System and method for automated annotation of radiology findings

1. A method for automatically annotating a set of radiology images, the method comprising:

receiving, at a remote computing system, a set of Digital Imaging and Communications in Medicine (DICOM) images, wherein the set of DICOM images is associated with a patient;

processing the set of DICOM images, comprising:

with a first convolutional neural network (CNN), the CNN comprising at least one of a V-net and a U-net, determining an anatomical feature associated with the set of DICOM images;

based on the anatomical feature and with a lookup table, determining a second CNN, the second CNN comprising at least one of a V-net and a U-net;

with the second CNN:

determining a set of abnormal findings associated with the set of DICOM images; and

determining a set of normal findings associated with the set of DICOM images;

assigning a set of labels to the set of DICOM images, wherein the set of labels is determined based on the set of abnormal findings and the set of normal findings;

automatically determining a set of annotations based on the set of labels, wherein the set of annotations comprises:

a first subset of annotations, wherein the first subset of annotations is associated with the set of abnormal findings; and

a second subset of annotations, wherein the second subset of annotations is associated with the set of normal findings;

producing a set of DICOM overlays associated with the set of DICOM images and comprising the set of annotations;

automatically displaying, at a display of a radiology workstation, the set of DICOM images and the set of DICOM overlays, wherein:

each of the first subset of annotations is displayed automatically upon display of an associated DICOM image; and

each of the second subset of annotations is displayed in an event that a user hovers a cursor proximal to an annotation of the second subset of annotations and for a duration of time while the user hovers the cursor proximal to the annotation of the second subset of annotations;

in response to detecting a hotkey press from the user while the annotation of the second subset of annotations is displayed, increasing the duration of time such that the annotation of the second subset of annotations is displayed throughout an entire remaining duration of display of the associated DICOM image;

contemporaneously and automatically with displaying the set of DICOM images, displaying a second set of DICOM images associated with the patient, wherein the second set of DICOM images is recorded prior to the set of DICOM images, and wherein the second set of DICOM images is automatically processed in parallel with the set of DICOM images to determine a second set of annotations.

2. The method of claim 1 , wherein each of the second set of DICOM images is linked with a corresponding image of the set of DICOM images, wherein scrolling between images of the set of DICOM images triggers a scrolling between corresponding images of the second set of DICOM images.

3. The method of claim 2 , further comprising, in the event that the user hovers the cursor proximal to the annotation of the second subset of annotations, displaying a corresponding annotation associated with the second set of DICOM images.

4. The method of claim 1 , further comprising preprocessing any or all of the set of DICOM images, wherein preprocessing any or all of the set of DICOM images comprises rotating one or more images.

5. The method of claim 1 , further comprising automatically filling in at least a portion of a radiologist report based on the set of DICOM overlays.

6. The method of claim 5 , wherein the portion comprises a findings section.

7. The method of claim 1 , wherein the set of DICOM images is received from a Radiology Information System (RIS).

8. The method of claim 1 , wherein the set of DICOM images is received from a Picture Archiving and Communication System (PACS).

9. A method for automatically annotating a set of radiology images, the method comprising:

receiving, at a remote computing system, a set of radiology images;

with a first set of deep learning models, determining an anatomical feature associated with the set of radiology images;

based on the anatomical feature and with a lookup table, determining a second set of deep learning models;

with the second set of deep learning models:

determining a set of abnormal findings associated with the set of radiology images; and

determining a set of normal findings associated with the set of radiology images;

assigning a set of labels to the set of radiology images, wherein the set of labels is determined based on the set of abnormal findings and the set of normal findings;

determining a set of annotations based on the set of labels, wherein the set of annotations comprises:

a first subset of annotations; and

a second subset of annotations;

producing a set of image overlays associated with the set of radiology images and comprising the set of annotations;

displaying, at a display of a radiology workstation, the set of radiology images and the set of image overlays, wherein:

each of the first subset of annotations is displayed automatically upon display of an associated radiology image and throughout an entire duration of display of the associated radiology image; and

each of the second subset of annotations is displayed in an event that a user hovers a cursor proximal to an annotation of the second subset of annotations and for a duration of time while the user hovers the cursor proximal to the annotation of the second subset of annotations; and

in response to detecting a hotkey press from the user while the annotation of the second subset of annotations is displayed, increasing the duration of time such that the annotation of the second subset of annotations is displayed throughout an entire remaining duration of display of the associated radiology image.

10. The method of claim 9 , wherein the first subset of annotations is associated with the set of abnormal findings and wherein the second subset of annotations is associated with the set of normal findings.

11. The method of claim 9 , wherein the set of radiology images is associated with a patient, the method further comprising, contemporaneously with displaying the set of radiology images, displaying a second set of radiology images associated with the patient, wherein the second set of radiology images is recorded prior to the set of radiology images.

12. The method of claim 11 , wherein each of the second set of radiology images is linked with a corresponding image of the set of radiology images, wherein scrolling between images of the set of radiology images triggers a scrolling between corresponding images of the second set of radiology images.

13. The method of claim 12 , further comprising, in the event that the user hovers the cursor proximal to the annotation of the second subset of annotations, displaying a corresponding annotation associated with the second set of radiology images.

14. The method of claim 9 , further comprising preprocessing any or all of the set of radiology images, wherein preprocessing any or all of the set of radiology images comprises rotating one or more images.

15. The method of claim 9 , further comprising automatically filling in a findings section of a radiologist report based on the set of image overlays.

16. The method of claim 9 , wherein the set of radiology images is received from a Radiology Information System (RIS).