Document creation support apparatus, method, and program

1. A document creation support apparatus comprising at least one processor,

wherein the processor is configured to

analyze an image to derive property information describing a property of a structure of interest included in the image,

input the property information into a learning model to generate a sentence related to the image, wherein the learning model is trained by a combination of a plurality of pieces of training property information derived from a plurality of training images and a plurality of medical sentences generated from the plurality of pieces of training property information,

analyze the sentence to specify a term representing the property related to the structure of interest included in the sentence,

collate the property information with the term, and

in response to the property information not being collated with the term, generate a parameter to indicate that the term is either excess information or deficiency information with respect to the property information and regenerate another sentence by using the learning model according to the parameter.

2. The document creation support apparatus according to claim 1 , wherein the processor is further configured to display the sentence on a display.

3. The document creation support apparatus according to claim 2 , wherein the processor is further configured to display a result of the collation on the display.

4. The document creation support apparatus according to claim 3 , wherein the processor is further configured to display the result of the collation by highlighting a different point in a case where the term and the property information are different from each other in the sentence.

5. The document creation support apparatus according to claim 1 , wherein the processor is further configured to regenerate the sentence in a case where the term and the property information are different from each other in the sentence.

6. The document creation support apparatus according to claim 1 , wherein the processor is further configured to receive correction of the sentence.

7. The document creation support apparatus according to claim 1 , wherein the processor is configured to

generate a plurality of sentences related to the image based on the property information,

analyze each of the plurality of sentences to specify a term representing the property related to the structure of interest included in each of the plurality of sentences,

collate the property information with the term for each of the plurality of sentences, and

select at least one sentence from the plurality of sentences based on a result of the collation.

8. The document creation support apparatus according to claim 1 , wherein the image is a medical image, and the sentence is a medical sentence related to the structure of interest included in the medical image.

9. The document creation support apparatus according to claim 1 , wherein the machine learning model is a recurrent neural network.

10. The document creation support apparatus according to claim 1 , wherein after a result of the collation is generated, the processor is further configured to

determine whether an automatic correction option is selected,

in response to the automatic correction option being selected, regenerate another sentence related to the image,

in response to the automatic correction option not being selected, determine whether a manual correction option is selected, and

in response to the manual correction option being selected, receive a manual correction to the sentence.

11. The document creation support apparatus according to claim 1 ,

wherein the machine learning model is a recurrent neural network constructed by learning an encoder and a decoder using the combination of the plurality of pieces of training property information and the plurality of medical sentences.

12. The document creation support apparatus according to claim 1 ,

wherein the plurality of pieces of training property information are derived from the training images through image analysis, and

wherein each of the plurality of pieces of training property information is at least one word that describes a property of a structure of interest included in the corresponding training image.

13. The document creation support apparatus according to claim 1 , wherein the processor is further configured to:

in response to the term being excessive information with respect to the property information, regenerate another sentence by using the learning model such that the term is designated to be not included in the another sentence, and

in response to the term being deficiency information with respect to the property information, regenerate another sentence by using the learning model such that the term is designated to be included in the another sentence.

14. A document creation support method comprising:

analyzing an image to derive property information describing a property of a structure of interest included in the image;

inputting the property information into a learning model to generate a sentence related to the image, wherein the learning model is trained by a combination of a plurality of pieces of training property information derived from a plurality of training images and a plurality of medical sentences generated from the plurality of pieces of training property information;

analyzing the sentence to specify a term representing the property related to the structure of interest included in the sentence;

collating the property information with the term, and

in response to the property information not being collated with the term, generating a parameter to indicate that the term is either excess information or deficiency information with respect to the property information and regenerating another sentence by using the learning model according to the parameter.

15. A non-transitory computer-readable storage medium that stores a document creation support program causing a computer to execute a procedure comprising:

analyzing an image to derive property information describing a property of a structure of interest included in the image;

inputting the property information into a learning model to generate a sentence related to the image, wherein the learning model is trained by a combination of a plurality of pieces of training property information derived from a plurality of training images and a plurality of medical sentences generated from the plurality of pieces of training property information;

analyzing the sentence to specify a term representing a property related to the structure of interest included in the sentence;

collating the property information with the term; and

in response to the property information not being collated with the term, generating a parameter to indicate that the term is either excess information or deficiency information with respect to the property information and regenerating another sentence by using the learning model according to the parameter.