Method of analyzing X-ray images and report generation

1. A method of analyzing images and generating a report, the method comprising:

inputting, by an image analyzing device, a test image to a trained prediction model, wherein the prediction model is a deep learning-based model;

obtaining, by the image analyzing device, at least one abnormality associated with the test image, using the trained prediction model;

identifying, by the image analyzing device, at least one relevant predetermined region from a plurality of predetermined regions associated with the test image, using the trained prediction model, based on the at least one abnormality, wherein each of the at least one relevant predetermined region is identified based on hierarchical classification; and

determining, by the image analyzing device, within each of the at least one relevant predetermined region, a location associated with the at least one abnormality; and

obtaining the report, based on the at least one identified abnormality and the location associated with the at least one abnormality for each of the at least one relevant predetermined region, from a trained report generating model, wherein the report generating model is a deep learning-based model, wherein obtaining the report comprises:

inputting each of the at least one abnormality, latent embeddings associated with the test image, and the location associated with each of the at least one abnormality to the trained report generating model, wherein the report generating model is trained based on a Radiological Finding Quality Index (RFQI) loss with LSM loss; and

receiving the report for the abnormality from the report generation model.

2. The method as claimed in claim 1 comprises:

training the prediction model and the report generating model using a training dataset, wherein the training dataset comprises a plurality of training images and a diagnostic report corresponding to each of the plurality of training images.

3. The method as claimed in claim 1 , wherein the training further comprises:

analyzing each of a plurality of diagnostic reports, based on one or more tokens, to:

identify one or more keywords matching with each of the one or more tokens, based on a Location Synonym Map (LSM);

identify an abnormality, associated with the diagnostic report; and

generate at least one sentence, using the one or more keywords,

wherein each of the plurality of diagnostic reports is in unstructured format,

wherein the one or more tokens are retrieved from a domain knowledge database; and

categorizing a plurality of abnormalities associated with the plurality of diagnostic reports into the at least one predetermined region within the image, corresponding to each of the plurality of abnormalities, to obtain a structured report.

4. The method as claimed in claim 3 , wherein obtaining the structured report comprises:

creating a plurality of region graphs for each of the at least one predetermined region within the training image;

mapping each of the plurality of abnormalities with each of the at least one predetermined region, based on attention weights predicted using a Graph Attention model; and

creating the structured report, based on the mapping.

5. The method as claimed in claim 2 , wherein the report generating model is trained using the training dataset comprising the one or more keywords, the identified abnormality associated with the report, and the at least one sentence, and the structured report.

6. The method as claimed in claim 1 , wherein identifying each of the at least one relevant predetermined region based on the hierarchical classification comprises:

classifying each of the at least one abnormality associated with the at least one relevant predetermined region into a plurality of classes, wherein each of the plurality of classes comprises a set of related abnormalities.

7. The method as claimed in claim 1 , wherein the test image is one of a radiography image, an ultrasonography image, a Computed Tomography (CT) scan image, a Magnetic Resonance Imaging (MRI) image, and a radiation therapy image.

8. A system for analyzing images and generating a report, the system comprising:

a processor; and

a memory communicatively coupled to the processor, wherein the memory stores processor-executable instructions, which, on execution, causes the processor to:

input a test image to a trained prediction model, wherein the prediction model is a deep learning-based model;

obtain at least one abnormality associated with the test image, using the trained prediction model;

identify at least one relevant predetermined region from a plurality of predetermined regions associated with the test image, using the trained prediction model, based on the at least one abnormality, wherein each of the at least one relevant predetermined region is identified based on hierarchical classification; and

determine within each of the at least one relevant predetermined region, a location associated with the at least one abnormality; and

obtain the report, based on the at least one identified abnormality and the location associated with the at least one abnormality for each of the at least one relevant predetermined region, from a trained report generating model, wherein the report generating model is a deep learning-based model, wherein to obtain the report the processor-executable instructions further cause the processor to:

input each of the at least one abnormality, latent embeddings associated with image, and the location associated with each of the at least one abnormality to the trained report generating model, wherein the report generating model is trained based on a Radiological Finding Quality Index (RFQI) loss with LSM loss; and

receive the report for the abnormality from the report generation model.

9. The system as claimed in claim 8 , wherein the processor-executable instructions further cause the processor to train the prediction model and the report generating model using a training dataset, wherein the training dataset comprises a plurality of training images and a diagnostic report corresponding to each of the plurality of training images.

10. The system as claimed in claim 8 , wherein to train the prediction model and the report generating model, the processor-executable instructions further cause the processor to:

analyze each of a plurality of diagnostic reports, based on one or more tokens, to:

identify one or more keywords matching with each of the one or more tokens, based on a Location Synonym Map (LSM);

identify an abnormality, associated with the diagnostic report; and

generate at least one sentence, using the one or more keywords,

wherein each of the plurality of diagnostic reports is in unstructured format,

wherein the one or more tokens are retrieved from a domain knowledge database; and

categorize a plurality of abnormalities associated with the plurality of diagnostic reports into the at least one predetermined region within the image, corresponding to each of the plurality of abnormalities, to obtain a structured report.

11. The system as claimed in claim 10 , wherein to obtain the structured report, the processor-executable instructions further cause the processor to:

create a plurality of region graphs for each of the at least one predetermined region within the training image;

map each of the plurality of abnormalities with each of the at least one predetermined region, based on attention weights predicted using a Graph Attention model; and

create the structured report, based on the mapping.

12. The system as claimed in claim 9 , wherein the report generating model is trained using the training dataset comprising the one or more keywords, the identified abnormality associated with the report, and the at least one sentence, and the structured report.

13. The system as claimed in claim 8 , wherein to identify each of the at least one relevant predetermined region based on the hierarchical classification, the processor-executable instructions further cause the processor to:

classify each of the at least one abnormality associated with the at least one relevant predetermined region into a plurality of classes, wherein each of the plurality of classes comprises a set of related abnormalities.

14. The system as claimed in claim 8 , wherein the test image is one of a radiography image, an ultrasonography image, a Computed Tomography (CT) scan image, a Magnetic Resonance Imaging (MRI) image, and a radiation therapy image.

15. A computer program product for analyzing images and generating a report, the computer program product being embodied in a non-transitory computer readable storage medium of an image analyzing device and comprising computer instructions for:

inputting, by an image analyzing device, a test image to a trained prediction model, wherein the prediction model is a deep learning-based model;

obtaining, by the image analyzing device, at least one abnormality associated with the test image, using the trained prediction model;

identifying, by the image analyzing device, at least one relevant predetermined region from a plurality of predetermined regions associated with the test image, using the trained prediction model, based on the at least one abnormality, wherein each of the at least one relevant predetermined region is identified based on hierarchical classification;

determining, by the image analyzing device, within each of the at least one relevant predetermined region, a location associated with the at least one abnormality; and

obtaining the report, based on the at least one identified abnormality and the location associated with the at least one abnormality for each of the at least one relevant predetermined region, from a trained report generating model, wherein the report generating model is a deep learning-based model, wherein obtaining the report comprises:

inputting each of the at least one abnormality, latent embeddings associated with the test image, and the location associated with each of the at least one abnormality to the trained report generating model, wherein the report generating model is trained based on a Radiological Finding Quality Index (RFQI) loss with LSM loss; and

receiving the report for the abnormality from the report generation model.