Automated fracture detection using machine learning models
Techniques for fracture detection are provided. A first image is received to be processed to identify rib fractures. A first set of regions of interest (ROIs) is identified by processing the first image using a first machine learning model, where each ROI in the first set of ROIs corresponds to a first potential fracture. Further, a first ROI of the first set of ROIs is upsampled, and the system attempts to verify the first potential fracture in the first ROI by processing the upsampled first ROI using a second machine learning model.
1. A method comprising:
receiving a first image to be processed to identify rib fractures;
identifying a first set of regions of interest (ROIs), each corresponding to a respective potential fracture, by processing the first image using a first machine learning model trained to identify fractures, wherein at least one portion of the first image is not included in the set of ROIs; and
in response to identifying a first ROI, of the first set of ROIs, corresponding to a first potential fracture:
upsampling the first ROI to yield an upsampled first ROI that is higher resolution than the first ROI; and
attempting to verify the first potential fracture in the first ROI by processing the upsampled first ROI using a second machine learning model.
2. The method of claim 1 , the method further comprising, upon failing to verify the first potential fracture in the first ROI:
upsampling the first image; and
identifying a second set of ROIs by processing the upsampled first image using the first machine learning model.
3. The method of claim 1 , the method further comprising, upon verifying the first potential fracture in the first ROI:
providing an indication of the first potential fracture; and
terminating a processing pipeline being used to process the first image to identify rib fractures.
4. The method of claim 1 , the method further comprising:
receiving a second image to be processed to identify rib fractures;
determining that there are no ROIs in the second image, by processing the second image using the first machine learning model; and
upon determining that there are no ROIs in the second image:
upsampling the second image; and
processing the upsampled second image using the first machine learning model.
5. The method of claim 1 , the method further comprising:
identifying one or more ribs in the first image using a third machine learning model;
generating a first set of patches along a first rib of the one or more identified ribs;
identifying a second potential fracture in a first patch of the first set of patches, by processing the first patch using a fourth machine learning model;
upsampling the first patch; and
attempting to verify the second potential fracture in the first patch by processing the upsampled first patch using the second machine learning model.
6. The method of claim 5 , the method further comprising, upon failing to verify the second potential fracture in the first patch:
upsampling the first image; and
generating a second set of patches along the first rib of the one or more identified ribs.
7. The method of claim 5 , the method further comprising:
generating a second set of patches along a second rib of the one or more identified ribs;
upon failing to identify any potential fractures in a second patch of the second set of patches:
upsampling the first image;
generating a third set of patches along the second rib in the upsampled first image; and
processing a third patch of the third set of patches using the fourth machine learning model.
8. The method of claim 7 , the method further comprising:
determining, using the fourth machine learning model, that none of the third set of patches include a potential fracture;
providing an indication that a patient associated with the first image does not have a rib fracture; and
terminating a processing pipeline being used to process the first image to identify rib fractures.
9. The method of claim 1 , wherein the first image is received in a first resolution, the method further comprising:
prior to identifying the first set of ROIs, downsampling the first image to a second resolution that is relatively lower than the first resolution.
10. A computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more computer processors to perform an operation comprising:
receiving a first image to be processed to identify rib fractures;
identifying a first set of regions of interest (ROIs), each corresponding to a respective potential fracture, by processing the first image using a first machine learning model trained to identify fractures, wherein at least one portion of the first image is not included in the set of ROIs; and
in response to identifying a first ROI, of the first set of ROIs, corresponding to a first potential fracture:
upsampling the first ROI to yield an upsampled first ROI that is higher resolution than the first ROI; and
attempting to verify the first potential fracture in the first ROI by processing the upsampled first ROI using a second machine learning model.
11. The computer-readable storage medium of claim 10 , the operation further comprising, upon failing to verify the first potential fracture in the first ROI:
upsampling the first image; and
identifying a second set of ROIs by processing the upsampled first image using the first machine learning model.
12. The computer-readable storage medium of claim 10 , the operation further comprising:
receiving a second image to be processed to identify rib fractures;
determining that there are no ROIs in the second image, by processing the second image using the first machine learning model; and
upon determining that there are no ROIs in the second image:
upsampling the second image; and
processing the upsampled second image using the first machine learning model.
13. The computer-readable storage medium of claim 10 , the operation further comprising:
identifying one or more ribs in the first image using a third machine learning model;
generating a first set of patches along a first rib of the one or more identified ribs;
identifying a second potential fracture in a first patch of the first set of patches, by processing the first patch using a fourth machine learning model;
upsampling the first patch; and
attempting to verify the second potential fracture in the first patch by processing the upsampled first patch using the second machine learning model.
14. The computer-readable storage medium of claim 13 , the operation further comprising:
generating a second set of patches along a second rib of the one or more identified ribs;
upon failing to identify any potential fractures in a second patch of the second set of patches:
upsampling the first image;
generating a third set of patches along the second rib in the upsampled first image; and
processing a third patch of the third set of patches using the fourth machine learning model.
15. The computer-readable storage medium of claim 14 , the operation further comprising:
determining, using the fourth machine learning model, that none of the third set of patches include a potential fracture;
providing an indication that a patient associated with the first image does not have a rib fracture; and
terminating a processing pipeline being used to process the first image to identify rib fractures.
16. A system comprising:
one or more computer processors; and
a memory containing a program which when executed by the one or more computer processors performs an operation, the operation comprising:
receiving a first image to be processed to identify rib fractures;
identifying a first set of regions of interest (ROIs), each corresponding to a respective potential fracture, by processing the first image using a first machine learning model trained to identify fractures, wherein at least one portion of the first image is not included in the set of ROIs; and
in response to identifying a first ROI, of the first set of ROIs, corresponding to a first potential fracture:
upsampling the first ROI to yield an upsampled first ROI that is higher resolution than the first ROI; and
attempting to verify the first potential fracture in the first ROI by processing the upsampled first ROI using a second machine learning model.
17. The system of claim 16 , the operation further comprising, upon failing to verify the first potential fracture in the first ROI:
upsampling the first image; and
identifying a second set of ROIs by processing the upsampled first image using the first machine learning model.
18. The system of claim 16 , the operation further comprising:
receiving a second image to be processed to identify rib fractures;
determining that there are no ROIs in the second image, by processing the second image using the first machine learning model; and
upon determining that there are no ROIs in the second image:
upsampling the second image; and
processing the upsampled second image using the first machine learning model.
19. The system of claim 16 , the operation further comprising:
identifying one or more ribs in the first image using a third machine learning model;
generating a first set of patches along a first rib of the one or more identified ribs;
identifying a second potential fracture in a first patch of the first set of patches, by processing the first patch using a fourth machine learning model;
upsampling the first patch; and
attempting to verify the second potential fracture in the first patch by processing the upsampled first patch using the second machine learning model.
20. The system of claim 19 , the operation further comprising:
generating a second set of patches along a second rib of the one or more identified ribs;
upon failing to identify any potential fractures in a second patch of the second set of patches:
upsampling the first image;
generating a third set of patches along the second rib in the upsampled first image; and
processing a third patch of the third set of patches using the fourth machine learning model.