IMAGE DIAGNOSIS LEARNING DEVICE, IMAGE DIAGNOSIS DEVICE, IMAGE DIAGNOSIS METHOD, AND RECORDING MEDIUM FOR STORING PROGRAM

• US 20190311475A1
• Filed: 07/03/2017
• Published: 10/10/2019
• Est. Priority Date: 07/04/2016
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: HOSOI, Toshinori
• Granted Patent: US 10,803,582 B2
• Time in Patent Office: Days
• CPC Class Codes: A61B 1/000096 using artificial intelligence
• CPC Class Codes: A61B 1/045 Control thereof
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 7/00 Image analysis
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

An image diagnosis learning device includes: CNN configuration storage storing a network configuration of a convolutional neural network (CNN); parameter storage storing parameters of a learning subject in the CNN; inappropriate region detection unit that detects, an inappropriate region which is a region inappropriate for identification of an abnormal region where a diagnosis subject has a possibility of abnormality, in an image for learning in which the diagnosis subject is photographed; and inappropriate region invalidation unit invalidates a unit corresponding to the inappropriate region, among units of an input layer in the network configuration of the CNN to which the image for learning has been input. The image diagnosis learning device further includes loss value calculation unit performs calculation of the CNN by using the parameters in a state where the unit of the input layer, which corresponds to the inappropriate region, has been invalidated, and calculates a loss value; and parameter updating unit updates the parameters in the parameter storage.

Claims

1. An image diagnosis learning device comprising:

a CNN configuration storage that stores a network configuration of a convolutional neural network (CNN);

a parameter storage that stores parameters of a learning subject in the CNN;

a memory storing instructions; and

at least one of processor configured to process the instructions to;

detect, based on a predetermined criterion, an inappropriate region which is a region inappropriate for identification of an abnormal region where a diagnosis subject has a possibility of abnormality, in an image for learning in which the diagnosis subject is photographed;

invalidate a unit corresponding to the inappropriate region, among units of an input layer in the network configuration of the CNN to which the image for learning has been input;

perform calculation of the CNN by using the parameters in a state where the unit of the input layer, which corresponds to the inappropriate region, has been invalidated, and calculate a loss value based on a result of the calculation and information, the information indicating abnormality of the diagnosis subject and given to the image for learning in advance; and

update the parameters of the parameter storage, based on the loss value.

2. The image diagnosis learning device according to claim 1,

wherein the processor is further configured to process the instructions to generate the image for learning by executing a disturbance process on a learning sample of an image in which the diagnosis subject is photographed.

3. An image diagnosis device comprising:

a parameter storage the parameters of the CNN, which were updated by applying the image diagnosis learning device according to claim 1 to one or a plurality of images for learning;

a CNN configuration storage that stores a network configuration of the CNN used by the image diagnosis learning device when the parameters were updated; and

a memory storing instructions; and

at least one of processor configured to process the instructions to;

input information based on the diagnosis subject image, in which the diagnosis image is photographed, to the CNN, and perform calculation, thereby identifying an abnormal region where the diagnosis subject has a possibility of abnormality in the diagnosis subject image.

4. The image diagnosis device according to claim 3, further comprising:

wherein the processor is further configured to process the instructions to;

detect, based on a predetermined criterion, the inappropriate region in the diagnosis subject image; and

invalidate a unit corresponding to the inappropriate region, among units of an input layer in the CNN to which the diagnosis subject image has been input,

perform calculation of the CNN in a state in which the unit of the input layer, which corresponds to the inappropriate region, has been invalidated, thereby identifying the abnormal region in the diagnosis subject image.

5. The image diagnosis device according to claim 3,

wherein the processor is further configured to process the instructions to perform the identification by using the CNN with respect to each of partial regions included in the diagnosis subject image, thereby outputting an identification result for each partial region.

6. The image diagnosis device according to claim 3,

wherein the processor is further configured to process the instructions to:

input an entire area of the diagnosis subject image to the CNN, and

perform calculation of the CNN by using a configuration up to a layer from which a feature is extracted, in the network configuration of the CNN, thereby outputting a feature relating to the abnormal region in the diagnosis subject image.

7. A method in which a computer device executes:

detecting, based on a predetermined criterion, an inappropriate region which is a region inappropriate for identification of an abnormal region where a diagnosis subject has a possibility of abnormality, in an image for learning in which the diagnosis subject is photographed;

invalidating a unit corresponding to the inappropriate region, among units of an input layer in a network configuration of a convolutional neural network (CNN) to which the image for learning has been input;

performing calculation of the CNN in a state where the unit of the input layer, which corresponds to the inappropriate region, has been invalidated, and calculating a loss value based on a result of the calculation and information, the information indicating abnormality of the diagnosis subject and given to the image for learning in advance; and

updating parameters of a learning subject in the CNN, based on the loss value.

9. A method in which a computer device executes,

by using the parameters of the CNN, which were updated by executing the method according to claim 7 on one or a plurality of images for learning, and a network configuration of the CNN used when the parameters were updated,

inputting information based on the diagnosis subject image, in which the diagnosis image is photographed, to the CNN, and performing calculation, thereby identifying an abnormal region where the diagnosis subject has a possibility of abnormality in the diagnosis subject image.

8. A non-transitory computer readable recording medium for storing a program which causes a computer device to execute:

detecting, based on a predetermined criterion, an inappropriate region which is a region inappropriate for identification of an abnormal region where a diagnosis subject has a possibility of abnormality, in an image for learning in which the diagnosis subject is photographed;

invalidating a unit corresponding to the inappropriate region, among units of an input layer in a network configuration of a convolutional neural network (CNN) to which the image for learning has been input;

performing calculation of the CNN in a state where the unit of the input layer, which corresponds to the inappropriate region, has been invalidated, and calculating a loss value based on a result of the calculation and information, the information indicating abnormality of the diagnosis subject and given to the image for learning in advance; and

updating parameters of a learning subject in the CNN, based on the loss value.

10. A non-transitory computer readable recording medium for storing a program which causes a computer device to execute,

by using the parameters of the CNN, which were updated by causing the computer device to execute the program stored in the recording method according to claim 8 on one or a plurality of images for learning, and a network configuration of the CNN used when the parameters were updated,

inputting information based on the diagnosis subject image, in which the diagnosis image is photographed, to the CNN, and performing calculation, thereby identifying an abnormal region where the diagnosis subject has a possibility of abnormality in the diagnosis subject image.