JP6990540B2 - Video processing equipment, video processing methods, and video processing programs - Google Patents

Description

The present invention relates to a video processing apparatus, a video processing method, and a video processing program, and more particularly to a video processing device, a video processing method, and a video processing program that detects and displays a lesion portion from an input medical image.

Conventionally, image diagnosis is known in which a diagnosis is made by displaying an image of the morphology of an organ, the presence or absence of a lesion, etc., which cannot be understood only by diagnosing from the outside of the body. Image diagnosis is made by a doctor based on images made using, for example, X-rays, ultrasound, magnetic resonance, etc., or inspection images taken under visible light such as endoscopy and fundus cameras. ing.

However, it is not easy to decipher the information from such inspection images, and a lot of specialized knowledge, abundant experience and cases are required. On the other hand, it takes a lot of time and cost to train specialists who specialize in diagnostic imaging, so there is a shortage of specialists, and in medical facilities such as clinics, general doctors who have few opportunities to come into contact with cases perform diagnostic imaging. Sometimes I do it.

In recent years, the technological level of computer functions and artificial intelligence technology has improved, and image judgment systems that utilize artificial intelligence technology have been applied in various industrial fields, and development using artificial intelligence is progressing in the medical field as well. I'm out.

For example, letting a computer analyze past inspection images and using algorithms obtained from useful rules and rules from the results, the computer recognizes new inspection images, and the presence or absence of lesions and whether the lesions are benign or malignant. A technique called machine learning that distinguishes between the two has become possible (see, for example, Patent Document 1).

Japanese Unexamined Patent Publication No. 2015-154918

However, such a lesion candidate verifier can grasp the presence or absence of a lesion candidate in a specific area and the reliability, but clearly grasp the detailed position of the lesion, the shape of the lesion, and the degree of progression of the lesion. I couldn't.

FIG. 21 is a block diagram showing an example of a conventional lesion detection device.
As shown in FIG. 21, the lesion detection device 10 includes a video input unit 11, a lesion candidate detector 12, a peripheral object detector 13, a lesion candidate verifier 14, a candidate remover 15, and a display unit 16.

The image input unit 11 receives a medical image taken by the medical image device 20 of a specific part of the human body using, for example, ultrasonic waves or X-rays.
The lesion candidate detector 12 analyzes the medical image to detect whether or not the lesion candidate is possible and the position of the existing lesion candidate.

The peripheral object detector 13 detects skin, fat, glandular tissue, muscle, bone, etc. in the image as a plurality of anatomical objects in the medical image.
The lesion candidate verifier 14 uses the lesion information 30 including the information related to the anatomical vein to the lesion candidate detected by the lesion candidate detector 12 and the anatomical object detected by the peripheral object detector 13. Then, it is verified whether or not the detected lesion candidate is actually a malignant lesion.

The candidate remover 15 receives the result verified by the lesion candidate verifier 14, removes the lesion candidate determined to be negative by the lesion candidate verifier 14 from the detected lesion candidates, and then removes the remaining lesion candidate. Is output to the display unit 16.

For example, the lesion candidate detector 12 analyzes the medical image received from the image input unit 11 and then determines that a certain area in the medical image is a “lesion candidate with a reliability of 85%”, so that the lesion candidate detector 12 can be a lesion candidate. Is detected.

The peripheral object detector 13 detects a lesion candidate by analyzing the medical image received from the image input unit 11 and then determining a certain area in the medical image as "subcutaneous fat with a reliability of 95%". do.

As a result, the lesion candidate verifier 14 compares the specific region A detected by the lesion candidate detector 12 as the lesion candidate with the position where the anatomical object exists, and by referring to the anatomical information 30, the specific region A can be determined. Determine if it is actually a lesion or negative.

FIG. 22 is a diagram showing a detection image in which a lesion candidate is detected from an inspection image using a lesion detection device.
As shown in FIG. 22, in the lesion candidate detection image 40, two lesion candidate regions 41 and 42 and a glandular tissue region 43 are displayed in a plurality of specific regions in the image.

The lesion candidate region 41 is displayed with a reliability of 69%, and the lesion candidate region 42 is displayed with a reliability of 12%. In this way, lesion candidates can be easily identified by the user. The specific area can be automatically detected by the lesion candidate detector 12, but the doctor determines the specific area, and the lesion candidate verifier 14 verifies the specific area determined by the doctor to determine whether the specific area is a lesion. Alternatively, it can be determined whether it is negative and displayed.

However, by verifying the region identified by the lesion candidate detector 12 or the doctor, it is possible to display whether the region is a lesion or a negative region, but it is not possible to grasp the shape of the entire lesion.

Specifically, the doctor determines a specific rectangular area in the examination image, and displays the result of verifying the lesion candidate in the specific area numerically. In other words, you can only see the results within a specific area determined by the doctor. Therefore, it was difficult to clearly display the boundary line around the lesion, and it was not possible to grasp the shape of the entire lesion.

The present invention has been made in view of these points, and an object of the present invention is to provide an image processing device, an image processing method, and an image processing program that clearly display the position and shape of a lesion site in a medical image. do.

In the present invention, in order to solve the above problem, in an image processing device that detects and displays a specific lesion from an input medical image, a determination range specifying means for specifying a determination range and the presence or absence of a specific lesion in the determination range are determined. An image processing characterized by comprising a determination means for determining, a determination range moving means for moving the determination range, and a highlighting means for highlighting and displaying the determination range according to the result of determination by the determination means. Equipment is provided.

As a result, the determination range specifying means specifies the determination range, the determination means determines the presence or absence of the specific lesion in the determination range, the determination range moving means moves the determination range, and the highlighting means is the determination means. The judgment range is highlighted and displayed according to the judgment result.

Further, in the present invention, in the image processing method of detecting and displaying a specific lesion from the input medical image, the determination range specifying means steps to specify the determination range, and the determination means determines the specific lesion in the determination range. It includes a step of determining the presence / absence, a step of moving the determination range by the determination range moving means, and a step of highlighting and displaying the determination range according to the result of the determination by the highlighting means. A video processing method characterized by this is provided.

As a result, the determination range specifying means specifies the determination range, the determination means determines the presence or absence of the specific lesion in the determination range, the determination range moving means moves the determination range, and the highlighting means is the determination means. The judgment range is highlighted and displayed according to the judgment result.

Further, in the present invention, in an image processing program that detects and displays a specific lesion from an input medical image, a computer is used as a determination range specifying means for specifying a determination range, and a determination for determining the presence or absence of a specific lesion in the determination range. Provided is a video processing program characterized by functioning as a means, a determination range moving means for moving the determination range, and a highlighting means for highlighting and displaying the determination range according to a result determined by the determination means. ..

As a result, the determination range specifying means specifies the determination range, the determination means determines the presence or absence of the specific lesion in the determination range, the determination range moving means moves the determination range, and the highlighting means is the determination means. The judgment range is highlighted and displayed according to the judgment result.

According to the image processing apparatus, the image processing method, and the image processing program of the present invention, the determination range specifying means specifies the determination range, the determination means determines the presence or absence of a specific lesion in the determination range, and the determination range moving means. However, since the determination range is moved and the highlighting means highlights and displays the determination range according to the result of the determination by the determination means, the position and shape of the specific lesion in the medical image can be clearly displayed. ..

It is a figure which shows the concept of the image processing apparatus which concerns on this embodiment. It is a figure which shows the detail of the determination means. It is a figure which shows the detail of the determination processing part. It is a figure which shows the detail of the image processing means. The determination means provided in the video processing apparatus of this embodiment is a flowchart showing processing until learning is performed. It is a flowchart which shows the processing until the image processing means provided in the image processing apparatus of this embodiment process an image, and the processed image is displayed by a display apparatus. It is a flowchart which shows the process from inputting instruction information to performing each process in the image processing means provided in the image processing apparatus of this embodiment. The determination means provided in the video processing apparatus of this embodiment is a flowchart showing processing until learning. It is a flowchart which shows the process until the determination means provided in the image processing apparatus of this embodiment verify the learned learning data. It is a flowchart which shows the process until the determination means provided in the image processing apparatus of this embodiment output a determination result. It is a flowchart which shows the processing until the image processing means provided in the image processing apparatus of this embodiment performs image processing, and outputs the processed image. It is a flowchart which shows the process until the image processing means provided in the image processing apparatus of this embodiment performs a determination exclusion process. It is a flowchart which shows the process until the image processing means provided with the image processing apparatus of this embodiment changes a color to the determination medical image which performed the determination exclusion process. It is a mammography image which shows the processing performed by the judgment exclusion processing unit. It is a figure which shows the neural network used in deep learning. It is a figure which shows the conventional learning method which makes a determination means learn, and the learning method in the image processing apparatus which concerns on this embodiment. The correct diagnosis rate of machine learning using training data and test data processed by the learning information adjustment department for the amount of information in the learning medical video, and machine learning using training data and test data that the learning information adjustment department does not process the amount of information in the learning medical video. It is a figure which shows the correct diagnosis rate of. It is a figure which shows the moving method of the determination frame. It is a figure which shows the state which the color change part changed the color of the determination range in the determination frame. It is a figure which shows the state which clarified the site of a malignant lesion in the judgment medical image. It is a block diagram which shows an example of the conventional lesion detection apparatus. It is a figure which shows the detection image which detected the lesion candidate from the inspection image using the lesion detection apparatus.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a diagram showing a concept of a video processing device according to the present embodiment.
The image processing device 100 includes a determination unit 110 and an image processing unit 120.

The determination means 110 uses, for example, a machine learning technique called deep learning using a neural network that imitates the function of the brain, and is used for image recognition, voice recognition, language processing, and the like.

The determination means 110 was made to learn a benign medical image and a malignant medical image which are learning medical images in advance, and later, the determination means 110 was made to recognize the determination medical image for determining the presence or absence of a malignant lesion. Judgment It is possible to judge whether the medical image is benign or malignant.

The image processing means 120 is connected to the determination means 110, and by inputting the determination medical image taken by the medical image device 200 connected from the outside into the image processing means 120, the image processing means 120 inputs the determination medical image. Let the determination means 110 make a determination. From the result of the determination by the determination means 110, the image processing means 120 performs image processing for clarifying the position and shape of the malignant lesion.

Specifically, the image processing means 120 cuts out a part of the determination medical image taken by the medical image device 200, and the determination means 110 determines whether it is benign or malignant. The image processing means 120 receives the result determined by the determination means 110, and for example, changes the color of the corresponding portion only when the result is malignant to highlight the relevant portion in the determination medical image. As a result, the color of only the malignant part in the judgment medical image is changed, so that the position and shape of the malignant lesion can be clarified.

The medical image device 200 is a medical image imaged by, for example, an X-ray device, a computer tomography device, a nuclear magnetic resonance imaging device, an ultrasonic inspection device, or the like. The image processing device 100 processes the still images and moving images taken by the medical image device 200, and outputs an image in which the position and shape of the lesion are clarified to the display device 300 connected to the image processing device 100. be able to.

Examples of the display device 300 include a display that displays information given by being connected to a personal computer, a printer that prints and displays the input information, and a developing machine that develops photographs and the like.

Since the position and shape of the lesion are clearly displayed in the judgment medical image output via the image processing device 100, it is possible to alert the doctor not to overlook when viewing the judgment medical image. Accurate diagnosis rate increases.
By inputting an instruction value by an externally connected device connected to the image processing means 120, it is possible to instruct the start of determination of the input image, the resolution of the output image, and the like.

FIG. 2 is a diagram showing details of the determination means.
As shown in FIG. 2, the determination means 110 includes a determination processing unit 111 and a learning data storage unit 112.

The determination processing unit 111 causes the determination processing unit 111 to learn by inputting the learning medical image in advance, and the learned data is stored in the learning data storage unit 112. The learning is performed using training data called a training data set, which is information for training the determination processing unit 111, and test data called a test data set for verifying the learned data.

Further, a part of the determination medical image cut and input by the image processing means 120 uses the learning data stored in the learning data storage unit 112 to determine whether the determination processing unit 111 is malignant or benign. After performing image processing for clarifying the malignant lesion by the image processing means 120, the determination medical image in which the malignant lesion is clarified is displayed on the display device 300.

FIG. 3 is a diagram showing details of the determination processing unit.
As shown in FIG. 3, the determination processing unit 111 includes a learning information adjusting unit 1110, a learning unit 1111, and a determination unit 1112.

The learning information adjusting unit 1110 is for adjusting the amount of information of the learning medical image input from the outside in order to make the learning unit 1111 learn efficiently. The input learning medical image is a medical image having a lesion divided in a grid pattern in advance, and is labeled so that it can be identified as benign or malignant. In addition, these labeled learning medical images are divided into test data and training data, respectively.

Since the learning medical images classified into benign and malignant have a bias in the amount of information between benign and malignant, the learning information adjustment unit 1110 adjusts the balance of the learning medical images having different ratios. 1111 can be trained correctly.

First, the learning information adjustment unit 1110 divides the benign training data into a plurality of groups. One of the divided benign training data is divided by the learning information adjustment unit 1110 so as to be equal to the amount of malignant training data.

While exchanging the divided benign training data for one malignant training data, the learning information adjusting unit 1110 is set to the learning unit 1111 so that the amount of malignant training data and the amount of benign training data are equal to each other. To learn.

As a result, malignant training data with a relatively small amount of data can be trained in correspondence with benign training data with a large amount of data, so the balance between benign and malignant can be achieved without reducing the overall amount of learning. It is possible to make the learning unit 1111 learn a large amount of training data while taking the above.

The learning unit 1111 is for actually learning the learning medical image adjusted by the learning information adjusting unit 1110, and is connected to the learning data storage unit 112 and the determination unit 1112.
As a specific learning example, the learning medical image adjusted by the learning information adjustment unit 1110 is input to the learning unit 1111, and the input learning medical image is the input learning by deep learning using a convolutional neural network. The learning unit 1111 learns by extracting the features of the medical image. The learned learning data is stored in the learning data storage unit 112.

The determination unit 1112 uses the learning data stored in the learning data storage unit 112 to determine whether a part of the determination medical image is benign or malignant by using a part of the determination medical image cut out by the image processing means 120. The determination is made, and the result of the determination by the determination unit 1112 is transmitted to the video processing means 120.

FIG. 4 is a diagram showing details of the video processing means.
As shown in FIG. 4, the image processing means 120 includes a determination exclusion processing unit 121, a determination range setting unit 122, a determination range moving unit 123, and a color changing unit 124.

The determination exclusion processing unit 121 divides the determination medical image input by the medical image device 200 into a grid pattern, and removes a portion of the divided determination medical images other than the determination target.

Specifically, the determination medical image taken by the medical image device 200 includes an image including a portion other than the determination target. For example, in an image taken by a mammography examination used in an examination such as breast cancer, the breast is not always photographed in the entire image, and in fact, a part other than the breast is also reflected.

By removing the portion other than the breast as a target of determination, the amount of information determined by the determination processing unit 111 can be reduced. As a result, the determination time of the entire determination medical image determined by the determination processing unit 111 can be shortened.

The judgment range setting unit 122 cuts out a part of the judgment medical video for which the judgment exclusion processing has been performed by the judgment exclusion processing unit 121, and determines a judgment range for the judgment processing unit 111 to make a judgment, which is not shown. It is for setting F.

The determination range moving unit 123 is for moving the determination frame F set by the determination range setting unit 122 in the determination medical image. As a result, the judgment processing unit 111 can judge whether it is malignant or benign on the locus of the moved judgment frame F, and the judgment frame F moves the entire judgment medical image to perform the judgment processing of the entire judgment medical image. be able to.

In this determination frame F, the determination range setting unit 122 increases the determination range in the determination frame F, so that the movement time of the determination frame F for moving the entire determination medical image is shortened, and the determination processing unit 111 makes a determination. The judgment time for the entire medical image is shortened.

Further, by reducing the determination range in the determination frame F by the determination range setting unit 122, it is possible to determine a fine lesion portion in the determination medical image, so that the degree of clarity of the malignant lesion is increased. The determination range in the determination frame F to be set by the determination range setting unit 122 can be arbitrarily set by the connected externally connected device.

Further, in the determination frame F, the movement value of the determination frame F to be moved by the determination range moving unit 123 can be arbitrarily set, but as with the size of the determination range, if the movement value of the determination frame F is increased, the determination is made. The movement time of the determination frame F for moving the entire medical image is shortened, and the determination time of the entire determination medical image determined by the determination processing unit 111 is shortened.
Further, by reducing the movement value of the determination frame F, it is possible to determine a fine lesion portion in the determination medical image, so that the degree of clarity of the malignant lesion is increased.

The color changing unit 124 is for changing the color of the determination range according to the result of the determination means 110 determining the determination range in the determination frame F.
Specifically, the determination means 110 determines whether it is benign or malignant in the determination range within the determination frame F, and transmits the determination result to the color changing unit 124.

The color changing unit 124 that has received the determination result changes the color of the determined determination range to red only when it is malignant. As a result, the color of only the malignant lesion portion is changed in the determination range, so that the site and shape of the malignant lesion are clearly displayed.

Further, a transparent color can be used for the color change performed by the color changing unit 124. For example, when the transparent red color is used for the color change performed by the color changing unit 124, the determination range moving unit 123 superimposes the transparent red color on the malignant lesion portion while moving the determination frame F, so that the malignant lesion The transparent red color overlaps many times in the part.

The portion where the transparent red color overlaps and is displayed in deep red is the portion strongly determined by the determination means 110 to be malignant, and it can be said that the entire portion displayed in dark red is the shape of the malignant lesion. This makes it possible to clearly display the site and shape of the malignant lesion.

FIG. 5 is a flowchart showing a process until the determination means provided in the video processing device of the present embodiment performs learning. Hereinafter, the process shown in FIG. 5 will be described along with the step numbers.
[Step S11] The learning medical image is input to the determination means 110. Specifically, the learning medical video is divided into a grid pattern in advance, labeled as benign and malignant, and then divided into test data and training data. The learning information adjustment unit It is input to 1110.

[Step S12] The determination means 110 adjusts the amount of information in the learning medical image. Specifically, among the learning medical images input in step S11, the benign training data is divided into a plurality of groups, and the learning unit 1111 is made to learn while exchanging the divided training data groups, and corresponding to the division. In order for the learning unit 1111 to repeatedly learn the malignant training data, the learning information adjustment unit 1110 adjusts the amount of information in the learning medical image.

[Step S13] The determination means 110 performs learning. Specifically, the learning unit 1111 learns by deep learning using a convolutional neural network based on the training data and the test data adjusted in step S12.
The determination means 110 learned in this way can determine whether the determination medical image is benign or malignant based on the learned learning data.

FIG. 6 is a flowchart showing a process in which the image processing means included in the image processing device of the present embodiment processes the image and the processed image is displayed by the display device. Hereinafter, the process shown in FIG. 6 will be described along with the step numbers.

[Step S21] The determination medical image is input to the image processing means 120. Specifically, a determination medical image for determining the presence or absence of a lesion imaged by the medical image device 200 is input to the image processing means 120.

[Step S22] The image processing means 120 processes the input determination medical image. Specifically, the image processing means 120 performs image processing for clarifying the site and shape of the malignant lesion with respect to the determination medical image input in step S21.

[Step S23] The image processing means 120 outputs the image-processed determination medical image. Specifically, the image processing means 120 outputs the determination medical image in which the image is processed in step S22 and the site and shape of the malignant lesion are clarified to the display device 300.
[Step S24] The display device 300 displays the image-processed determination medical image. Specifically, the display device 300 displays the determination medical image output in step S23.

FIG. 7 is a flowchart showing a process from the input of instruction information to the process of performing each process in the video processing means included in the video processing device of the present embodiment. Hereinafter, the process shown in FIG. 7 will be described along with the step numbers.

[Step S31] Instruction information is input to the video processing means 120. Specifically, an instruction for determining a determination range, setting a movement value, or starting video processing is input by an externally connected device connected from the outside of the video processing device 100.

[Step S32] The video processing means 120 determines the input instruction information. Specifically, the video processing means 120 determines whether the instruction information input to the video processing apparatus 100 is an instruction for setting a determination range, an instruction for setting a movement value, or an instruction for starting video processing.

When it is determined that the judgment range setting instruction has been input, the process proceeds to step S33, and when it is determined that the movement value setting instruction has been input, the process proceeds to step S34 and the video processing start instruction is input. If it is determined that the processing has been performed, the process proceeds to step S35.

[Step S33] The video processing means 120 sets a determination range. Specifically, the determination range setting unit 122 sets the determination range in the determination frame F for cutting out and determining a part of the determination medical image input in step S21.
[Step S34] The video processing means 120 sets a movement value. Specifically, the determination range moving unit 123 sets the movement value at which the determination frame F set in step S33 moves.

[Step S35] The video processing means 120 starts video processing. Specifically, while the determination range moving unit 123 moves the determination frame F set by the determination range setting unit 122 in the determination medical image, the determination means 110 is used to determine the malignant lesion, and the determination result is used as the basis for the determination. The image processing means 120 starts the process of changing the color of the determination range by the color changing unit 124.

FIG. 8 is a flowchart showing a process until the determination means included in the video processing apparatus of the present embodiment learns. Hereinafter, the process shown in FIG. 8 will be described along with the step numbers.
[Step S41] The learning medical image is input to the determination means 110. Specifically, in order for the determination means 110 to learn by deep learning using a convolutional neural network, the training data whose amount of information is adjusted in step S12 is input to the learning unit 1111.

[Step S42] The learning unit 1111 extracts the features of the input learning medical image. Specifically, the training data input in step S41 is extracted from the characteristics of the training data by deep learning using a convolutional neural network, and the learning unit 1111 identifies what is drawn from the characteristics. ..

[Step S43] The learning unit 1111 determines the input learning medical image. Specifically, the learning unit 1111 determines whether the characteristics of the training data extracted in step S42 are benign or malignant.

[Step S44] The learning unit 1111 compares the determination result with the label. Specifically, the determination result determined in step S43 is compared with the content of the label attached to the training data.

[Step S45] The learning unit 1111 corrects the learning data to be stored. Specifically, in order to reduce the difference between the determination result of the characteristics of the training data determined in step S43 and the label attached to the training data compared in step S44, the learning data to be stored is corrected. ..

[Step S46] The learning unit 1111 determines whether the learning process has been repeated up to a predetermined number. Specifically, the learning unit 1111 determines whether the processes from step S42 to step S45 are repeated a predetermined number of times.

The predetermined number of times can be adjusted according to the number of training data and the image size. Specifically, the processing from step S42 to step S45 is set at 5000 times, 100,000 times, etc., and the learning unit 1111 is made to learn while repeating up to the set number of times, thereby increasing the correct judgment rate. ..

When it is determined that the learning unit 1111 has repeated the processes from step S42 to step S45 a predetermined number of times, the process proceeds to step S47, and it is determined that the learning unit 1111 has not repeated the processes from step S42 to step S45 a predetermined number of times. When the case, the process proceeds to step S42.

[Step S47] The learning unit 1111 stores the learned learning data. Specifically, the learning unit 1111 stores the learning data learned by repeating the processes from step S42 to step S45 a predetermined number of times in the learning data storage unit 112.

FIG. 9 is a flowchart showing a process until the determination means provided in the video processing apparatus of the present embodiment verifies the learned learning data. Hereinafter, the process shown in FIG. 9 will be described along with the step numbers.

[Step S51] The learning medical image is input to the determination means 110. Specifically, in order to repeat the processes from step S41 to step S47 and verify the learning data learned by the determination means 110, the test data in which the benign or malignant label is not visible from the determination means 110. Is input to the learning unit 1111.

[Step S52] The learning unit 1111 extracts the features of the input learning medical image. Specifically, a feature is extracted from the test data input in step S51, and the learning unit 1111 identifies what is drawn from the feature.

[Step S53] The determination means 110 predicts the input learning medical image. Specifically, the learning unit 1111 compares the features extracted in step S52 with the learning data stored in the learning data storage unit 112, and predicts whether the test data input in step S52 is benign or malignant. ..

[Step S54] The determination means 110 evaluates the prediction result. Specifically, the prediction made by the learning unit 1111 in step S53 is compared with the label attached to the test data, and the learning unit 1111 performs the evaluation. For example, if the value of the prediction and the label match, a positive evaluation is given, and if the value of the prediction and the label do not match, a negative evaluation is given.

FIG. 10 is a flowchart showing a process until the determination means provided in the video processing apparatus of the present embodiment outputs a determination result. Hereinafter, the process shown in FIG. 10 will be described along with the step numbers.

[Step S61] Information on the determination range is input to the determination means 110. Specifically, a part of the determination medical image cut out in the determination range in the determination frame F set by the image processing means 120 in step S33 is input to the determination unit 1112.

[Step S62] The determination means 110 extracts the characteristics of the input determination medical image. Specifically, a part of the feature of the determination medical image input in step S61 is extracted, and what is drawn from the feature is identified.

[Step S63] The determination means 110 determines the input determination medical image. Specifically, a part of the characteristics of the judgment medical image extracted in step S62 is compared with the learning data stored in the learning data storage unit 112, and whether a part of the judgment medical image input in step S61 is benign. The determination unit 1112 determines whether it is malignant.

[Step S64] The determination means 110 outputs the determination result. Specifically, the determination unit 1112 outputs the result of determining a part of the determination medical image in step S63 to the image processing means 120.

FIG. 11 is a flowchart showing processing until the video processing means included in the video processing apparatus of the present embodiment performs video processing and outputs the processed video. Hereinafter, the process shown in FIG. 11 will be described along with the step numbers.

[Step S71] The determination medical image is input to the image processing means 120. Specifically, the information of the determination medical image taken by the medical image device 200 is input to the image processing means 120.
[Step S72] The image processing means 120 performs a determination exclusion process of the input determination medical image. Specifically, the determination exclusion processing unit 121 excludes a portion of the determination medical image input in step S71 other than the determination target.

[Step S73] The video processing means 120 performs a determination process. Specifically, while the judgment range moving unit 123 moves the judgment range in the judgment frame F set by the judgment range setting unit 122 in the judgment medical image, the judgment unit 1112 cuts out a part of the judgment medical image. judge.

[Step S74] The image processing means 120 performs a color change process. Specifically, the color changing unit 124 changes the color according to the result of the determination in step S73. More specifically, the color changing unit 124 changes the color in the determination range only when the result of the determination in step S73 is malignant.

[Step S75] The video processing means 120 outputs the processed video. Specifically, the color changing unit 124 outputs the determination medical image processed in steps S71 to S74 to the display device 300.

FIG. 12 is a flowchart showing a process until the image processing means included in the image processing apparatus of the present embodiment performs the determination exclusion process. Hereinafter, the process shown in FIG. 12 will be described along with the step numbers.

[Step S81] The determination medical image is input to the image processing means 120. Specifically, the information of the determination medical image taken by the medical image device 200 is input to the image processing means 120.
[Step S82] The image processing means 120 divides the determination medical image. Specifically, the judgment exclusion processing unit 121 divides the judgment medical image input in step S81 into a fine grid pattern, for example, the input judgment medical image into 80 grids of 10 vertical × 8 horizontal.

[Step S83] The video processing means 120 selects the first video to be subjected to the determination exclusion process. Specifically, among the determination medical images divided into fine grids in step S82, the determination exclusion processing unit 121 selects the image for which the determination exclusion process is first performed.

[Step S84] The image processing means 120 determines whether or not the determination target image is included in a part of the determination medical image. Specifically, the determination exclusion processing unit 121 determines whether or not a determination target such as a part of the human body is included in a part of the determination medical image selected in step S83.

If it is determined that the judgment target is included in a part of the judgment medical video, the process proceeds to step S85, and if it is determined that the judgment target is not included in a part of the judgment medical video, the process is stepped. Proceed to S86.

[Step S85] The image processing means 120 determines whether the determination exclusion process has been performed until the end of the determination medical image. Specifically, it is determined whether or not a part of the determination medical image determined in step S84 is the last divided image among the determination medical images divided into a plurality of parts.

When it is determined that a part of the judgment medical video is the last divided video of the judgment medical video divided into a plurality of parts, the processing is terminated and a part of the judgment medical video is divided into a plurality of judgment medical images. If it is determined that the video is not the last divided video, the process proceeds to step S87.

[Step S86] The video processing means 120 performs a determination exclusion process. Specifically, as a result of the determination by the determination exclusion processing unit 121 in step S84, if the determination target image is not included in a part of the determination medical image, the determination means 110 needs to determine a part of the determination medical image. Therefore, the determination exclusion processing unit 121 sets a determination exclusion flag for the determination means 110 not to determine.

[Step S87] The video processing means 120 determines the next video to be subjected to the determination exclusion process. Specifically, among the judgment medical images divided into fine grids in step S82, the judgment exclusion processing unit 121 selects a part of the judgment medical images to be subjected to the judgment exclusion processing next.

FIG. 13 is a flowchart showing a process until the image processing means included in the image processing apparatus of the present embodiment changes the color of the determination medical image to which the determination exclusion process has been performed. Hereinafter, the process shown in FIG. 13 will be described along with the step numbers.

[Step S91] A determination medical image obtained by performing a determination exclusion process on the image processing means 120 is generated. Specifically, the judgment exclusion processing unit 121 performs the judgment exclusion processing on the judgment medical image input to the image processing means 120, and the judgment medical image after the judgment exclusion processing is generated.

[Step S92] The video processing means 120 initializes. Specifically, the determination range moving unit 123 moves the determination frame F to an initial position when the determination frame F moves while making a determination, for example, a position such as the upper leftmost stage of the determination medical image.

[Step S93] The video processing means 120 determines whether or not the determination range in the determination frame F is a determination target. Specifically, the determination range moving unit 123 determines whether or not the determination range is the determination target depending on whether or not the determination exclusion flag is set in the divided video to which the determination frame F is moved.

More specifically, when the judgment frame F moves on the divided video in which the judgment exclusion flag is set, the judgment range moving unit 123 determines that the judgment is excluded, and on the divided video in which the judgment exclusion flag is not set. When the determination frame F moves to, the determination range moving unit 123 determines that it is a determination target.

Further, when the determination frame F moves on the joint portion between the divided video in which the determination exclusion flag is set and the divided video in which the determination exclusion flag is not set, the determination range moving unit 123 determines that the determination target is the determination target. ..

When it is determined that the determination range in the determination frame F is the determination target, the process proceeds to step S94, and when it is determined that the determination range in the determination frame F is not the determination target, the process proceeds to step S97.

[Step S94] The image processing means 120 determines whether or not the determination range in the determination frame F is a malignant lesion. Specifically, the determination range moving unit 123 transmits the determination range in the determination frame F moved by the determination range moving unit 123 to the determination means 110 to make a determination.
When it is determined that the determination range in the determination frame F is malignant, the process proceeds to step S95, and when it is determined that the determination range in the determination frame F is benign, the process proceeds to step S96.

[Step S95] The image processing means 120 changes colors. Specifically, the color changing unit 124 changes the color of the determination range in the determination frame F determined to be malignant in step S94.

[Step S96] The image processing means 120 determines whether or not the determination has been made to the end of the determination medical image. Specifically, the judgment range moving unit 123 moves the judgment frame F from the initial position moved by the judgment range moving unit 123 in step S92, and all of the judgment medical images except the range that is not the judgment target are displayed. Judge whether it was judged.

For example, when the judgment is made while the uppermost position on the left side of the judgment medical image is set as the initial position and the judgment frame F is moved to the right, and the judgment frame F is moved from the upper part to the lower part of the judgment medical image in order, the judgment is made. When the frame F reaches the bottom on the right side, it is determined that the determination is completed.

When it is determined that the determination range moving unit 123 moves the determination frame F to the end and makes a judgment, the process is terminated and the judgment range moving unit 123 does not move the judgment frame F to the end to make a judgment. When the determination is made, the process proceeds to step S97.

[Step S97] The video processing means 120 moves the determination frame F to the next position. Specifically, the determination range moving unit 123 moves the determination frame F to a position where the determination frame F has not been moved and the determination has not been completed.

In the video processing apparatus 100 of the present embodiment, the judgment exclusion processing unit 121 has performed the judgment exclusion processing in step S91, and the judgment range moving unit 123 has the judgment range in the judgment frame F in step S93. An example is shown in which the determination means 110 determines whether or not the lesion is a determination target and whether or not the determination range is a malignant lesion in step S94, but the process in step S91 is omitted. The judgment range moving unit 123 judges whether or not the judgment range in the judgment frame F is the judgment target based on the input judgment medical image, and determines whether or not the judgment range is a malignant lesion in the case of judgment coping. 110 may determine.

FIG. 14 is a mammography image showing the processing performed by the determination exclusion processing unit.
As shown in FIG. 14A, the judgment exclusion processing unit 121 divides the judgment medical image into a fine grid pattern in order to set the judgment exclusion flag in the range not including the judgment target in the judgment medical image.

As shown in FIG. 14B, the judgment exclusion processing unit 121 determines whether or not the judgment target is included in a part of the divided judgment medical images, and sets the judgment exclusion flag in the portion not including the judgment target. By standing, it is excluded from the determination process performed by the determination unit 1112.

Specifically, images other than the mammary gland in FIG. 14 (A) are excluded. Since the portion displayed in black in FIG. 14B is a range not including the mammary gland, it is excluded from the determination target and the determination exclusion flag is set.

In the case of the example of FIG. 14B, it can be seen that the range of the determination target becomes less than half of the whole when the determination exclusion process is performed. Since the range of the determination target is limited by this, the amount of information of the determination process performed by the determination unit 1112 in the entire determination medical image is reduced, so that the processing speed until the determination process is completed can be improved.

Further, in FIG. 14, an example is shown in which the judgment exclusion processing unit 121 divides the judgment medical image and excludes items other than the judgment target in order to perform the judgment exclusion processing, but the judgment exclusion processing performed by the judgment exclusion processing unit 121 is a judgment. Not only those that process medical images but also learning medical images may be processed.

For example, when the learning medical video is input to the judgment exclusion processing unit 121, the learning medical video is divided by the judgment exclusion processing unit 121 as shown in FIG. 14 (A), and the divided video is divided as shown in FIG. 14 (B). By excluding the image other than the mammary gland by the determination exclusion processing unit 121, the process of leaving only the divided image including the determination target is performed.

The learning unit possessed by the determination means 110 by labeling the divided images left as these determination targets in order to distinguish between the images in which the malignant lesions are visualized and the images in which the malignant lesions are not visualized. It can be used as a learning medical image for 1111 to learn.

Further, the learning medical image processed by the determination exclusion processing unit 121 is recognized by the learning unit 1111 as a different image by inverting the image left and right, and the amount of information of the learning medical image input to the learning unit 1111. Can also be increased.

FIG. 15 is a diagram showing a neural network used in deep learning.
As shown in FIG. 15, in the neural network, the entire processing can be divided into an input layer, an intermediate layer, and an output layer.

The input layer is a place for inputting an image, sound, or the like, and is used in the present invention to input image data of a medical image finely divided in a grid pattern shown on the left side of FIG.
The intermediate layer is a place for extracting various characteristic shapes and patterns from the input data, and is used in the present invention to extract shapes and patterns from the medical image input in the input layer.

The output layer is a place to identify the input data from the features extracted in the intermediate layer, and is used in the present invention to identify whether the input medical image is benign or malignant.
As a general form, in a neural network, a convolution layer that plays a role in extracting features of input data and a pooling layer that gives invariance to the position by extracting features that strongly appear from the features extracted by the convolution layer. Is used as a set and is a fully connected layer used for the final output layer, which is classified based on the extracted characteristics.

It is also possible to insert a plurality of intermediate layers to make it complicated. In the image processing apparatus 100 according to the present embodiment, deep learning was performed using a neural network in which the convolution layer and the pooling layer were combined into 10 layers. As a result, for example, the characteristics of breast cancer can be extracted from a mammography image, and it can be identified that the cancer is present with a probability of 70%, the cancer is not present with a probability of 30%, and the like.

In the learning by deep learning performed by the learning unit 1111 according to the present embodiment, a neural network having 10 layers including the convolutional layer and the pooling layer was used, but the number of units of the convolutional layer and the pooling layer in the intermediate layer was used. Can be made to learn by the learning unit 1111 while making appropriate adjustments. Further, the learning unit 1111 can be made to learn while appropriately adjusting the input layer and the output layer.

FIG. 16 is a diagram showing a conventional learning method in which a determination means is trained and a learning method in a video processing device according to the present embodiment.
As shown in FIG. 16A, the conventional machine learning learning method divides a medical image into a grid pattern and distinguishes between an image in which a malignant lesion is visualized and an image in which a malignant lesion is not visualized. Therefore, it was labeled and divided into training data and test data, and machine learning was performed using these training data and test data.

However, when the medical images were divided, the number of benign images was overwhelmingly large, and there was a large difference in the learning data between benign and malignant. In reality, nearly 90% of the images are benign images, and correct machine learning could not be performed due to this biased learning data.

For example, if malignant video is 10% and benign video is 90%, even if the test is performed without learning at all, if all are diagnosed as benign, the correct answer rate will be 90%, so such a learning medical video is correct. It was difficult to learn.

As one of these solutions, it is conceivable to increase the number of malignant images according to the benign images, but it is very difficult to collect malignant images with nearly 90% of benign images, which is not realistic. I didn't.

It is also possible to balance by training benign images with a relatively large amount of data in accordance with malignant images with a relatively small amount of data, but this is efficient because the overall amount of learning is significantly reduced. It was not a typical learning.

Therefore, as shown in FIG. 16B, the learning information adjusting unit 1110 included in the image processing apparatus 100 according to the present embodiment adjusts the amount of information of the learning medical image, so that the benign image and the malignant image can be obtained. You will be able to learn correctly even if the ratio is different.

Specifically, first, 47,282 divided medical images are prepared by combining the image in which the malignant lesion is visualized and the image in which the malignant lesion is not visualized, and the divided medical images are labeled as benign and malignant. After attaching, divide each into test data and training data.

Of the divided test data and training data, the training data in the benign video with a relatively large amount of data is divided into multiple groups such as training data 1, training data 2, training data 3, and so on. The learning information adjustment unit 1110 performs the process of dividing.

Next, the learning information adjustment unit 1110 creates training data in which the training data in the benign video is replaced with training data 1, training data 2, training data 3, and so on, and is malignant so as to correspond to the training data. The video is also duplicated by the learning information adjusting unit 1110 and is made to learn by the learning unit 1111.

By learning the learning unit 1111 while repeatedly exchanging the training data in the benign, many learning medical images can be obtained in the learning unit 1111 while balancing benign and malignant without reducing the amount of information in the learning medical images. Can be made to learn.

In the learning method with the image processing apparatus 100 according to the present embodiment, the benign images were 27 times as many as the malignant images for 47,282 images including the images in which the malignant lesions were not visualized, so all the images were used. The learning information adjustment unit 1110 performed a process of exchanging the training data in the benign video 27 times, and the learning unit 1111 was repeatedly trained.

FIG. 17 shows the correct diagnosis rate of machine learning based on the training data and test data processed by the learning information adjustment unit for the amount of information in the learning medical image, and the training data and test in which the learning information adjustment unit does not process the amount of information in the learning medical image. It is a figure which shows the correct diagnosis rate of machine learning by data.

In FIG. 17A, the learning information adjustment unit 1110 processes the amount of information of all lesions and learning medical images other than the lesions in the mammography examination, and the learning medical images are learned by deep learning using a neural network. It is a figure which shows the result of having learned part 1111.

FIG. 17A shows the result of learning 47282 divided medical images. In addition, 788 test data and 3156 training data are prepared for the learning medical video, and the learning information adjustment unit 1110 replaces the benign data among the training data so that any video can be used 27 times. It is the result of learning.

As a result of learning, the correct diagnosis rate, which is the correct diagnosis rate, is 96.6%, the sensitivity, which is the rate of positive results among those who actually have the disease, is 93.9%, and among those who do not have the disease. The specificity, which is the percentage of negatives, is 99.2%, and the positive predictive value of PPV (Positive Predictive Value), which is the percentage of those who actually have the disease among those who test positive, is 99.2. The NPV (Negative Predictive Value), which is the negative predictive value of the percentage of those who did not actually suffer from the negative, was 94.2%.

In this way, the learning information adjustment unit 1110 processes the amount of information in the learning medical image to balance the malignant image and the benign image, and increase the amount of information in the learning medical image to be learned in the learning unit 1111. Since I was able to learn, I was able to obtain a high rate of correct diagnosis. In the image processing apparatus 100 according to the present embodiment, the correct diagnosis rate can be obtained by using the learning medical image processed by the learning information adjusting unit 1110.

In FIG. 17B, the learning medical image of the disorder of construction in the mammography examination is learned by the learning unit 1111 by deep learning using a neural network without being processed by the learning information adjusting unit 1110. It is a figure which shows the result of having made it.

FIG. 17B shows the result of learning 716 divided medical images. As a result of learning, the correct diagnosis rate was 93.1%, the sensitivity was 98.4%, the specificity was 61.9%, the PPV was 93.8%, and the NPV was 86.7%.

FIG. 17C is a deep learning using a neural network without processing the learning medical image of FAD (focal asymmetric density), which is a local asymmetric shadow in a mammography examination, without being processed by the learning information adjusting unit 1110. It is a figure which shows the result of having the learning unit 1111 learn the learning medical image.

FIG. 17C shows the result of learning 464 divided medical images. As a result of learning, the correct diagnosis rate was 91.3%, the sensitivity was 86.0%, the specificity was 88.9%, the PPV was 92.5%, and the NPV was 80.0%.

FIG. 17D shows a learning medical image of calcification in a mammography examination, which is learned by the learning unit 1111 by deep learning using a neural network without being processed by the learning information adjusting unit 1110. It is a figure which shows the result.

FIG. 17D shows the result of learning 2456 divided medical images. As a result of learning, the correct diagnosis rate was 85.6%, the sensitivity was 76.8%, the specificity was 92.2%, the PPV was 88.0%, and the NPV was 84.1%.

FIG. 17 (E) shows a learning medical image of a tumor in a mammography examination, which was learned by the learning unit 1111 by deep learning using a neural network without being processed by the learning information adjusting unit 1110. It is a figure which shows the result.

FIG. 17 (E) shows the result of learning 1816 divided medical images. As a result of learning, the correct diagnosis rate was 90.1%, the sensitivity was 95.8%, the specificity was 72.0%, the PPV was 90.0%, and the NPV was 86.7%.

In FIG. 17F, the learning medical image of all lesions in the mammography examination is learned by the learning unit 1111 by deep learning using a neural network without being processed by the learning information adjusting unit 1110. It is a figure which shows the result of having made it.

FIG. 17F shows the result of learning 3802 divided medical images. As a result of learning, the correct diagnosis rate was 89.5%, the sensitivity was 90.7%, the specificity was 88.1%, the PPV was 89.8%, and the NPV was 89.1%.

As can be seen from the above, as shown in FIG. 17A, the learning information adjustment unit 1110 processes the amount of information in the learning medical image, and learns a large amount of learning medical image in which the malignant image and the benign image are well-balanced. It can be seen that a high accuracy rate was obtained by learning by Department 1111.
Further, the learning unit 1111 can see that the correct diagnosis rate is improved as the amount of information of the learning medical image to be learned increases.

FIG. 18 is a diagram showing a method of moving the determination frame.
As shown in FIG. 18A, the determination frame F set by the determination range setting unit 122 causes the determination means 110 to determine the determination range in the determination frame F while moving in the determination medical image. Here, for convenience, the range excluded from the determination by the determination exclusion processing unit 121 is not displayed.

For example, when the initial position of the determination frame F is the uppermost position on the left side, the determination range moving unit 123 moves the determination frame F from left to right at the uppermost stage of the determination medical image, and the determination means 110 moves the determination range. 123 makes a judgment.

When the determination frame F reaches the right side of the determination medical image, the determination range moving unit 123 causes the determination means 110 to make a determination while moving the determination frame F from the left side to the right side at a position slightly lowered from the uppermost stage.

As shown in FIG. 18B, the determination range moving unit 123 moves the determination frame F so that the moving trajectories of the determination frames F overlap.
Specifically, the determination frame F makes a determination while the determination range moving unit 123 moves the height of F1 from left to right. When the determination frame F reaches the right side at the height of F1, the determination range moving unit 123 moves from left to right at the height of F2 so that the determination frame F next overlaps with the locus passed by F1. Make a judgment while letting it.

Similarly, when the determination frame F reaches the right side at the height of F2, the determination frame F is moved from left to right at the height of F3 so that the determination frame F overlaps with the locus passed by F2. The determination is made while the 123 is moving.

Based on the result of the determination by the determination unit 1112, the color change unit 124 changes the color, so that the color of the site of the malignant lesion is changed, and the position and shape of the malignant lesion can be clearly displayed. Details are shown below.

FIG. 19 is a diagram showing how the color changing unit changes the color of the determination range in the determination frame.
FIG. 19A shows that the determination range in the determination frame F is benign and the color changing unit 124 does not change the color.

As described above, as long as the determination range in the moving determination frame F is benign, the color of the determination medical image is not changed by the color changing unit 124, so that no change is seen in the determination medical image.

FIG. 19B shows that the determination range in the determination frame F is malignant and the color changing unit 124 changes the color.
In this way, when the determination range in the moving determination frame F is malignant, the color of the determination medical image is changed by the color changing unit 124, so that the malignant portion becomes clear. Further, by superimposing the semi-transparent color of the color change performed by the color changing unit 124 on the determination range, the colors of the overlapping portions are overlapped and the color is more emphasized.

By superimposing the translucent color of the color change performed by the color change unit 124 on the determination range in this way, even if it is an elongated malignant lesion or a lesion site having a complicated shape, which was difficult in the past, the figure is shown. The position and shape of the malignant lesion can be clearly displayed as in 19 (C).

In FIG. 19C, the boundary portion of the malignant lesion is displayed lightly because the determination ranges in the determination frame F whose color has been changed by the color change portion 124 do not overlap, and the central portion of the malignant lesion is determined to have changed color. It can be seen that the areas are displayed darkly because they overlap.

The size of the movement of the judgment frame F can be arbitrarily set, but as described above, if the movement value of the judgment frame F is large, the time for the judgment frame F to reach the end is shortened. The time to do this can be shortened.

Further, by reducing the size of the determination frame F and reducing the movement value of the determination frame F, fine determination can be made. Therefore, the display of the boundary portion of the lesion portion becomes clear, and the boundary portion around the lesion portion can be clearly displayed as shown in FIG. 20 (C).

FIG. 20 is a diagram showing a state in which the site of a malignant lesion is clarified in a judgment medical image.
As shown in FIG. 20, by changing the color of the site determined to be malignant by the determination unit 1112 by the color changing unit 124, the lesion site can be clarified as shown in the black circle. For this reason, the doctor can be alerted not to overlook when viewing the judgment medical image, and the accurate diagnosis rate of the doctor increases.

In the example shown in the present embodiment, a still image of a mammography inspection, which is a determination medical image taken by the medical image device 200, is taken as an example, but the image processing device 100 not only processes the still image but also medical treatment. The determination means 110 performs the determination processing on the moving image captured by the image device 200, and the image processing means 120 performs the image processing, so that the lesion site can be clarified even on the moving image.

Further, the image processing performed by the image processing apparatus 100 is not limited to the one imaged by using X-rays, ultrasonic waves, magnetic resonance, or the like, but if the determination means 110 learns the necessary learning data, endoscopy or the fundus of the eye can be obtained. The position and shape of the malignant lesion can be clarified by the image processing apparatus 100 performing image processing on the inspection image taken under visible light such as a camera.

Further, in the present embodiment, the position and shape of the malignant lesion are shown by the image processing performed by the image processing means 120 on the entire input determination medical image, but the image processing performed by the image processing means 120 in the determination medical image is performed. The range may be specified in advance, and the position and shape of the malignant lesion may be displayed within the specified range.

10 Lesion detector 11 Video input unit 12 Lesion candidate detector 13 Peripheral object detector 14 Lesion candidate verifier 15 Candidate remover 16 Display unit 20, 200 Medical imaging device 30 Correlation information 40 Lesion candidate detection image 41, 42 Lesion candidate area 43 Glandular tissue area 100 Image processing device 110 Judgment means 111 Judgment processing unit 112 Learning data storage unit 120 Video processing means 121 Judgment exclusion processing unit 122 Judgment range setting unit 123 Judgment range movement unit 124 Color change unit 300 Display device 1110 Learning information adjustment Part 1111 Learning part 1112 Judgment part A Specific area F Judgment frame

Claims (10)

In an image processing device that detects and displays a specific lesion from the input medical image
Judgment range specifying means to specify the judgment range and
A determination means for determining the presence or absence of a specific lesion in the determination range,
A highlighting means for displaying a color transparent color, which is a transparent color whose color has been changed by the color changing means, overlaid on the range determined by the determination range on the medical image according to the result of the determination by the determination means.
A determination range moving means for moving the determination range so as to overlap another determination range determined by the determination means on almost the entire area in one determination range determined by the determination means.
Equipped with
The process by the determination means, the process by the highlighting means, and the process by the determination range moving means are repeated so that the determination range scans on the medical image.
A video processing device characterized by. Judgment range changing means that can arbitrarily change the area of the judgment range,
The video processing apparatus according to claim 1, further comprising. A moving value changing means that can arbitrarily change the moving value of the determination range to be moved by the determination range moving means,
The video processing apparatus according to claim 1, further comprising. A determination exclusion means for excluding a portion other than the detection target portion from the medical image from the determination range performed by the determination means.
The video processing apparatus according to claim 1, further comprising. The determination means is
A learning means for determining the presence or absence of a specific lesion to be determined by the determination means,
The video processing apparatus according to claim 1, further comprising. The learning means is
The image processing apparatus according to claim 5, wherein the characteristics of a specific lesion are extracted from a medical image for learning by deep learning using a neural network. A lesion information adjusting means for equalizing the ratio of information with a specific lesion and information without a specific lesion in the medical image for learning.
6. The video processing apparatus according to claim 6. The lesion information adjusting means is
The image processing apparatus according to claim 7, wherein the information of no specific lesion divided into a plurality of pieces is replaced with the information of the presence of a specific lesion to adjust the ratio. In the video processing method that detects and displays a specific lesion from the input medical video,
The step in which the determination range specifying means specifies the determination range,
A step in which the determination means determines the presence or absence of a specific lesion within the determination range,
The highlighting means displays a color transparent color, which is a transparent color whose color has been changed by the color changing means, overlaid on the range determined by the determination range on the medical image according to the result determined by the determination means. Steps and
A step in which the determination range moving means moves the determination range so that the other determination ranges determined by the determination means overlap with almost the entire area in one determination range determined by the determination means.
Equipped with
The process by the determination means, the process by the highlighting means, and the process by the determination range moving means are repeated so that the determination range scans on the medical image.
A video processing method characterized by. In a video processing program that detects and displays a specific lesion from the input medical video
Computer,
Judgment range specifying means to specify the judgment range,
A determination means for determining the presence or absence of a specific lesion in the determination range,
A highlighting means for displaying a color transparent color, which is a transparent color whose color has been changed by the color changing means, overlaid on the range determined by the determination range on the medical image according to the result of the determination by the determination means.
A determination range moving means for moving the determination range so that the other determination ranges determined by the determination means are overlapped with almost the entire area in one determination range determined by the determination means.
To function as
The process by the determination means, the process by the highlighting means, and the process by the determination range moving means are repeated so that the determination range scans on the medical image.
A video processing program featuring.

Images