KR102307995B1 - The diagnostic system of lymph node metastasis in thyroid cancer using deep learning and method thereof - Google Patents

Abstract

A method of operating a system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention includes a pre-processing step in which a multi-dimensional mask image is generated from a radioiodine plane image, and the residual tissue ( Remnant tissue) or lymph node metastasis (Lymph Node metastasis) is calculated, the pre-processed radioiodine plane image and the neural network (Neural Network) learning model is applied to the calculated likelihood of the learning is performed and diagnosing whether thyroid cancer has metastasized to a lymph node based on a result of neural network learning.

Description

Thyroid cancer lymph node metastasis diagnosis system using deep learning and its operation method

The present invention relates to a lymph node diagnosis system for thyroid cancer using deep learning and an operating method thereof, and more particularly, by generating a multi-dimensional mask image for a radioiodine plane image and calculating the likelihood of residual tissue or lymph node metastasis. A system for diagnosing lymph node metastasis of thyroid cancer using deep learning in which a neural network learning model is applied to a generated image and a calculated likelihood is used to diagnose lymph node metastasis, and a method of operating the same.

Lymph node metastases are one of the major prognostic factors of Differentiated Thyroid Carcinoma (DTC), and the prevalence of cervical lymph node metastasis in patients with differentiated thyroid cancer ranges from 20% to 90%. is reported to be

Radioactive iodine planar scans, one of the medical images, are helpful in step-by-step diagnosis by visualizing the degree of radioactive-iodine (RAI) metastatic lymph node metastasis hidden in thyroid cancer patients after thyroidectomy. However, the lack of anatomical features in the radioactive iodine plane image and the intake of radioiodine by the thyroid residue negatively affect the diagnostic sensitivity and specificity, resulting in a number of false positive or false negative diagnoses. There is a risk of over- or under-treatment. To overcome this problem, an accurate diagnosis of lymph node metastasis can be made by using single photon emission computed tomography (SPECT) images, but radiation exposure problems and cost problems caused by tomography and time problem, it is difficult to continuously use the tomography method.

To solve the above problems, a computer aided diagnosis (CAD) system may be used. The computer-assisted diagnosis system may refer to a system capable of assisting a doctor in diagnosis by displaying diagnosis-related contents on a predetermined portion of a medical image according to a result of analyzing a medical image. A computer-aided diagnostic method for thyroid cancer metastasis to lymph nodes can be a useful tool not only for doctors but also for patients.

However, subdivision and examination of radioiodine plane images for use in computer-aided diagnosis is difficult due to the possibility of residual thyroid tissue with radioiodine binding properties. Currently, we are in a limited situation where we have to focus on CT images for the detection of cervical lymph node metastases, and even if CT images are used, there are cases of metastases that cannot be identified from adjacent tissues, which is not suitable.

Recently, a deep-learning algorithm is used to solve various medical diagnostic problems, and complex imaging and anatomical images are analyzed and utilized for diagnosis. However, in the radioiodine plane image, lymph node metastasis and the remaining tissue region are generally affected by noise artifacts and have a problem that may overlap with other regions. This is inappropriate, and the development of technology for diagnosing lymph node metastasis through pixel-wise regional-based neural network training is required.

Republic of Korea Patent Publication No. 10-2018-0022440 (published date: 2018.03.06)

The present invention is a countermeasure to the above-described problem, but analyzes a radioiodine plane image using a deep learning algorithm, rather than directly learning the radioiodine plane image itself, a pixel-unit preprocessed mask image and An object of the present invention is to provide a lymph node metastasis diagnosis system that can utilize the result of learning with respect to the calculated likelihood and an operating method thereof.

As an embodiment of the present invention, a method of operating a system for diagnosing lymph node metastasis of thyroid cancer using deep learning may be provided.

A method of operating a system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention includes a pre-processing step in which a multi-dimensional mask image is generated from a radioiodine plane image, and the residual tissue ( Remnant tissue) or lymph node metastasis (Lymph Node metastasis) is calculated, the pre-processed radioiodine plane image and the neural network (Neural Network) learning model is applied to the calculated likelihood of the learning is performed and diagnosing whether thyroid cancer has metastasized to a lymph node based on a result of neural network learning.

The pre-processing step of generating a multi-dimensional mask image of the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention is the step of detecting the region of the tracer ingestion site from the radioiodine plane image, the detected tracer intake The method may include performing Gaussian filtering on the partial region and generating a multidimensional mask image in which pixel features are expressed in a plurality of levels based on the Gaussian filtering result.

In the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, in the step of diagnosing whether lymph node metastasis of thyroid cancer is diagnosed, it is determined as any one of residual tissue and lymph node metastasis for each region ingested by the tracer. Depending on the results obtained, it can be diagnosed whether the thyroid cancer has metastasized to the lymph nodes.

In the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, after filtering is performed in units of pixels to remove an area erroneously determined in the stage of diagnosing whether lymph node metastasis of thyroid cancer is performed, Further processing steps may be included.

In the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, the likelihood in the step of calculating the likelihood is the GT of all pixels of the preprocessed radioiodine plane image. (Ground Truth) may be calculated by performing normalization on the value.

As an embodiment of the present invention, a system for diagnosing lymph node metastasis of thyroid cancer using deep learning may be provided.

A system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention includes a preprocessing unit that generates a multidimensional mask image from a radioiodine plane image, and a residual tissue for each pixel in the preprocessed radioiodine planar image. Alternatively, a neural network in which learning is performed by applying a Neural Network learning model to the likelihood determining unit where the likelihood of lymph node metastasis is calculated, the preprocessed radioiodine plane image, and the calculated likelihood. A diagnosis predictor for diagnosing whether thyroid cancer has metastasized to a lymph node based on the learning results of the learning unit and the neural network learning unit may be included.

In the preprocessor of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, a tracer intake site region is detected from a radioiodine plane image, and Gaussian filtering is performed on the detected tracer intake site region, A multidimensional mask image in which pixel features are expressed in a plurality of levels may be generated based on the Gaussian filtering result.

In the diagnostic prediction unit of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, the diagnosis of whether the lymph node metastasis of thyroid cancer is diagnosed according to the result determined as either residual tissue or lymph node metastasis for each region of the tracer intake site can be

The system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention may further include a post-processing unit in which filtering in units of pixels is performed to remove a region erroneously determined by the diagnosis prediction unit.

In the likelihood determination unit of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, normalization is performed on the GT (Ground Truth) values of all pixels of the pre-processed radioiodine plane image, thereby increasing the likelihood. can be calculated.

Meanwhile, as an embodiment of the present invention, a computer-readable recording medium in which a program for implementing the above-described method is recorded may be provided.

According to an embodiment of the present invention, learning may be performed by minimizing the influence of noise artifacts by learning the pixel-by-pixel preprocessed image of the radioiodine image.

In addition, by adding information on the likelihood of lymph node metastasis and residual tissue as a learning target, it is possible to more accurately determine whether lymph node metastasis of thyroid cancer is diagnosed.

1 is a flowchart illustrating an operating method of a system for diagnosing lymph node metastasis according to an embodiment of the present invention.
2 is a flowchart illustrating a pre-processing method for generating a multi-dimensional mask image in the operating method of the lymph node metastasis diagnosis system according to an embodiment of the present invention.
3 is an exemplary diagram illustrating an image on which Gaussian filtering is performed in the operating method of the system for diagnosing lymph node metastasis according to an embodiment of the present invention.
4 is an exemplary diagram illustrating an image in which the calculated likelihood of lymph node metastasis and residual tissue is reflected in the operating method of the lymph node metastasis diagnosis system according to an embodiment of the present invention.
5 is an exemplary diagram showing the results of displaying lymph node metastasis and residual tissue for each patient in the operating method of the lymph node metastasis diagnosis system according to an embodiment of the present invention.
6 is an exemplary diagram illustrating incorrect classification results in the operating method of the system for diagnosing lymph node metastasis according to an embodiment of the present invention.
7 is an exemplary diagram illustrating a result of performing a method according to the number of hidden layers of a neural network in an operating method of a system for diagnosing lymph node metastasis according to an embodiment of the present invention.
8 is a block diagram illustrating a system for diagnosing lymph node metastasis according to an embodiment of the present invention.
9 is an exemplary diagram illustrating a system for diagnosing lymph node metastasis according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement them. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Terms used in this specification will be briefly described, and the present invention will be described in detail.

The terms used in the present invention have been selected as currently widely used general terms as possible while considering the functions in the present invention, but these may vary depending on the intention or precedent of a person skilled in the art, the emergence of new technology, and the like. In addition, in a specific case, there is a term arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the corresponding invention. Therefore, the term used in the present invention should be defined based on the meaning of the term and the overall content of the present invention, rather than the name of a simple term.

When a part "includes" a certain element throughout the specification, this means that other elements may be further included, rather than excluding other elements, unless otherwise stated. In addition, terms such as "...unit" and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software, or a combination of hardware and software. . In addition, when a part is "connected" to another part throughout the specification, it includes not only a case in which it is "directly connected" but also a case in which it is connected "with another element in the middle".

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

As an embodiment of the present invention, a method of operating a system for diagnosing lymph node metastasis of thyroid cancer using deep learning may be provided. In addition, the operating method of the lymph node metastasis diagnosis system may be performed by the lymph node metastasis diagnosis system 1000 to be described later. explain about it.

1 is a flowchart illustrating an operating method of a system for diagnosing lymph node metastasis according to an embodiment of the present invention.

Referring to FIG. 1 , the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention includes a pre-processing step (S100) of generating a multi-dimensional mask image from a radioactive iodine plane image, the pre-processed radioactive iodine A step of calculating the likelihood of residual tissue or Lymph Node metastasis for each pixel in the planar image (S200), the preprocessed radioiodine planar image and the calculated likelihood of the neural network Network) A step of performing learning by applying a learning model ( S300 ) and a step of diagnosing whether lymph node metastasis of thyroid cancer is diagnosed based on the result of neural network learning ( S400 ) may be included.

The radioactive iodine planar image (whole-body planar scans) is one of the medical images as described above, and refers to an image that visualizes the degree of radioactive-iodine (RAI) metastatic lymph node metastasis hidden in a thyroid cancer patient after thyroidectomy. can do. In addition, the radioiodine plane image may correspond to an image taken using a gamma-ray camera after a predetermined time has elapsed after administration of 131-I iodine to a thyroid cancer patient.

First, in the pre-processing step ( S100 ), pre-processing is performed on the input radioiodine plane image, so that multi-dimensional masked images can be generated.

2 is a flowchart illustrating a pre-processing method for generating a multi-dimensional mask image in an operating method of a lymph node metastasis diagnosis system according to an embodiment of the present invention, and FIG. 3 is a diagram illustrating a lymph node metastasis diagnosis system according to an embodiment of the present invention. In the method of operation, it is an exemplary diagram showing an image on which Gaussian filtering is performed.

2 to 3, the pre-processing step (S100) of generating a multidimensional mask image of the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention is a tracer from a radioiodine plane image. A step of detecting the intake site region (S110), the step of performing Gaussian filtering on the detected tracer intake region region (S120), and a multi-dimensional mask image in which pixel features are expressed in a plurality of levels based on the Gaussian filtering result is generated Step S130 may be included.

Specifically, in the step (S110) of detecting the region of the tracer intake site from the radioiodine plane image, the region of the tracer intake site for diagnosing whether or not metastasis of thyroid cancer is detected in the radioiodine plane image is first detected, and the tracer intake site is the residual tissue. and metastatic tissue, the tracer may refer to a substance emitting positrons in a radiographic image generating device. That is, in the step of detecting the region of the tracer ingestion site (S110), the classification result for the metastatic tissue or the residual tissue has not yet been obtained, but the region in which the tracer is ingested around the thyroid gland, which is the subject of diagnosis, is separately detected according to the present invention, thereby further deep learning learning. can be used as a target for

Next, in the step S120 of performing Gaussian filtering on the detected tracer intake region, Gaussian filtering may be performed to improve image quality and generate multidimensional pixel features. Gaussian filtering, one of the image processing techniques, may refer to a filtering method that generates a mask using a Gaussian function and performs a mask operation with an input image. can As shown in FIG. 3 , by setting standard deviation values of 2, 4, and 8 with respect to the original image A, the filtered images may appear as B, C, and D, respectively.

Next, that is, a multidimensional mask image may be generated through a mask operation using the image generated through Gaussian filtering in step S120 with respect to the tracer intake site detected in step S110. That is, as a result of the pre-processing step ( S100 ), a multidimensional mask image in which pixel features are expressed in a plurality of levels based on the Gaussian filtering result may be generated ( S130 ).

After the pre-processing step (S100), a step (S200) of calculating the likelihood of residual tissue or lymph node metastasis for each pixel in the pre-treated radioiodine plane image may be performed.

4 is an exemplary diagram illustrating an image in which the calculated likelihood of lymph node metastasis and residual tissue is reflected in the operating method of the lymph node metastasis diagnosis system according to an embodiment of the present invention.

4 , in the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, the likelihood in the step of calculating the likelihood is the preprocessed radioiodine plane. It may be calculated by performing normalization on GT (Ground Truth) values of all pixels of the image.

The GT (Ground Truth) value is used in the image processing algorithm performance evaluation, and may refer to an actual value serving as a comparison standard when evaluating how accurate a specific result value is. A more detailed description is omitted as it is irrelevant to the content of the present invention.

Specific likelihood calculation may be performed by [Equation 1] and [Equation 2] for the following normalization calculation.

[Equation 1]

[Equation 2]

는 전술한 GT(Ground Truth) 값에 해당되고, [수학식 1]에서

는 해당 전처리 영상의 픽셀 위치가 잔여 조직(remnant tissue)에 해당될 확률을 나타내며, [수학식 2]에서

는 해당 전처리 영상의 픽셀 위치가 림프절 전이(lymph node metastasis)에 해당될 확률을 나타낼 수 있다. 즉, 위 [수학식 1] 및 [수학식 2]를 사용하여 전처리된 영상에 대한 모든 픽셀의 GT 값을 정규화함으로써 림프절 전이 혹은 잔여 조직 여부에 대한 가능도가 계산될 수 있다.In the above [Equation 1] and [Equation 2]

corresponds to the above-described GT (Ground Truth) value, and in [Equation 1]

represents the probability that the pixel position of the pre-processed image corresponds to residual tissue, and in [Equation 2]

may represent the probability that the pixel position of the pre-processed image corresponds to lymph node metastasis. That is, the likelihood of lymph node metastasis or residual tissue can be calculated by normalizing the GT values of all pixels for the preprocessed image using the above [Equation 1] and [Equation 2].

Next, a step (S300) of learning by applying a neural network learning model to the preprocessed radioiodine plane image and the calculated likelihood may be performed. Referring to FIG. 8 , the preprocessed radioiodine plane image and information about the likelihood may be stored in the database unit 600 for learning the neural network.

A variety of algorithms may be used in the deep learning algorithm used in the present invention. Hereinafter, a learning and diagnosis process using a deep neural network (DNN) will be described. The deep neural network (DNN) may refer to an artificial neural network (ANN) including several hidden layers between an input layer and an output layer.

7 is an exemplary diagram showing the performance result according to the number of hidden layers of the neural network in the operating method of the system for diagnosing lymph node metastasis according to an embodiment of the present invention. As shown in FIGS. 7 and 9, the number of hidden layers can be variously set, such as 3(A), 7(B), 10(C), but it can be seen that the higher the number of hidden layers, the better the performance (ex. accuracy, loss, etc.). That is, in case of C rather than A and B in FIG. 7 , it can be seen that the performance is superior.

In addition, the number of nodes in the input layer may be determined to be the same as the number of image pixel features extracted in the above-described steps S100-S200.

In addition, the deep neural network may include a Rectified Linear Unit (ReLU) layer, a tanh(r) layer, etc. for performing an activation function.

In the learning of the deep neural network, as the labeling information, a result diagnosed by a single photon emission computed tomography (SPECT) image may be utilized. That is, labeling can be performed for deep neural network learning using the results of diagnosis of lymph node metastasis or residual tissue based on the tomography results for each patient.

After the learning is performed (S300) through the deep neural network learning model as described above, based on the learning result, whether the lymph node metastasis of the thyroid cancer is diagnosed (S400). In the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, in the step of diagnosing whether lymph node metastasis of thyroid cancer is diagnosed, it is determined as any one of residual tissue and lymph node metastasis for each region ingested by the tracer. Depending on the results obtained, it can be diagnosed whether the thyroid cancer has metastasized to the lymph nodes.

In order to perform the lymph node metastasis diagnosis step, a softmax layer may be used as a classification layer at the output end of the deep neural network. That is, in the softmax layer, the classification of lymph node metastasis or residual tissue can be performed in units of pixels according to the score output according to the deep neural network learning result.

5 is an exemplary view showing the results of displaying lymph node metastasis and residual tissue for each patient in the operating method of the lymph node metastasis diagnosis system according to an embodiment of the present invention, and FIG. 6 is a lymph node according to an embodiment of the present invention. In the operating method of the metastasis diagnosis system, it is an exemplary diagram showing incorrect classification results.

Referring to FIG. 5 , according to the operating method of the lymph node metastasis diagnosis system of the present invention, the green part corresponds to the marked area classified as the residual tissue 20 in the radioiodine plane image, and the red part corresponds to the lymph node metastasis tissue 30 . The sorted and displayed results are shown. That is, in the radioiodine plane image 50 input according to the implementation of the present invention, the result of classification into residual tissue and lymph node metastasis tissue in pixel-wise manner may be displayed.

In addition, in the operating method of the system for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, filtering in units of pixels to remove an erroneously determined region in the stage of diagnosing whether lymph node metastasis of thyroid cancer is performed is performed. A post-processing step may be further included.

Various methods may be used for the post-processing method. For example, a size filtering method for a predetermined pixel may be included in an image classified as metastatic tissue or lymph node metastasis tissue.

As an embodiment of the present invention, a system 1000 for diagnosing lymph node metastasis of thyroid cancer using deep learning may be provided. In relation to the system 1000 for diagnosing lymph node metastasis to be described below, the above-described method may be applied. Accordingly, descriptions of the same contents as those described above in relation to the system are omitted.

8 is a block diagram illustrating a lymph node metastasis diagnosis system 1000 according to an embodiment of the present invention, and FIG. 9 is an exemplary diagram illustrating a lymph node metastasis diagnosis system 1000 according to an embodiment of the present invention.

8 and 9 , the system 1000 for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention includes a preprocessing unit 100 that generates a multidimensional mask image from a radioiodine plane image; The likelihood determination unit 200, which calculates the likelihood of residual tissue or Lymph Node metastasis for each pixel in the radioiodine plane image, the preprocessed radioiodine plane image and the calculated likelihood With respect to the figure, a neural network learning unit 300 to which a Neural Network learning model is applied and learning is performed, and a diagnosis prediction unit 400 for diagnosing whether lymph node metastasis of thyroid cancer is diagnosed based on the learning results of the neural network learning unit 300 ) may be included.

In the preprocessor 100 of the system 1000 for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, a tracer intake site region is detected from a radioiodine plane image, and the detected tracer intake site region is Gaussian filtering is performed, and a multidimensional mask image in which pixel features are expressed in a plurality of levels may be generated based on the Gaussian filtering result.

In the diagnosis prediction unit 400 of the system 1000 for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, the thyroid cancer lymph node metastasis can be diagnosed.

In the system 1000 for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, a post-processing unit 500 in which filtering is performed in units of pixels to remove an area erroneously determined by the diagnosis prediction unit 400 . may be further included.

In the likelihood determination unit 200 of the system 1000 for diagnosing lymph node metastasis of thyroid cancer using deep learning according to an embodiment of the present invention, the GT (Ground Truth) values of all pixels of the preprocessed radioiodine plane image are normalized. By performing , the likelihood can be calculated.

Meanwhile, the above-described method can be written as a program that can be executed on a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable medium. In addition, the structure of the data used in the above-described method may be recorded in a computer-readable medium through various means. A recording medium for recording an executable computer program or code for performing various methods of the present invention should not be construed as including temporary objects such as carrier waves or signals. The computer-readable medium may include a storage medium such as a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.) and an optically readable medium (eg, a CD-ROM, a DVD, etc.).

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention. do.

100: preprocessing unit 200: likelihood determining unit
300: neural network learning unit 400: diagnosis prediction unit
500: post-processing unit 600: database unit
1000: lymph node metastasis diagnosis system

Claims (11)

In the operating method of a system for diagnosing lymph node metastasis of thyroid cancer using deep learning,
A pre-processing step of generating a multi-dimensional mask image from the radioiodine plane image in the pre-processing unit;
calculating the likelihood of residual tissue or Lymph Node metastasis for each pixel in the pre-processed radioiodine plane image in the likelihood determination unit;
performing learning by applying a neural network learning model to the preprocessed radioiodine plane image and the calculated likelihood in the neural network learning unit; and
A step of diagnosing whether the lymph node metastasis of thyroid cancer is included based on the result of the neural network learning in the diagnosis prediction unit,
The pre-processing step in which the multidimensional mask image is generated is,
detecting a region of a tracer intake site from the radioiodine plane image;
performing Gaussian filtering on the detected tracer intake region; and
The method of operating a system for diagnosing lymph node metastasis of thyroid cancer using deep learning, further comprising generating a multidimensional mask image in which pixel features are expressed in a plurality of levels based on the Gaussian filtering result.
delete The method of claim 1,
In the step of diagnosing whether or not the lymph node metastasis of the thyroid cancer is present, the lymph node metastasis diagnosis of thyroid cancer using deep learning is diagnosed according to the result determined as either residual tissue or lymph node metastasis for each region of the tracer intake site. How the system works.
4. The method of claim 3,
In the post-processing unit, a post-processing step in which filtering in a pixel unit is performed to remove an erroneously determined region in the step of diagnosing whether the lymph node metastasis of the thyroid cancer is diagnosed. how it works.
The method of claim 1,
Lymph node metastasis of thyroid cancer using deep learning is calculated by normalizing GT (Ground Truth) values of all pixels of the preprocessed radioiodine plane image in the step of calculating the likelihood. How the diagnostic system works.
In a system for diagnosing lymph node metastasis of thyroid cancer using deep learning,
a preprocessor for generating a multidimensional mask image from the radioiodine plane image;
a likelihood determining unit for calculating the likelihood of residual tissue or lymph node metastasis for each pixel in the pre-processed radioiodine plane image;
a neural network learning unit that performs learning by applying a neural network learning model to the preprocessed radioiodine plane image and the calculated likelihood; and
a diagnosis prediction unit for diagnosing whether or not metastasis of thyroid cancer to lymph nodes is diagnosed based on the learning result of the neural network learning unit;
In the preprocessor, a region of a tracer intake site is detected from the radioiodine plane image, Gaussian filtering is performed on the detected tracer intake region, and a multidimensional mask in which pixel features are expressed in a plurality of levels based on the result of the Gaussian filtering. A system for diagnosing lymph node metastasis of thyroid cancer, characterized in that an image is generated.
delete 7. The method of claim 6,
The thyroid cancer lymph node metastasis diagnosis system in which the diagnosis predicting unit diagnoses whether the thyroid cancer has metastasized to the lymph node according to a result determined as any one of residual tissue and lymph node metastasis for each region of the tracer intake site.
9. The method of claim 8,
The system for diagnosing lymph node metastasis of thyroid cancer further includes a post-processing unit that performs pixel-by-pixel filtering to remove the region erroneously determined by the diagnosis prediction unit.
7. The method of claim 6,
In the likelihood determination unit, the likelihood is calculated by normalizing the GT (ground truth) values of all pixels of the pre-processed radioiodine plane image.
A computer-readable recording medium in which a program for implementing the method of any one of claims 1 and 3 to 5 is recorded.

Images