KR102208358B1 - Lung Cancer Diagnosis System Using Artificial Intelligence, Server and Method for Diagnosing Lung Cancer - Google Patents

Abstract

The present invention relates to a system for predicting the probability that a lung nodule is cancer using CT images. Through the embodiment, individuals and clinics can quickly obtain a reading result of a lung cancer screening CT image and improve the accuracy of the CT image reading result. In addition, through online, users can view their CT image reading results at any time, and manage their own medical image information by comparing them with previous results from time to time during follow-up tests.

Description

Lung Cancer Diagnosis System Using Artificial Intelligence, Server and Method for Diagnosing Lung Cancer}

The present invention relates to a system for predicting the probability that a lung nodule is cancer using CT images. Specifically, it relates to a lung cancer screening CT reading server and a CT reading method using artificial intelligence.

Unless otherwise indicated herein, the content described in this section is not prior art to the claims of this application, and inclusion in this section is not admitted to be prior art.

Computed tomography (CT), unlike X-ray imaging, acquires a cross-sectional image of a human body horizontally cut. CT has the advantage of being able to see structures and lesions more clearly because structures overlap less than simple X-rays. In addition, CT scanning is a basic examination method when a lesion is suspected in most organs and diseases and requires detailed examination because it is cheaper than MRI and the examination time is short.

Recently, with the spread of multi-channel CT (MDCT, multi-detector CT), it is possible to freely obtain desired cross-sectional and three-dimensional (3D) images like MRI by reconstructing images after shooting, and CT images can be collected at medical institutions. .

However, the number of specialists who read CT images is limited, and there is no system for requesting medical image readings. For this reason, even if an individual owns CT images, it is not possible to request reliable reading online. In addition, there is no system capable of improving the reliability of image reading results, which is very important for deriving accurate determination results, such as cancer determination of nodules.

1. Korean Patent Application No. 10-2018-0070722 (2018.06.20) 2. Korean Patent Application No. 10-2016-0131976 (2016.10.12)

It provides a CT reading server, a lung cancer diagnosis method, and a lung cancer diagnosis system that predicts the probability that a lung nodule is cancer using CT images.

The lung cancer screening CT reading server using artificial intelligence according to the embodiment includes a pre-processed pre-nodule image and a follow up-nodule image or a three-dimensional pre-nodule cube and a follow-up nodule cube. Masking each continuous CT (Computed Tomography) image as a primary layer divided into a plurality of unit images, extracting features of each unit image included in the primary layer, and calculating the number of features for each extracted unit image. And a calculation module for calculating a doubling time (DT) of a nodule included in the pre-nodule image and the follow-up-nodule image and a percentage change in pulmonary nodule volume (PVC); An extraction module that collects feature information extracted from each of the continuous CT images, reflects the learned weight in the collected feature information, and filters feature information whose weight is less than a certain level to extract analysis valid information; A generation module for generating analysis data by adding the doubling time and pulmonary nodule volume change calculated as a result of continuous CT image analysis to the extracted valid information, and extracting final analysis valid information from the analysis data to generate final analysis data; And a calculation module for calculating a cancer probability of a nodule by comparing the generated final analysis data with the learned cancer diagnosis data set. Includes.

An online lung cancer diagnosis system using artificial intelligence according to another embodiment includes a user terminal for collecting personal health information including X-rays, CT images of lung nodules, and health checkup information; A medical institution server for collecting CT images of pulmonary nodules including a patient's pre-nodule image and follow up-nodule image; And a lung cancer screening CT reading server that receives the CT image of a lung nodule from a user terminal or a medical institution server, calculates a probability that the lung nodule is cancer, and transmits the calculated result to the medical institution server and the user terminal. Includes.

The lung cancer diagnosis process of the CT reading server for lung cancer screening using artificial intelligence according to another embodiment is (A) the calculation module includes a pre-processed pre-nodule image and a follow up-nodule image or 3 Each consecutive CT image including the dimensional pre-nodule cube and the follow-up nodule cube is masked with a primary layer divided into a plurality of unit images, and special extraction of each unit image included in the primary layer, and the extracted unit image Calculating the number of star features; (B) the calculation module includes calculating a nodule doubling time (DT, Doubling Time) and a pulmonary nodule volume percentage change (PVC) included in the pre-nodule image and the follow-up-nodule image; (C) the extraction module collects feature information extracted from each of the continuous CT images, reflects the learned weight in the collected feature information, and filters feature information whose weight is less than a certain level to extract analysis valid information; (D) The generation module generates analysis data by adding the doubling time and pulmonary nodule volume change calculated as a result of continuous CT image analysis to the extracted valid information, and extracts the final analysis valid information from the analysis data to generate final analysis data. Step to do; And (E) calculating the cancer probability of the nodule by comparing the generated final analysis data with the learned cancer diagnosis data set. Includes.

It is possible to quickly obtain the reading result of the CT image for lung cancer screening conducted in individuals and clinics, and improve the accuracy of the CT image reading result. In addition, through online, the user can request to read his or her CT image, view the result at any time, and manage his/her medical image information by comparing it with previous results at any time during follow-up examination.

The CT image reading server can analyze image data using artificial intelligence, improve data processing efficiency, and improve image reading accuracy.

Through the introduction of a CT image reading system, it is possible to create profits for individuals and medical institutions by allowing individual readers, clinics, and hospitals to perform medical image reading online.

The effects of the present invention are not limited to the above effects, and should be understood to include all effects that can be inferred from the configuration of the invention described in the detailed description or claims of the present invention.

1 is a diagram showing an online lung cancer diagnosis system using artificial intelligence according to an embodiment
Figure 2 is a diagram for explaining the lung cancer probability calculation process of the CT reading server 100 according to the embodiment
3 is a block diagram of a CT reading server 100 according to an embodiment
4 is a diagram showing a lung cancer probability diagnosis process according to an embodiment
5 is a diagram showing a detailed data processing block of a calculation module according to an embodiment
6 is a diagram showing an image feature extraction process according to an embodiment
7 is a diagram showing a data processing flow of a lung cancer diagnosis method of a lung cancer screening CT reading server using artificial intelligence according to an embodiment
8 is a view showing a feature extraction process of a continuous image in step S710
9 is a diagram showing an example of a continuous CT image (prenodule image, follow-up nodule image)

Advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in a variety of different forms, only the present embodiments are intended to complete the disclosure of the present invention, and the general knowledge in the technical field to which the present invention pertains. It is provided to completely inform the scope of the invention to those who have it, and the invention is only defined by the scope of the claims. The same reference numerals refer to the same elements throughout the specification.

In describing embodiments of the present invention, if it is determined that a detailed description of a known function or configuration may unnecessarily obscure the subject matter of the present invention, a detailed description thereof will be omitted. In addition, terms to be described later are terms defined in consideration of functions in an embodiment of the present invention, which may vary according to the intention or custom of users or operators. Therefore, the definition should be made based on the contents throughout this specification.

1 is a diagram showing an online lung cancer diagnosis system using artificial intelligence according to an embodiment.

Referring to FIG. 1, the lung cancer diagnosis system according to the embodiment may include a CT reading server 100, a user terminal 200, and a medical institution server 600.

The user terminal 200 collects patient's personal health information, such as X-rays, CT images, and CT images of lung nodules of the patient who performed CT scans, and transmits individual health information to a server that requests health information for reading the CT image. have.

The medical institution server 600 collects continuous pulmonary nodule CT images including pre-nodule images and follow up-nodule images of patients, and collects and updates health information of patients. I can.

The lung cancer diagnosis CT reading server 100 receives a CT image of a lung nodule from the user terminal 200 or the medical institution server 600, calculates the probability that the lung nodule is cancer, and transmits the calculated result to the medical institution server 600 and the user terminal 200. Transfer to.

In the embodiment, the lung cancer diagnosis system may further include a payment system server 500, a database server 300, a medical personnel terminal 400, and a security system server 700. The payment system server 500 pays the reading cost when the CT image reading is performed in the lung cancer diagnosis system. The database server 300 may store reference data, a data set, and medical information of patients necessary for calculating a lung cancer probability. The reference data may include lung nodule image data extracted from a continuous CT image.

The medical practitioner terminal 400 is a user terminal of a doctor subscribed to the lung cancer diagnosis system, and may check a CT image reading result, directly perform a CT image reading, or perform abnormal nodule diagnosis and findings. The security system server 700 secures data and personal information of each server included in the lung cancer diagnosis system.

2 is a view for explaining a lung cancer probability calculation process of the CT reading server 100 according to an embodiment.

Referring to FIG. 2, network A of the reading server 1000 receives a pre-nodule image and a follow up-nodule image on which an image preprocessing process has been performed. Thereafter, features are extracted from two consecutive CT images, and the doubling time of the nodule and the rate of change in the volume of the nodule are calculated from the two consecutive images. Feature information of each image extracted from network A is transmitted to network B of the reading server. In Network B, the extracted features are integrated, and analysis valid information and analysis data are selected using a 3D convolutional neural network. Afterwards, the pulmonary nodule volume and volume change rate information are added to the analysis data. After that, the final analysis data is extracted, and the final analysis data is compared with the training data set to output the probability of lung nodule cancer.

3 is a block diagram of a CT reading server 100 according to an embodiment.

Referring to Figure 3, the CT reading server 100 according to the embodiment may be composed of a network A and a network B. Network A extracts features of continuous CT images and calculates the doubling time and pulmonary nodule volume change rate using the analysis results of the continuous CT images. Network B collects features extracted from network A, generates final analysis data, and calculates the cancer probability of the nodule. To this end, network A may be configured to include a calculation module 110, and network B may be configured to include an extraction module 130, a generation module 150, and a calculation module 170. The term'module' used in this specification should be interpreted as being capable of including software, hardware, or a combination thereof, depending on the context in which the term is used. For example, the software may be machine language, firmware, embedded code, and application software. As another example, the hardware may be a circuit, a processor, a computer, an integrated circuit, an integrated circuit core, a sensor, a MEMS (Micro-Electro-Mechanical System), a passive device, or a combination thereof.

Hereinafter, in order to help understand the function of the data processing block of the server, a description will be given with reference to FIG. 4 showing a lung cancer probability diagnosis process according to an embodiment.

The calculation module 110 receives the pre-processed pre-nodule image and follow-up-nodule image, masks each image with a first layer divided into a plurality of unit images, Features of each unit image included in the first layer are extracted. In an embodiment, the unit image feature may include information indicating detailed information of an object included in the image, such as an edge, color, and size of an object included in the image. More specifically, the unit image feature extracted from the first layer may be an edge of each segmented image area, and the unit image feature extracted from the second layer may be a line or a plane, Features extracted from the third layer may be an overall shape or a contour. That is, the more complex the two-dimensional and three-dimensional image features are extracted from the simple one-dimensional feature as the first layer proceeds to the higher-order layer. In addition, the calculation module 110 calculates the number of features for each extracted unit image. In addition, based on the analysis result of the continuous CT image, the doubling time (DT, Doubling Time) of the nodules included in the pre-nodule image and the follow-up-nodule image, and the percentage change in the pulmonary nodule volume (PVC) are determined. Calculate.

In the embodiment, the image input to the calculation module 110 may include not only a continuous CT image, but also a continuous voxcel. A voxel is a volume element and represents a value in a regular grid unit in a three-dimensional space. The term voxel is a hybrid word that combines volume and pixel, and is a metaphor for displaying two-dimensional image data as pixels. Voxels are often used for visualization and analysis of medical and scientific data. Not only the pixels but also the voxels themselves do not have spatial coordinates, but can be guessed from the positional relationship with other voxel groups (that is, each position of the data structure constituting one stereoscopic image). That is, in the embodiment, the calculation module receives the voxel type 3D prenodule cube and follow-up nodule cube, extracts each feature of the input cube, and calculates the doubling type and the volume change rate of the nodule through successively input cube information. can do. In an embodiment, the CT reading server may further include a preprocessing module (not shown) to convert a continuous CT image into a voxel-shaped 3D cube.

The extraction module 130 of the network B collects each feature information extracted from the pre-nodule image and the follow-up-nodule image or the three-dimensional pre-nodule cube and the follow-up nodule cube, and applies the learned weight to the collected feature information. Reflect. After that, the feature information whose weight is less than a certain level is filtered to extract the analysis valid information. Referring to FIG. 4, the extraction module 130 may generate a total of 1024 bytes of feature information (a) by collecting 512 byte feature information of the pre-nodule and 512 byte feature information of the follow-up nodule from the calculation module 110. have. Thereafter, the extraction module 130 may extract only the analysis valid information (b) of 64 bytes having a weight equal to or greater than a certain level by reflecting the weight learned in advance to the 1024 byte feature information. In an embodiment, the analysis valid information (b) may be information obtained by filtering information not necessary for calculating cancer probability from the feature information. In the embodiment, the weight is reflected in the order in which the metadata of the feature information (a) is many, or the information having a weight of a certain level or higher is extracted as the analysis valid information (b) by reflecting the weight according to the degree to which the edge portion of the object is included. In addition, artificial intelligence algorithms such as deep learning can be used to reflect weights.

The generation module 150 extracts only data having metadata of a certain level or higher from the analysis valid information, or generates analysis preliminary data by applying an artificial intelligence algorithm, and the generated preliminary analysis data received from the calculation module 110 Analysis data (c) is generated by adding information on the volume change of the lung nodule of the nodule image and the follow-up-nodule image. Here, examples of pulmonary nodule volume change information may be doubling time and Percentage Volume Change (PVC). There can be a total of 4 data used in the previous step to generate the final analysis data, and two of these data are image features extracted and selected from the pre-nodule image and the follow-up-nodule image, and the remaining two data. Is the pulmonary nodule volume change information, and the doubling time and PVC illustrated above can be used.

The analysis preliminary data may be data obtained by re-extracting only important information from the analysis valid information. Thereafter, the generation module 150 selects only the most important data from the analysis data c through an artificial intelligence algorithm such as a deep learning model to generate the final analysis data d.

The calculation module 170 compares the final analysis data d with a previously learned cancer diagnosis data set to calculate a cancer probability of a nodule.

Referring to FIG. 4, in the case of a 3D CNN model used for CT image diagnosis in an embodiment, it is composed of two networks (Network A and Network B). In an embodiment, network A and network B may be separated from each other to perform learning. One nodule in the form of a 3D cube is input to each network A, and all nodules are used as learning data to learn the network A, and learn to find features related to lung cancer. In the case of network B, the characteristics and volume change information of two pulmonary nodules are input, and as a result, the probability of cancer of the pulmonary nodule is provided. In an embodiment, a data augmentation technique using rotation and flip may be used for network learning.

Referring to FIG. 4, in network A, a prenodule image and a follow-up nodule image or a 3D pre-nodule cube and a follow-up nodule cube are input to extract features from two consecutive images and cubes. Fig. 4 shows a feature extraction process from two consecutive images and three orders of magnitude of the cube. In the first feature extraction process, two consecutive images and a cube can be received and two three-dimensional features can be calculated. As a result, in the first feature extraction process, feature data of (32x32x32)x2 size is calculated.

In the second feature extraction process, from the (32x32x32)x2 size data generated as a result of the first feature extraction, the feature extraction process (Max pooling) is performed again, leaving only the unit images with the maximum features. As a result, in the second extraction process, four feature values are calculated, and data having a size of (16x16x16)x4 may be generated. When the third feature extraction process is repeatedly performed, in the third feature extraction process, only the unit images having the largest feature are filtered again from the data generated as a result of the second feature extraction. As a result, data of (8x8x8)x8 size can be generated in the third feature extraction process. In the embodiment, the maximum feature extraction (Max pooling) process is repeated to extract only feature data from a continuous image and transmit it to the network B. After that, the network B generates an analysis image (a) in which the feature data of the consecutive images are merged, and it repeatedly extracts valid data necessary for cancer determination based on the machine learning result. After that, the analysis data (c) in which the volume change information of the prenodule and the follow-up nodule is inserted is generated, and the final analysis data (d) is extracted from the analysis data (c). In an exemplary embodiment, in the extraction process, the final analysis data (d) may be extracted using the results of deep learning and machine learning for cancer determination that have been built in advance. Thereafter, network B determines whether the lung nodule is cancer based on the final analysis data.

5 is a diagram illustrating a detailed data processing block of a calculation module according to an embodiment.

Referring to FIG. 5, the calculation module 110 may include a generator 111, an operation unit 113, and a determination unit 115. When FIG. 5 is described to help understand the function of the calculation module 110, FIG. 6 showing an image feature extraction process will be described together.

The generation unit 111 generates a secondary layer consisting of a certain number of unit images, extracts the maximum feature value image having the largest number of features among the unit images constituting the secondary layer, and extracts from each of the plurality of secondary layers. The nth order analysis image is generated by collecting the maximum feature value images. As shown in FIG. 6, in the embodiment, the generation unit 111 generates secondary layers L21, L22, L23, and L24 including four unit images, and features among the unit images constituting the secondary layer. The maximum feature value image with the largest number is extracted. In the embodiment, an image representing 7 as the maximum feature value is extracted from the secondary layers 2,1,1,7, and an image representing 6 as the maximum feature value is extracted from the secondary layers 2,1,6,1. Through the process, images representing 6 and 8 as maximum feature values are extracted.

Thereafter, the generation unit 111 collects the maximum feature value images extracted from each of the plurality of secondary layers to generate an n-th analysis image. Referring to FIG. 6, a second analysis image e2 is generated by collecting unit images representing the maximum feature values as 7,6,6,8. Thereafter, the third analysis image e3 may be generated by extracting 8, which is the maximum feature value among 7,6,6,8. In the embodiment, up to the third analysis embodiment has been described, but the feature extraction through the maximum pooling process may be repeated up to n times, thereby generating an nth analysis image.

The operation unit 113 calculates the doubling time DT and the nodule volume change rate by using the input continuous CT image and characteristic information of the continuous 3D nodule cube data.

The doubling time may be calculated through Equation 1.

Equation 1

DT=doubling time

V1: Nodule volume of pre-nodule image

V2: Nodule volume in follow up-nodule images

T1: Baseline, T2: Diagnostic scan time

The percentage change in the pulmonary nodule volume may be calculated through Equation 2.

Equation 2

PVC: Percent change in pulmonary nodule volume

V2: follow up-nodule image nodule volume

V1: Volume of pre-nodule image nodules

When the n-th analysis image includes the maximum number of features, the determination unit 115 transmits the pre-nodule image and the maximum feature value image information of the follow-up-nodule image to the extraction module. In addition, the determination unit 115 transmits the doubling time and pulmonary nodule volume percentage change data to the extraction module of the network B.

Hereinafter, an online lung cancer diagnosis method according to an embodiment will be sequentially described. Since the function (function) of the online lung cancer diagnosis method is essentially the same as the function of the lung cancer screening CT reading server and the lung cancer screening system, the overlapping description of FIGS. 1 to 6 will be omitted.

7 is a diagram showing a data processing flow of a lung cancer diagnosis method of a lung cancer screening CT reading server using artificial intelligence according to an embodiment.

In step S710, each feature is extracted from the continuous CT image, the pre-nodule image and the follow-up nodule image. In step S710, a 3D pre-nodule cube and a follow-up nodule cube are generated by converting the continuous CT image into a voxel form, and features included in the 3D cube are extracted by analyzing each of the pre-nodule cube and the follow-up nodule cube. I can.

In step S720, the calculation module calculates the nodule doubling time and the pulmonary nodule volume percentage.

In step S730, the extraction module collects feature information extracted from each successive CT image, reflects the learned weight in the collected feature information, and filters feature information whose weight is less than a certain level to generate analysis valid information.

In step S740, the generation module generates analysis data by adding the doubling time and pulmonary nodule volume change calculated as the result of the continuous CT image analysis to the extracted valid information. In step S750, the final analysis valid information is extracted from the analysis data Generate analysis data.

In step S760, the calculation module calculates the cancer probability of the nodule by comparing the generated final analysis data with the learned cancer diagnosis data set.

8 is a view showing a feature extraction process of a continuous image in step S710.

In step S711, the generation unit generates a secondary layer composed of a predetermined number of unit images.

In step S713, the maximum feature value image having the largest number of features among the unit images constituting the second layer is extracted, and the maximum feature value images extracted from each of the plurality of secondary layers are collected to generate an nth-order analysis image.

In step S715, when the nth analysis image includes the maximum feature value in the determination unit, each of the maximum feature value image information of the continuous CT images is transmitted to the generation module.

9 is a diagram showing an example of a continuous CT image (prenodule image, follow-up nodule image).

As shown in Fig. 9, the lung cancer screening CT reading server using artificial intelligence according to the embodiment includes prenodule images (1-1, 2-1, 3-1) and follow-up nodule images (1-2,2- 2,3-2) are continuously inputted, analysis data and final analysis data are generated using two consecutive images, and cancer determination is performed through the final analysis data.

The lung cancer diagnosis system using artificial intelligence according to the embodiment, the lung cancer screening CT reading server, and the lung cancer diagnosis method can quickly obtain the reading results of the lung cancer screening CT images conducted in individuals and clinics, and improve the accuracy of the CT image reading results. Can be improved. In addition, through online, users can view their CT image reading results at any time, and manage their own medical image information by comparing them with previous results from time to time during follow-up tests. The CT image reading server minimizes the size of image data to be analyzed using artificial intelligence, thereby improving data processing efficiency and improving image reading accuracy. In addition, through the introduction of a CT image reading system, individual medical personnel and tertiary hospitals can perform online medical image reading, thereby promoting profit creation for individuals and medical institutions.

The disclosed contents are only examples, and various changes can be made by those of ordinary skill in the art without departing from the gist of the claims claimed in the claims, so the scope of protection of the disclosed contents is It is not limited to the examples.

Claims (14)

In the lung cancer screening CT reading server using artificial intelligence,
Pre-processed pre-nodule image and follow up-nodule image or continuous CT (Computed Tomography) image including three-dimensional pre-nodule cube and follow-up nodule cube, respectively, as a plurality of units Masking with a primary layer divided into images, extracting features of each unit image included in the primary layer, calculating the number of features for each extracted unit image, and calculating the number of features for each of the extracted unit images, and the pre-nodule image and the follow-up-nodule image. A calculation module for calculating a doubling time (DT) and a percentage change in a pulmonary nodule volume (PVC) of the included nodule;
An extraction module that collects feature information extracted from each of the continuous CT images, reflects the learned weight in the collected feature information, filters feature information whose weight is less than a certain level, and extracts analysis valid information;
A generation module for generating analysis data by adding the doubling time and pulmonary nodule volume change calculated as the result of the continuous CT image analysis to the extracted valid information, and extracting final analysis valid information from the analysis data to generate final analysis data ; And
A calculation module for calculating a cancer probability of a nodule by comparing the generated final analysis data with a learned cancer diagnosis data set; Lung cancer screening CT reading server comprising a.
The method of claim 1, wherein the calculation module
A secondary layer consisting of a certain number of unit images is generated, the maximum feature value image with the largest number of features is extracted from the unit images constituting the secondary layer, and the maximum feature value image extracted from each of the plurality of secondary layers A generator that collects the data and generates an n-th analysis image;
A determination unit for transmitting each of the maximum feature value image information of the consecutive CT images to a generation module when the n-th analysis image includes a maximum feature value; And
A calculation unit for calculating a doubling time (DT) of a nodule included in the pre-nodule image and the follow-up-nodule image, and a percentage change in a pulmonary nodule volume (PVC); Including
The determination unit
When the n-th analysis image does not include a maximum feature value, the n-th analysis image is input again to the generation unit to generate an n+1st analysis image.
The method of claim 1, wherein the calculation module
Equation 1

DT=doubling time
V1: Nodule volume of pre-nodule image
V2: Nodule volume in follow up-nodule images
T1: baseline, T2: time after completion of scan shooting (diagnostic scan time)
Calculate the doubling time using
Equation 2

PVC: Percent change in pulmonary nodule volume
V2: follow up-nodule image nodule volume
V1: Volume of pre-nodule image nodules
Lung cancer screening CT reading server, characterized in that calculating the percentage change in the volume of lung nodules using.
The method of claim 1, wherein the calculation module
In the unit image included in the primary layer of each of the pre-processed pre-nodule image and follow up-nodule image, the number of features and features including edges, objects and colors Lung cancer screening CT reading server, characterized in that to calculate. The method of claim 1, wherein the lung cancer screening CT reading server
A pre-processing module that performs pre-processing of a pre-nodule image and a follow up-nodule image to generate a 3D pre-nodule cube and a 3D follow-up nodule cube in a voxcel format; Lung cancer screening CT reading server, characterized in that it further comprises.
In the online lung cancer diagnosis system using artificial intelligence,
A user terminal for collecting personal health information including X-rays, CT images of pulmonary nodules, and health examination information;
A medical institution server for collecting CT images of lung nodules including pre-nodule images and follow up-nodule images of patients; And
Including; a lung cancer screening CT reading server for receiving the CT image of the lung nodule from the user terminal or the medical institution server, calculating the probability that the lung nodule is cancer, and transmitting the calculation result to the medical institution server and the user terminal,
The lung cancer screening CT reading server,
A pre-processing module that performs pre-processing of a pre-nodule image and a follow up-nodule image to generate a 3D pre-nodule cube and a 3D follow-up nodule cube in a voxcel format;
A pre-processed pre-nodule image and a follow up-nodule image or a continuous CT image including a three-dimensional pre-nodule cube and a follow-up nodule cube are divided into a plurality of unit images. Masking with the first layer, extracting features of each unit image included in the first layer, calculating the number of features for each extracted unit image, and determining the number of nodules included in the pre-nodule image and the follow-up-nodule image. A calculation module for calculating a doubling time (DT) and a percentage change in a pulmonary nodule volume (PVC);
An extraction module that collects feature information extracted from each of the continuous CT images, reflects the learned weight in the collected feature information, filters feature information whose weight is less than a certain level, and extracts analysis valid information;
A generation module for generating analysis data by adding the doubling time and pulmonary nodule volume change calculated as the result of the continuous CT image analysis to the extracted valid information, and extracting final analysis valid information from the analysis data to generate final analysis data ; And
And an operation module that compares the generated final analysis data with a learned cancer diagnosis data set to calculate a cancer probability of a nodule.
delete The method of claim 6, wherein the calculation module
A secondary layer consisting of a certain number of unit images is generated, the maximum feature value image with the largest number of features is extracted from the unit images constituting the secondary layer, and the maximum feature value image extracted from each of the plurality of secondary layers A generator that collects the data and generates an n-th analysis image;
A determination unit for transmitting each of the maximum feature value image information of the consecutive CT images to a generation module when the n-th analysis image includes a maximum feature value; And
A calculation unit for calculating a doubling time (DT) of a nodule included in the pre-nodule image and the follow-up-nodule image, and a percentage change in a pulmonary nodule volume (PVC); Including
The determination unit
When the n-th analysis image does not include a maximum feature value, the n-th analysis image is input again to the generation unit to generate an n+1st analysis image.
The method of claim 6, wherein the calculation module
Equation 1

DT=doubling time
V1: Nodule volume of pre-nodule image
V2: Nodule volume in follow up-nodule images
T1: Baseline, T2: Diagnostic scan time
Calculate the doubling time using
Equation 2

PVC: Percent change in pulmonary nodule volume
V2: follow up-nodule image nodule volume
V1: Volume of pre-nodule image nodules
Online lung cancer diagnosis system, characterized in that to calculate the percentage change in the volume of the lung nodule using.
delete delete delete delete delete

Images