CN111724391B

CN111724391B - Lymph cancer image fine segmentation method based on dynamic threshold

Info

Publication number: CN111724391B
Application number: CN202010425083.4A
Authority: CN
Inventors: 胡海根; 杜超; 苏一平; 管秋; 周乾伟; 肖杰; 陈胜勇
Original assignee: Zhejiang University of Technology ZJUT
Current assignee: Zhejiang University of Technology ZJUT
Priority date: 2020-05-19
Filing date: 2020-05-19
Publication date: 2023-04-07
Anticipated expiration: 2040-05-19
Also published as: CN111724391A

Abstract

A lymph cancer image fine segmentation method based on dynamic threshold, use the method of the double prediction precision to guarantee the accuracy that ROI chooses at first, and guarantee that the selective area includes the whole lymph cancer; secondly, regarding a single lymph cancer focus newly appeared in the optimization process as a new ROI area, and continuously optimizing the edge by using an independent threshold criterion so as to adjust the edge fine segmentation of the single lymph cancer focus in the dense area. According to the method, the initial segmentation result of the single focus image in the dense region is optimized by adopting a dynamic ROI threshold method, and a more accurate lymph cancer image is obtained by utilizing double-prediction precision information.

Description

Lymph cancer image fine segmentation method based on dynamic threshold

Technical Field

The invention relates to a lymph cancer image fine segmentation method based on a dynamic threshold value.

Technical Field

PET imaging of 18-fluorodeoxyglucose (18F-FDG) is a major means and tool for medical analysis of lymphoid cancers, and in PET images, standardized Uptake Value (SUV) is widely used to locate and segment lymphoid cancers. The PET image resolution of lymphoma is very low, and even an experienced expert, there are slight differences in manual identification of the same lymphoma, so in many cases, it is common to use 41% of the local maximum as a measure. However, using local thresholding as a segmentation criterion generally requires additional provision of regions of interest, and thresholding cannot be used in whole-body PET data. The diversity of lymphoma is clearly not satisfied by using the thresholding method directly. By being limited to a single maximum, surrounding images of lymphoid cancer masses are easily ignored, and some normal tissue images are mistaken for images of lymphoid cancer.

Disclosure of Invention

In order to overcome the defects of the prior art and solve the problem that the edge of a local focus image of some rough segmentation results is under-segmented, the invention provides a method for finely segmenting a lymph cancer image based on a dynamic threshold, and the initial segmentation result of a single focus image in a dense region is optimized by adopting a dynamic ROI threshold method.

In order to solve the above technical problems, the present invention can provide the following technical solutions:

a method for fine segmentation of a lymph cancer image based on a dynamic threshold, the method comprising the steps of:

step 1, obtaining an approximate position preparation stage, wherein probs refers to network output probability, T1 is a low threshold value, T2 is a high threshold value, and the process is as follows:

1.1 Obtaining a coordinate cluster C1 with lesion probability greater than T1 in probs;

1.2 C2 coordinates where the probability of foci in probs is greater than T2 were obtained.

1.3 For each C1, it is anded with C2, i.e., C1 ≈ C2, if not leaving C1 for the empty set, otherwise discarding.

Step 2, a local optimization stage, namely optimizing each cluster in the C1, wherein the cluster is a single focus set, and the process is as follows:

2.1 For a cluster of total pixel redundancy N1, if SUV minimum < SUV maximum 0.41: deleting the lowest value coordinate; otherwise, storing the cluster and performing the next cluster (the cluster is optimized);

2.2 For the total pixel points of the cluster less than the rest N1, storing the cluster and performing next cluster (the cluster is optimized);

2.3 Detect whether to classify into two clusters, delete the original cluster and add two clusters after splitting to C1 if split; not splitting: the next iteration continues to optimize the cluster;

2.4 Update the ROI;

2.5 ) repeat 2.1) -2.4) until C1 is

Further, in the step 1, the approximate position of the lymphoma is obtained, the low-precision of the lymphoma is indicated in a low threshold value T1 to ensure complete inclusion, and the high-precision of the lymphoma is indicated in a high threshold value T2 to ensure the accuracy of the focus, so that a good segmentation effect can be obtained in an undersegmented edge region to the maximum extent; excluding points with low SUV values until the lowest brightness value is greater than 41% of the highest brightness value; however, the remaining regions are monitored at regular intervals during the exclusion process, and if the region is split into two degree independent regions, there is no link between the two regions, and each region continues to optimize individually according to its own internal threshold, up to a 41% stopping criterion.

The technical conception of the invention is as follows: firstly, a double prediction precision method is used for ensuring the accuracy of ROI selection and ensuring that a selected region contains the whole lymph cancer; secondly, regarding a single lymph cancer focus newly appeared in the optimization process as a new ROI area, and continuously optimizing the edge by using an independent threshold criterion so as to adjust the edge fine segmentation of the single lymph cancer focus in the dense area.

The invention has the following beneficial effects: and acquiring a more accurate lymph cancer image by using the double prediction precision information.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

Referring to fig. 1, a method for performing a fine segmentation of a lymph cancer image based on a dynamic threshold value is a potential method for predicting a local lesion region on the basis of a rough lymph cancer region which can be obtained by a network prediction. Regions of high confidence are used as the primary criteria for candidate regions, while regions of low confidence are used primarily to include potential lesion regions as well. However, if there is only a low confidence prediction at one location, then this region will also be excluded. The focus position can be screened from the prediction result through the coincidence of high and low confidence degrees, and the omission is reduced.

Referring to fig. 1, the method for finely segmenting the lymph cancer image based on the dynamic threshold comprises the following steps:

2.1 For a cluster of total pixel redundancy N1, if SUV minimum < SUV maximum 0.41: deleting the lowest value coordinate; otherwise, storing the cluster and performing next cluster (the cluster is optimized);

2.3 Detecting whether the cluster is classified into two clusters, and if the cluster is split, deleting the original cluster and adding the two clusters after splitting into C1; not splitting: the next iteration continues to optimize the cluster;

2.4 Update the ROI;

2.5 ) repeat 2.1) -2.4) until C1 is

In the scheme of this embodiment, in step 1, a candidate region is determined by two confidence levels; in the step 2, the ROI is updated through continuous iteration, so that different local thresholds are found to be more suitable for complex case characteristics of the lymph cancer. Thus, the good segmentation effect can be obtained in the marginal area of the lymph cancer undersegmented to the maximum extent.

Claims

1. A lymph cancer image fine segmentation method based on dynamic threshold is characterized by comprising the following steps:

1.2 Obtaining a coordinate C2 with a lesion probability in probs greater than T2;

1.3 For each C1, make it and C2 phase, namely C1 ^ C2, if not for the empty set to reserve C1, otherwise abandon;

2.1 For clusters with more than N1 total pixels, if SUV minimum < SUV maximum 0.41: deleting the lowest value coordinate; otherwise, storing the cluster and carrying out the next cluster;

2.2 For the total pixel points of the cluster are less than N1, storing the cluster and carrying out the next cluster;

2.4 Update the ROI;

2.5 ) repeat 2.1) -2.4) until C1 is

In the step 1, the approximate position of the lymph cancer is obtained, the low precision of the lymph cancer is indicated in a low threshold value T1 to ensure the completeness of the lymph cancer, and the high precision is indicated in a high threshold value T2 to ensure the accuracy of the focus, so that a good segmentation effect can be obtained in an undersegmented edge region to the maximum extent; in the step 2, points with low SUV values are excluded until the SUV lowest value is greater than 41% of the SUV maximum value; however, the remaining clusters are monitored at regular intervals during the exclusion process, if the cluster is split into two independent clusters, the two clusters are not linked to each other, and each cluster is individually optimized according to its own internal stopping criteria until the SUV minimum in the cluster is greater than 41% of the SUV maximum.