Multiple artificial neural network-based estimated dose assessment method

Abstract

The present invention relates to a multiple artificial neural network-based estimated dose assessment method capable of obtaining an estimated dose with high accuracy by correcting the frequency of dicentric chromosomes based on the estimated number of chromosomes, number of chromosome fragments, and frequency of dicentric chromosomes obtained through multiple neural network models, and then determining an estimated dose corresponding to the corrected frequency of dicentric chromosomes.

Description

Estimation dose examination method based on multiple artificial neural networks

[0001] The present invention relates to a method for examining an estimated dose by passing a chromosome image through a sequentially connected multiple artificial neural network.

[0002]

[0003] In general, the estimated radiation dose (Gy) can be calculated by calculating the frequency of the exposed mobile chromosome and inputting this into the estimated dose calculation formula.

[0004] To this end, a process of identifying the mobile chromosome is first performed. However, when exposed to a low dose of radiation, typically 0 to 1 Gy, many false mobile chromosomes are identified, which reduces the accuracy of the mobile chromosome frequency, and there is a disadvantage in that it is difficult to estimate an accurate radiation dose.

[0005] Considering the above-mentioned points, there is a need for the development of technology to obtain more accurate radiation dose estimates by preventing the identification of false mobile source chromosomes.

[0006]

[0007] The present invention aims to correct the frequency of a mobile chromosome based on the number of chromosomes, the number of chromosome fragments, and the frequency of the mobile chromosome estimated through a multiple neural network model, and to determine an estimated dose corresponding to the frequency of the corrected mobile chromosome.

[0008] The purposes of the present invention are not limited to those mentioned above. Other purposes and advantages of the present invention not mentioned above can be understood through the following description and will be more clearly understood through the embodiments of the present invention. Furthermore, it will be readily apparent that the purposes and advantages of the present invention can be realized by the means and combinations thereof set forth in the claims.

[0009]

[0010] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention includes the steps of inputting a plurality of chromosome images, which are targets of dose assessment, into a first neural network model to determine whether they are readable, inputting a chromosome image for reading, which is determined to be readable as a result of the determination, into a second neural network model to identify the number of chromosomes in the chromosome image for reading, inputting the chromosome image for reading into a third neural network model to identify the number of chromosome fragments in the chromosome image for reading, inputting the chromosome image for reading into a fourth neural network model to identify the number of source chromosomes, calculating a frequency of source chromosomes based on the number of source chromosomes, inputting the number of chromosomes, the number of chromosomes, and the frequency of source chromosomes into a fifth neural network model to correct the frequency of source chromosomes, and determining an estimated dose corresponding to the corrected frequency of source chromosomes.

[0011] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention further includes a step of collecting a learning chromosome image, a step of labeling whether a chromosome in the learning chromosome image can be read, and a step of setting the learning chromosome image as input data of the first neural network model and setting the readability as output data of the first neural network model to perform supervised learning of the first neural network model.

[0012] The above first neural network model outputs one of two classes corresponding to readability and unreadability of the chromosome image, respectively.

[0013] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention further includes a step of collecting a learning chromosome image, a step of labeling the position of an individual chromosome in the learning chromosome image, and a step of setting the learning chromosome image as input data of the second neural network model and setting the labeled position as output data of the second neural network model to perform supervised learning of the second neural network model.

[0014] The second neural network model outputs the location of each chromosome within the chromosome image for reading.

[0015] The step of identifying the number of chromosomes includes the step of counting the number of bounding boxes corresponding to the positions of individual chromosomes output from the second neural network model.

[0016] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention further includes a step of collecting a learning chromosome image, a step of labeling the location of an individual chromosome segment in the learning chromosome image, and a step of setting the learning chromosome image as input data of the third neural network model and setting the labeled location as output data of the third neural network model to perform supervised learning of the third neural network model.

[0017] The third neural network model outputs the location of individual chromosome segments within the chromosome image for reading.

[0018] The step of identifying the number of chromosome fragments includes the step of counting the number of bounding boxes corresponding to the positions of individual chromosome fragments output from the third neural network model.

[0019] The fourth neural network model includes a fourth neural network model that outputs the location of each mobile chromosome within the chromosome image for reading, and a fourth neural network model that performs a classification operation on the mobile chromosome corresponding to each location output from the fourth neural network model.

[0020] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention further includes a step of cropping a bounding box output to the 4-1 neural network model and inputting an image of the cropped bounding box to the 4-2 neural network.

[0021] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention further includes the steps of collecting a learning chromosome image, labeling the position of an individual mobile source chromosome in the learning chromosome image, and setting the learning chromosome image as input data of a 4-1 neural network model and setting the labeled position as output data of the 4-1 neural network model to perform supervised learning of the 4-1 neural network model.

[0022] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention further includes a step of collecting a learning chromosome image, a step of labeling a class determined based on whether a chromosome expressed in the learning chromosome image is a mobile source chromosome, and a step of setting the learning chromosome image as input data of a 4-2 neural network model and setting the labeled class as output data of the 4-2 neural network model to perform supervised learning of the 4-2 neural network model.

[0023] The step of identifying the number of the above-mentioned mobile chromosomes includes the step of counting the number of classes output as mobile chromosomes in the above-mentioned 4-2 neural network model.

[0024] The step of calculating the frequency of the above-mentioned mobile chromosome includes the step of calculating the frequency by dividing the number of the above-mentioned mobile chromosome by the number of the above-mentioned chromosome images for reading.

[0025] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention further includes a step of collecting the number of chromosomes, the number of chromosome fragments, and the frequency of the source chromosome, a step of labeling the actual frequency of the source chromosome in the chromosome image for reading, and a step of setting the number of chromosomes, the number of chromosomes, and the number of source chromosomes as input data of the fifth neural network model, and setting the actual frequency as output data to supervise learning of the fifth neural network model.

[0026] The step of determining the estimated dose includes the step of determining the estimated dose by substituting the corrected frequency of the mobile chromosome into an equation defining a correlation between the frequency of the mobile chromosome and the radiation dose.

[0027]

[0028] The present invention has the advantage of being able to obtain an estimated dose with high accuracy by correcting the frequency of the mobile chromosome based on the number of chromosomes, the number of chromosome fragments, and the frequency of the mobile chromosome estimated through a multiple neural network model, and determining an estimated dose corresponding to the frequency of the corrected mobile chromosome.

[0029] In addition to the effects described above, specific effects of the present invention are described below while explaining specific details for carrying out the invention.

[0030]

[0031] FIG. 1 is a flowchart illustrating a multi-artificial neural network-based estimation dose inspection method according to one embodiment of the present invention.

[0032] Figure 2 is a diagram for explaining the operation of the first neural network model.

[0033] Figure 3a is a chromosome image for reading.

[0034] Figure 3b is an image of an unreadable chromosome.

[0035] Figure 4 is a diagram for explaining the operation of the second neural network model.

[0036] Figure 5 is a chromosome image for reading showing the identified chromosomes and their number.

[0037] Figure 6 is a diagram for explaining the operation of the third neural network model.

[0038] Figure 7 is a chromosome image for reading showing the identified chromosome fragments and their numbers.

[0039] Figure 8 is a diagram for explaining the operation of the 4-1 neural network model.

[0040] Figure 9 is a chromosome image for reading showing the identified mobile chromosomes and their number.

[0041] Figures 10 and 11 are diagrams for explaining the operation of the 4-2 neural network model.

[0042] Figure 12 is a diagram for explaining the operation of the fifth neural network model.

[0043] Figure 13 is a diagram showing the average estimated dose error depending on whether the present invention is applied.

[0044]

[0045] The above-described objects, features, and advantages will be described in detail below with reference to the accompanying drawings, so that those skilled in the art can easily practice the technical idea of ​​the present invention. In describing the present invention, if it is determined that a detailed description of known technologies related to the present invention may unnecessarily obscure the gist of the present invention, a detailed description thereof will be omitted. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals are used to indicate the same or similar components.

[0046] Although terms like "first" and "second" are used herein to describe various components, these components are not limited by these terms. These terms are used merely to distinguish one component from another, and unless otherwise specified, a "first" component may also be a "second" component.

[0047] Additionally, in this specification, the phrase "upper (or lower)" or "upper (or lower)" of any component may mean that any component is positioned not only in contact with the upper surface (or lower surface) of the component, but also that other components may be interposed between the component and any component positioned on (or below) the component.

[0048] Additionally, when it is described herein that a component is “connected,” “coupled,” or “connected” to another component, it should be understood that the components may be directly connected or connected to each other, but that other components may be “interposed” between the components, or that each component may be “connected,” “coupled,” or “connected” through another component.

[0049] Additionally, the singular expressions used herein include plural expressions unless the context clearly dictates otherwise. In this application, terms such as "consist of" or "comprises" should not be construed to necessarily include all of the various components or various steps described in the specification, and should be construed to mean that some of the components or some of the steps may not be included, or that additional components or steps may be included.

[0050] Also, in this specification, when it says "A and / or B", it means A, B or A and B, unless otherwise specifically stated, and when it says "C to D", it means C or more and D or less, unless otherwise specifically stated.

[0051]

[0052] A method for estimating dose based on a multi-artificial neural network according to one embodiment of the present invention may include a step of determining whether a plurality of chromosome images can be read (S100), a step of identifying the number of chromosomes in the chromosome images for reading (S200), a step of identifying the number of chromosome fragments in the chromosome images for reading (S300), a step of identifying a mobile source chromosome in the chromosome images for reading (S400), a step of calculating a frequency of the mobile source chromosome based on the number of the mobile source chromosomes (S500), a step of correcting the frequency of the mobile source chromosome (S600), and a step of determining an estimated dose corresponding to the frequency of the corrected mobile source chromosome (S700).

[0053] However, the multi-artificial neural network-based estimated dose inspection method illustrated in FIG. 1 is according to one embodiment, and the steps forming the invention are not limited to the embodiment illustrated in FIG. 1, and some steps may be added, changed, or deleted as needed.

[0054] Each of the steps illustrated in FIG. 1 may be performed by a processor, and for this purpose, the processor may include at least one physical element among application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, and micro-controllers.

[0055]

[0056] Hereinafter, each step (S100 to S700) illustrated in Fig. 1 will be described in detail.

[0057] The processor can input multiple chromosome images that are the subject of dose evaluation into the first neural network model (10) to determine whether they are readable (S100). Specifically, as illustrated in FIG. 2, the processor can input multiple chromosome images that are the subject of dose evaluation into the first neural network model (10), and the first neural network model (10) can output one of two classes corresponding to readability and unreadability of the input chromosome images, respectively.

[0058] At this time, the chromosome image may be a chromosome image that was not used for learning the first neural network model (10), and the first neural network model (10) may receive the chromosome image as input and output a class regarding whether it is readable or unreadable.

[0059] For example, the first neural network model (10) can receive the chromosome image illustrated in FIG. 3A as input. The chromosome image illustrated in FIG. 3A can be an image in which the shape of the chromosome expressed in the chromosome region (A) is clear, and the focus is clear and distinct. Accordingly, the first neural network model (10) can output a class for readability. As a result of the output, a chromosome image output as readable can be called a chromosome image for reading.

[0060] On the other hand, the first neural network model (10) may receive the chromosome image depicted in Fig. 3b as input. The chromosome image depicted in Fig. 3b may be an image in which the shape of the chromosome represented in the chromosome region (B) is unclear, and may be out of focus or unclear. Accordingly, the first neural network model (10) may output a class indicating "unreadable."

[0061] Meanwhile, for the above-described operations, a training process of the first neural network model (10) may be performed in advance. To this end, the processor may collect a first training chromosome image. As described above, the first training chromosome image may be an image in which the chromosome shape is clear, with a clear and distinct focus, or an image in which the chromosome shape is unclear, with a blurred or unclear focus.

[0062] Next, the processor can label the readability of the first training chromosome image according to the user's command. In more detail, each first training chromosome image can be labeled with a different class (Class) corresponding to the ground truth (GT), i.e., whether the first training chromosome image is readable or not.

[0063] Next, the processor may set the first learning chromosome image as input data of the first neural network model (10) and set the readability as output data of the first neural network model (10) to perform supervised learning in advance on the first neural network model (10). Specifically, the processor may set the first learning chromosome image as input data of the first neural network model (10) and perform supervised learning in advance on the first neural network model (10) based on a training dataset having as output data a class labeled according to the readability, i.e., a measured value corresponding to the input data.

[0064] Through this supervised learning, the first neural network model (10) can learn the correlation between the first learning chromosome image and whether it can be read or not, and after learning is completed, it can classify whether it can be read or not even for chromosome images that were not used for learning.

[0065] Next, the processor inputs the chromosome image for reading, which is determined to be readable through the first neural network model (10), into the second neural network model (20), thereby identifying the number of chromosomes in the chromosome image for reading (S200).

[0066] To this end, as illustrated in FIG. 4, the processor can input a chromosome image for reading into a second neural network model (20), and the second neural network model (20) can output the location of an individual chromosome within the input chromosome image for reading.

[0067] Meanwhile, for the aforementioned operations, a training process of the second neural network model (20) may be performed in advance. To this end, the processor may collect a second training chromosome image. At this time, the second training chromosome image may be an image with a clear chromosome shape and a sharp focus, as described above.

[0068] Next, referring to FIG. 5, the processor may label bounding boxes (C) that delimit individual chromosomes within a second training chromosome image according to a user's command. Specifically, the processor may label, according to a user's command, the actual values ​​corresponding to the second training chromosome image, i.e., the center coordinates, width, and height of the bounding boxes (C) that delimit each chromosome expressed in the second training chromosome image.

[0069] Next, the processor may set the second learning chromosome image as input data of the second neural network model (20) and set the labeled location, i.e., the bounding box (C), as output data of the second neural network model (20) to perform supervised learning in advance on the second neural network model (20). Specifically, the processor may set the second learning chromosome image as input data of the second neural network model (20) and perform supervised learning in advance on the second neural network model (20) based on a training dataset that has as output data the measured values ​​corresponding to the input data, i.e., the center coordinates, width, and height of the bounding box (C) that demarcates individual chromosomes in the second learning chromosome image.

[0070] Through this supervised learning, the second neural network model (20) can learn the correlation between the second learning chromosome image and the location of the chromosome, and after learning is completed, the location of each chromosome can be identified even for chromosome images that were not used.

[0071] Next, the processor can count the number of bounding boxes (C) corresponding to the positions of individual chromosomes output from the second neural network model (20). This allows the processor to identify the number of individual chromosomes. That is, as shown in FIG. 5, the processor can identify 45 individual chromosomes.

[0072] Next, the processor inputs the chromosome image for reading, which is determined to be readable through the first neural network model (10), into the third neural network model (30), thereby identifying the number of chromosome fragments in the chromosome image for reading (S300).

[0073] To this end, as illustrated in FIG. 6, the processor can input a chromosome image for reading into a third neural network model (30), and the third neural network model (30) can output the location of individual chromosome fragments within the input chromosome image for reading.

[0074] Meanwhile, the training process of the third neural network model (30) may be performed prior to the aforementioned operations. To this end, the processor may collect a third training chromosome image. At this time, the third training chromosome image may likewise be an image with a clear chromosome shape and a sharp focus.

[0075] Next, referring to FIG. 7, the processor may label bounding boxes (D) that delimit individual chromosome segments within a training chromosome image according to a user's command. Specifically, the processor may label, according to a user's command, the actual values ​​corresponding to the third training chromosome image, i.e., the center coordinates, width, and height of the bounding boxes (D) that delimit each chromosome segment expressed in the third training chromosome image.

[0076] Next, the processor may set the third learning chromosome image as input data of the third neural network model (30) and set the labeled location, i.e., the bounding box (D), as output data of the third neural network model (30) to perform supervised learning in advance on the third neural network model (30). Specifically, the processor may set the third learning chromosome image as input data to the third neural network model (30) and perform supervised learning in advance on the third neural network model (30) based on a training dataset that has as output data the measured values ​​corresponding to the input data, i.e., the center coordinates, width, and height of the bounding box (D) that demarcates individual chromosome segments in the third learning chromosome image.

[0077] Through this supervised learning, the third neural network model (30) can learn the correlation between the third learning chromosome image and the location of the chromosome fragment, and after learning is completed, the location of each chromosome fragment can be identified even for chromosome images that were not used.

[0078] Next, the processor can count the number of bounding boxes (C) corresponding to the positions of individual chromosome segments output from the third neural network model (30). This allows the processor to identify the number of individual chromosome segments. That is, as shown in FIG. 7, the processor can identify one individual chromosome.

[0079] Next, the processor inputs the chromosome image for reading into the fourth neural network model to identify the number of mobile chromosomes (S400). At this time, the operation of identifying the number of mobile chromosomes is identical to the operation of identifying the number of chromosomes or chromosome fragments described above, and can be accomplished with just one fourth neural network model.

[0080] However, to increase the accuracy of identification of the mobile chromosome, the fourth neural network model may include the 4-1 neural network model and the 4-2 neural network model.

[0081] First, regarding the operation of the 4-1 neural network model, the processor inputs a chromosome image for reading, which is determined to be readable through the first neural network model (10), into the 4-1 neural network model (40a), thereby identifying the location of the moving source chromosome within the chromosome image for reading.

[0082] Specifically, as illustrated in FIG. 8, the processor can input a chromosome image for reading into the 4-1 neural network model (40a), and the 4-1 neural network model (40a) can output the location of each mobile source chromosome within the input chromosome image for reading.

[0083] Meanwhile, for the aforementioned operations, the training process of the 4-1 neural network model (40a) may be preceded. To this end, the processor may collect a 4-1 training chromosome image. At this time, the 4-1 training chromosome image may likewise be an image with a clear chromosome shape and a sharp focus.

[0084] Next, referring to FIG. 9, the processor can label bounding boxes (E) that delimit individual mobile source chromosomes within the 4-1 training chromosome image according to a user's command. Specifically, the processor can label, according to the user's command, the actual values ​​corresponding to the 4-1 training chromosome image, i.e., the center coordinates, width, and height of the bounding boxes (E) that delimit each mobile source chromosome expressed in the 4-1 training chromosome image.

[0085] Next, the processor sets the 4-1 learning chromosome image as input data of the 4-1 neural network model (40a), and sets the labeled location, i.e., the bounding box (E), as output data of the 4-1 neural network model (40a), so that the 4-1 neural network model (40a) can be supervised in advance. Specifically, the processor sets the 4-1 learning chromosome image as input data to the 4-1 neural network model (40a), and performs supervised learning of the 4-1 neural network model (40a) in advance based on a training dataset that has as output data the measured values ​​corresponding to the input data, i.e., the center coordinates, width, and height of the bounding box (E) that demarcates the individual moving source chromosomes in the 4-1 learning chromosome image.

[0086] Through this supervised learning, the 4-1 neural network model (40a) can learn the correlation between the positions of the training chromosome images and the source chromosomes, and after learning is completed, the positions of individual source chromosomes can be identified even for chromosome images that were not used.

[0087] Next, the processor can crop the bounding box (E) output by the 4-1 neural network model (40a). Accordingly, an image expressing one mobile chromosome can be generated.

[0088] Next, regarding the operation of the 4-2 neural network model, the processor can input an image of a cropped bounding box (E) into the 4-2 neural network model (40b) to perform a classification operation on the moving source chromosome.

[0089] Specifically, as illustrated in FIGS. 10 and 11, the processor can input a cropped bounding box (E) image into the 4-2 neural network, and perform a classification operation on the moving source chromosome within the input bounding box (E) image.

[0090] Specifically, as illustrated in FIG. 10, the processor can input a cropped bounding box (E) image into the 4-2 neural network model, and the 4-2 neural network model can output a class corresponding to the input bounding box (E) image.

[0091] At this time, the cropped bounding (E) box image may be an image that was not used in the learning of the 4-2 neural network (40b) model, and the 4-2 neural network model (40b) may output a class for the cropped bounding (E) box image.

[0092] For example, the 4-2 neural network model (40b) can receive the cropped bounding box (E) image illustrated in FIG. 10 as input. The bounding box (E) image illustrated in FIG. 10 may be a moving chromosome image (dic). Accordingly, the 4-2 neural network model (40b) can output a class corresponding to the moving chromosome.

[0093] On the other hand, the 4-2 neural network model (40b) can receive the cropped bounding box (E) image illustrated in FIG. 11 as input. The bounding box (E) image illustrated in FIG. 11 may be a non-motile chromosome (non-dic). Accordingly, the 4-2 neural network model (40b) can output a class corresponding to a non-motile chromosome.

[0094] Meanwhile, for the aforementioned operations, the training process of the 4-2 neural network model (40b) may be performed first. To this end, the processor may collect a cropped bounding box image. At this time, the bounding box image may be a cropped bounding box image.

[0095] Next, the processor can label whether a chromosome is a mobile chromosome within a bounding box according to a user's command. Specifically, the processor can label the bounding box image with a corresponding ground truth value, i.e., a class of a mobile chromosome or a non-mobile chromosome expressed in the bounding box image, according to the user's command.

[0096] Next, the processor may set the bounding box image as input data of the 4-2 neural network model (40b) and set whether there is a mobile chromosome as output data of the 4-2 neural network model (40b) to perform supervised learning in advance on the 4-2 neural network model (40b). Specifically, the processor may set the bounding box image as input data to the 4-2 neural network model (40b) and perform supervised learning in advance on the 4-2 neural network model (40b) based on a training dataset that has as output data the measured value corresponding to the input data, that is, the class of a mobile chromosome or a non-mobile chromosome in the bounding box image.

[0097] Through this supervised learning, the 4-2 neural network model (40b) can learn the correlation between the bounding box image and whether it is a mobile chromosome, and after learning is completed, it can identify whether it is a mobile chromosome or a non-mobile chromosome even for unused bounding box images.

[0098] Next, the processor can count the number of classes output as mobile chromosomes in the 4-2 neural network model (40b). This allows the processor to identify the number of mobile chromosomes. That is, as illustrated in FIG. 9, the processor can identify two individual mobile chromosomes.

[0099] Next, the processor can calculate the frequency of the mobile chromosome based on the number of identified mobile chromosomes (S500). At this time, the processor can calculate the frequency by dividing the number of mobile chromosomes by the number of chromosome images for reading. For example, if the number of mobile chromosomes is 30 and the number of chromosome images for reading is 3000, the frequency of the mobile chromosome can be 30 / 3000, or 0.01.

[0100] Next, the processor can input the number of chromosomes, the number of chromosome fragments, and the frequency of the mobile chromosome into the fifth neural network model (50) to correct the frequency of the mobile chromosome (S600).

[0101] Specifically, as illustrated in FIG. 12, the processor can input the number of chromosomes, the number of chromosome fragments, and the frequency of the mobile chromosome into the fifth neural network model (50), and the fifth neural network model (50) can output the frequency of the corrected mobile chromosome.

[0102] Meanwhile, the learning process of the fifth neural network model (50) may be performed in advance for the above-described operations. To this end, the processor may collect the number of chromosomes, the number of chromosome fragments, and the frequency of the mobile chromosome.

[0103] Next, the processor can label the actual frequency ratio in the chromosome image for reading according to the user's command. Specifically, the processor can label the actual value corresponding to the chromosome image for reading according to the user's command, i.e., the actual frequency ratio of the mobile chromosome expressed in the chromosome image for reading.

[0104] Next, the processor may set the number of chromosomes, the number of chromosome fragments, and the frequency of the mobile chromosome as input data of the fifth neural network model (50), and may set the actual frequency as output data of the fifth neural network model (50) to perform supervised learning in advance for the fifth neural network model (50). Specifically, the processor may set the number of chromosomes, the number of chromosome fragments, and the frequency of the mobile chromosome as input data of the fifth neural network model (50), and may perform supervised learning in advance for the fifth neural network model (50) based on a training dataset that has the actual values ​​corresponding to the input data, i.e., the actual frequency of the mobile chromosome in the chromosome image for reading, as output data.

[0105] Through this supervised learning, the fifth neural network model (50) can learn the correlation between the number of chromosomes, the number of chromosome fragments, and the frequency of the source chromosome and the actual frequency, and after learning is completed, the actual frequency can also be output for the number of unused chromosomes, the number of chromosome fragments, and the frequency of the source chromosome.

[0106] Next, the processor can determine an estimated dose corresponding to the frequency of the corrected mobile chromosome (S700). Specifically, the processor can determine the estimated dose by substituting the frequency of the corrected mobile chromosome into an equation defining the correlation between the frequency of the mobile chromosome and the radiation dose.

[0107] At this time, the processor inputs the frequency of the corrected mobile chromosome into the calculation formula presented in the Cytogenetic dosimetry (2011 International Atomic Energy Agency) booklet, so that the final estimated dose can be calculated.

[0108] Accordingly, referring to FIG. 13, it can be seen that the error rate of the estimated dose calculated by inputting the corrected mobile source frequency rate through the fifth neural network model (50) into the equation is lower than the error rate of the estimated dose calculated by inputting the uncorrected mobile source chromosome frequency rate.

[0109]

[0110] Although the present invention has been described with reference to the drawings exemplified above, it is to be understood that the present invention is not limited to the embodiments and drawings disclosed herein, and that various modifications may be made by those skilled in the art within the scope of the technical idea of ​​the present invention. Furthermore, even if the operational effects according to the configuration of the present invention have not been explicitly described while describing the embodiments of the present invention, it is natural that the effects predictable by the corresponding configuration should also be acknowledged.

Claims

1. A step of inputting multiple chromosome images that are the subject of quality assessment into a first neural network model to determine whether they are readable; A step of inputting a chromosome image for reading, which is determined to be readable as a result of the above judgment, into a second neural network model to identify the number of chromosomes in the chromosome image for reading; A step of inputting the above-mentioned chromosome image for reading into a third neural network model to identify the number of chromosome fragments in the above-mentioned chromosome image for reading; A step of inputting the above-mentioned chromosome image for reading into a fourth neural network model to identify the number of mobile chromosomes; A step of calculating the frequency of the mobile chromosome based on the number of the mobile chromosome; A step of inputting the number of chromosomes, the number of chromosomes fragments, and the frequency of the mobile chromosome into the fifth neural network model to correct the frequency of the mobile chromosome; and A step of determining an estimated dose corresponding to the frequency of the corrected mobile chromosome; A multi-artificial neural network-based estimation dose examination method.

2. In paragraph 1, Step of collecting chromosome images for training; A step of labeling whether a chromosome in the above learning chromosome image can be read; and A step of setting the learning chromosome image as input data of the first neural network model and setting the readability as output data of the first neural network model to perform supervised learning of the first neural network model is further included. A multi-artificial neural network-based estimation dose examination method.

3. In paragraph 1, The above first neural network model outputs one of two classes corresponding to readability and unreadability of the chromosome image, respectively. A multi-artificial neural network-based estimation dose examination method.

4. In paragraph 1, Step of collecting chromosome images for training; A step of labeling the location of individual chromosomes within the above learning chromosome image; and A step of setting the learning chromosome image as input data of the second neural network model and setting the labeled location as output data of the second neural network model to perform supervised learning of the second neural network model is further included. A multi-artificial neural network-based estimation dose examination method.

5. In paragraph 1, The second neural network model outputs the location of each chromosome within the chromosome image for reading. A multi-artificial neural network-based estimation dose examination method.

6. In paragraph 1, The step of identifying the number of the above chromosomes is A step of counting the number of bounding boxes corresponding to the positions of individual chromosomes output from the second neural network model. A multi-artificial neural network-based estimation dose examination method.

7. In paragraph 1, Step of collecting chromosome images for training; A step of labeling the positions of individual chromosome segments within the above learning chromosome image; and A step of setting the above learning chromosome image as input data of the third neural network model and setting the labeled location as output data of the third neural network model to supervise learning of the third neural network model is further included. A multi-artificial neural network-based estimation dose examination method.

8. In paragraph 1, The third neural network model outputs the location of individual chromosome segments within the chromosome image for reading. A multi-artificial neural network-based estimation dose examination method.

9. In paragraph 1, The step of identifying the number of the above chromosome fragments is A step of counting the number of bounding boxes corresponding to the positions of individual chromosome segments output from the third neural network model. A multi-artificial neural network-based estimation dose examination method.

10. In paragraph 1, The above fourth neural network model is A 4-1 neural network model that outputs the location of each mobile chromosome in the chromosome image for reading, and a 4-2 neural network model that performs a classification operation on the mobile chromosome corresponding to each location output from the 4-1 neural network model. A multi-artificial neural network-based estimation dose examination method.

11. In paragraph 10, A method further comprising the step of cropping the bounding box outputted by the 4-1 neural network model and inputting the image of the cropped bounding box into the 4-2 neural network. A multi-artificial neural network-based estimation dose examination method.

12. In paragraph 10, Step of collecting chromosome images for training; A step of labeling the positions of individual mobile source chromosomes within the above learning chromosome image; A step of setting the above learning chromosome image as input data of the 4-1 neural network model and setting the labeled location as output data of the 4-1 neural network model to supervise learning of the 4-1 neural network model is further included. A multi-artificial neural network-based estimation dose examination method.

13. In paragraph 10, Step of collecting chromosome images for training; A step of labeling a class determined based on whether the chromosome represented in the above learning chromosome image is a mobile chromosome; and A step of setting the above learning chromosome image as input data of the 4-2 neural network model and setting the labeled class as output data of the 4-2 neural network model to supervise learning of the 4-2 neural network model is further included. A multi-artificial neural network-based estimation dose examination method.

14. In paragraph 10, The step of identifying the number of the above mobile chromosomes is Including a step of counting the number of classes output as moving chromosomes in the above 4-2 neural network model. A multi-artificial neural network-based estimation dose examination method.

15. In paragraph 1, The step of calculating the frequency of the above mobile chromosome is A step of calculating the number of the above mobile chromosomes by dividing the number of the above reading chromosome images. A multi-artificial neural network-based estimation dose examination method.

16. In paragraph 1, A step of collecting the number of chromosomes, the number of chromosome fragments and the frequency of the mobile chromosome; A step of labeling the actual frequency of the mobile chromosome in the above-mentioned reading chromosome image; and A step of setting the number of chromosomes, the number of chromosomes, and the number of mobile chromosomes as input data of the fifth neural network model, and setting the actual frequency as output data to supervise learning of the fifth neural network model is further included. A multi-artificial neural network-based estimation dose examination method.

17. In paragraph 1, The step of determining the above estimated dose is A step of determining an estimated dose by substituting the corrected mobile chromosome frequency into an equation defining a correlation between the frequency of the mobile chromosome and the radiation dose. A multi-artificial neural network-based estimation dose examination method.

Images