sequence, and after matching with the CT image, the area where the entire lymph node exists is obtained by adaptively determining the shadow part of the diseased lymph node in the PET image, excluding the image area that does not belong to the lymph node, and reducing the possibility of misidentification. The present application can use the information of the PET image to provide assistance for the lymph node segmentation of the CT image, so that the accuracy of the lymph node image segmentation in the CT image can be improved.

At the same time, based on the above steps, this method adaptively segments the areas where diseased lymph nodes may exist, so that lymph nodes of different sizes are trained at the same scale, achieving scale normalization. At the same scale, the shape and texture features of lymph nodes of different sizes are more consistent in the model, making the feature extraction process more accurate, thereby better identifying and segmenting lymph nodes of different sizes.

In a second aspect, the present application provides a method for auxiliary diagnosis of lymphoma, the method for auxiliary diagnosis of lymphoma comprising the following steps:

After segmenting the CT image of the diseased lymph node using any of the diseased lymph node image segmentation methods described in the above embodiments to obtain the diseased lymph node;

The image areas of the CT image and the PET image of the diseased lymph node belonging to the contour change pattern of the lymph node correspond to the image areas in the CT image, and are marked with different colors in the CT image to form an auxiliary judgment image;

The auxiliary judgment image is transmitted to the client to be displayed to the doctor.

It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. In addition, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. Moreover, in the description of the embodiments of the present application, unless otherwise specified, "/" means or, for example, A/B can represent A or B; "and/or" in this article is only a kind of association relationship describing the associated objects, indicating that there can be three relationships, for example, A and/or B can represent: A exists alone, A and B exist at the same time, and B exists alone. Furthermore, in the description of the embodiments of the present application, “plurality” refers to two or more than two.

The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims (6)

1. A diseased lymph node image segmentation method, characterized in that the diseased lymph node image segmentation method comprises the following steps: Acquire a PET image sequence of the part to be diagnosed of the subject, wherein the PET image sequence includes at least two or more PET images of the part to be diagnosed taken consecutively; Acquire a grayscale image of each PET image in the PET image sequence; Using an edge detection algorithm to obtain the image area of the shadow part of each PET grayscale image, and obtain a PET preprocessing image sequence; Inputting the PET preprocessed image sequence into a contour change pattern recognition model to identify image regions belonging to the contour change pattern of lymph nodes; The PET image is registered with the simultaneously scanned CT image; The image area of the PET image belonging to the contour change pattern of the lymph node corresponds to the image part in the CT image and is set as the region of interest; The diseased lymph node image is segmented and obtained according to the region of interest, the CT image and a preset image segmentation model. 2. The diseased lymph node image segmentation method according to claim 1, wherein the contour change pattern recognition model is trained by the following steps: Acquire multiple PET image sequences with diseased lymph nodes, and perform grayscale processing on each PET image to obtain a PET grayscale image sequence composed of PET grayscale images; Using an edge detection algorithm to process the PET grayscale image in each PET grayscale image sequence to obtain a shadow area contained in each PET grayscale image; Calculate the center point of the shadow area image in each PET grayscale image; The shadow area images with the same center point position in the PET grayscale image sequence are formed into an image set, and the image set is used as a separate shadow area image sequence; Each shadow area image sequence is manually labeled to determine whether it is a lymph node; Using a convolution module to extract features from the shadow area image of each shadow area sequence to obtain a first feature matrix, wherein the convolution module includes at least one convolution layer and one pooling layer; Calculate the central moment of the shadow area image in each shadow area sequence to obtain a second feature matrix; The first feature matrix and the second feature matrix are combined to obtain a feature representation of each shadow area image in the shadow area sequence; The feature representation of all shadow area images in the shadow area sequence is taken as input, the type of shadow area change sequence is used as the label, and a recurrent neural network is used as the neural network framework to train a contour change pattern recognition model for determining whether it belongs to a lymph node. 3. The diseased lymph node image segmentation method according to claim 2 is characterized in that the type of shadow area change sequence is set according to the normal organs that may appear in the high metabolic area so as to distinguish the diseased lymph nodes by excluding the normal organs. 4. The diseased lymph node image segmentation method according to claim 1, characterized in that segmenting the CT image according to the region of interest and a preset image segmentation model to obtain the diseased lymph node image comprises the following steps: Obtain the center point of the region of interest in the CT image, and obtain the farthest point of the region of interest from the center point as the first feature point; Taking the center point as the origin, extending a straight line to both ends simultaneously until one end of the straight line coincides with the first characteristic point, and setting the obtained straight line as the first straight line axis; Creating an elliptic curve, using the first linear axis as the major axis of the elliptic curve, and gradually increasing the minor axis of the elliptic curve until the range included by the elliptic curve encompasses all the areas of interest; The image range included in the elliptic curve is the first CT local image; Create a first rectangle, where the center point of the first rectangle coincides with the origin, the length of the first rectangle is equal to the length of the major axis and is parallel to the major axis, and the width of the first rectangle is equal to the length of the minor axis and is parallel to the minor axis; Taking the center point of the first rectangle as the center, create a second rectangle that includes the first rectangle, and the size of the second rectangle is such that the area occupied by the first rectangle in the second rectangle is equal to a preset ratio; The second rectangular area image in the CT image is input into a preset image segmentation model to obtain an area image where the diseased lymph node in the CT image is located. 5. The diseased lymph node image segmentation method according to claim 4, characterized in that the preset image segmentation model is trained by the following steps: Acquire multiple CT images of lymph nodes of different sizes in the body, and divide the boundaries of the lymph nodes in the CT images by an expert to obtain multiple CT training images of lymph nodes; Creating a third rectangle for each lymph node CT training image, so that the area of the third rectangle is minimal and includes the image area of the entire lymph node; Creating a fourth rectangle, wherein the center point of the fourth rectangle is the same as that of the third rectangle, and the size of the fourth rectangle is such that the area occupied by the third rectangle in the fourth rectangle is equal to a preset ratio; Cutting out the image included in the fourth rectangle to obtain a plurality of lymph node CT preprocessing images; The U-net++ neural network framework is trained using the multiple lymph node CT preprocessed images to obtain the preset image segmentation model. 6. A method for assisting diagnosis of lymphoma, characterized in that the method comprises the following steps: After segmenting the CT image of the diseased lymph node using the diseased lymph node image segmentation method described in any one of claims 1 to 5; The image areas of the CT image and the PET image of the diseased lymph node belonging to the contour change pattern of the lymph node correspond to the image areas in the CT image, and are marked with different colors in the CT image to form an auxiliary judgment image; The auxiliary judgment image is transmitted to the client to be displayed to the doctor.