System and method for quantifying the extent of disease from 2-D images
View Patent ↗A disease assessment method includes using a machine learning system and 2-D image data to quantify an extent of disease in the patient, wherein the machine learning system has been previously trained using 3-D image data for each of a plurality of 3-D images obtained from a plurality of subjects, and 2-D image data for each of a plurality of 2-D images obtained from the plurality of subjects, wherein each of the 2-D images is associated with a corresponding one of the 3-D images and is captured at the same time as the corresponding one of the 3-D images or within a certain predetermined time frame of capturing the corresponding one of the 3-D images, and wherein a ground truth for the 2-D images comprises for each of the 3-D images a quantification of an extent of the predetermined disease based on the 3-D image data for the 3-D image.
1. A method of training a predictive machine learning system for of lung disease, comprising:
receiving in a computer system (i) 3-D image data for each of a plurality of 3-D images obtained from a plurality of subjects, and (ii) 2-D image data for each of a plurality of 2-D images obtained from the plurality of subjects;
for each of the 3-D images, determining in the computer system a quantification of an extent of lung disease based on the 3-D image data for the 3-D image to create a ground truth for the 2-D images, wherein the quantification of the extent of lung disease comprises a percentage of lung low attenuation areas relative to total lung volume in the 3-D image based on a predetermined Hounsfield unit threshold; and
using each of the quantifications and the 2-D image data for each of the 2-D images to train the predictive machine learning system to be able to predict a quantitative extent of lung disease in a patient based on a received 2-D patient image.
2. The method according to claim 1 , wherein the using each of the quantifications and the 2-D image data for each of the 2-D images to train the predictive machine learning system comprises, for each of the 2-D images, labelling the 2-D image data of the 2-D image with the quantifications that was determined for the 3-D image that is associated with the 2-D image to create labeled 2-D image data, and using the labeled 2-D image data to train the predictive machine learning system.
3. The method according to claim 1 , wherein the predictive machine learning system is a deep learning system.
4. The method according to claim 3 , wherein the deep learning system is a deep neural network.
5. The method according to claim 4 , wherein the deep learning network is a convolution neural network.
6. The method according to claim 4 , wherein the deep learning network is a deep belief network.
7. The method according to claim 1 , wherein the 3-D images are each a CT image or an MRI image.
8. The method according to claim 1 , wherein each of the 3-D images is a CT image and each of the plurality of 2-D images is a 2-D CT scout radiograph.
9. The method according to claim 1 , wherein the received patient 2-D image is a 2-D x-ray image.
10. The method according to claim 1 , wherein each quantification comprises a percentage of lung voxels with a Hounsfield (HU) value below the predetermined Hounsfield unit threshold.
11. A computer program product, comprising a non-transitory computer usable medium having a computer readable program code embodied therein, the computer readable program code being adapted and configured to be executed to implement a method of training a predictive machine learning system for lung disease assessment according to claim 1 .
12. The method according to claim 1 , wherein each of the 2-D images is associated with but not derived form a corresponding one of the 3-D images and is captured by an imaging apparatus at the same time as the corresponding one of the 3-D images or within a certain predetermined time frame of capturing the corresponding one of the 3-D images.
13. The method according to claim 1 , wherein the predetermined Hounsfield unit threshold is a −950 Hounsfield unit threshold.
14. The method according to claim 1 , wherein the lung disease is emphysema.
15. A lung disease assessment method, comprising:
receiving 2-D patient image data representing a 2-D image of a patient, and
providing the 2-D patient image data to a predictive machine learning system and using the predictive machine learning system and the 2-D patient image data to quantify an extent of lung disease in the patient, wherein the predictive machine learning system has been previously trained using: (i) 3-D image data for each of a plurality of 3-D images obtained from a plurality of subjects, (ii) 2-D image data for each of a plurality of 2-D images obtained from the plurality of subjects; and (iii) a ground truth for the 2-D images comprising for each of the 3-D images a quantification of an extent of lung disease based on the 3-D image data for the 3-D image, wherein the quantification of the extent of lung disease comp percentage of lung low attenuation areas relative to total lung volume in the 3-D image based on a predetermined Hounsfield unit threshold.
16. The method according to claim 15 , wherein the predictive machine learning system is a deep learning system.
17. The method according to claim 16 , wherein the deep learning system is a deep neural network.
18. The method according to claim 17 , wherein the deep learning network is a convolution neural network.
19. The method according to claim 17 , wherein the deep learning network is a deep belief network.
20. The method according to claim 15 , wherein the 3-D images are each a CT image or an MRI image.
21. The method according to claim 15 , wherein each of the 3-D images is a CT image and each of the plurality of 2-D images is a 2-D CT scout radiograph.
22. The method according to claim 15 , wherein the 2-D image of the patient is a 2-D x-ray image.
23. The method according to claim 15 , wherein each quantification comprises a percentage of lung voxels with a Hounsfield (HU) value below the predetermined Hounsfield unit threshold.
24. A computer program product, comprising a non-transitory computer usable medium having a computer readable program code embodied therein, the computer readable program code being adapted to be executed to implement a lung disease assessment method as recited in claim 15 .
25. The method according to claim 15 , wherein each of the 2-D images is associated with but not derived form a corresponding one of the 3-D images and is captured by an imaging apparatus at the same time as the corresponding one of the 3-D images or within a certain predetermined time frame of capturing the corresponding one of the 3-D images.
26. The method according to claim 15 , wherein the predetermined Hounsfield unit threshold is a −950 Hounsfield unit threshold.
27. The method according to claim 15 , wherein the lung disease is emphysema.
28. A lung disease assessment system, comprising:
a computing device implementing a predictive machine learning system, the computing device having a processor apparatus structured and configured to:
receive 2-D patient image data representing a 2-D image of a patient; and
use the predictive machine learning system and the 2-D patient image data to quantify an extent of lung disease in the patient, wherein the predictive machine learning system has been previously trained using (1) 3-D image data for each of a plurality of 3-D images obtained from a plurality of subjects, (ii) 2-D image data for each of a plurality of 2-D images obtained from the plurality of subjects, and (iii) a ground truth for the 2-D images comprising for each of the 3-D images a quantification of an extent of lung disease based on the 3-D image data for the 3-D image, wherein the qualification of the extent of lung disease comprises a percentage of lung low attenuation areas relative to total lung volume in the 3-D image based on a predetermined Hounsfield unit threshold.
29. The system according to claim 28 , wherein the predictive machine learning system is a deep learning system.
30. The system according to claim 29 , wherein the deep learning system is a deep neural network.
31. The system according to claim 30 , wherein the deep learning network is a convolution neural network.
32. The system according to claim 30 , wherein the deep learning network is a deep belief network.
33. The system according to claim 28 , wherein the 3-D images are each a CT image or an MRI image.
34. The system according to claim 28 , wherein each of the 3-D images is a CT image and each of the plurality of 2-D images is a 2-D CT scout radiograph.
35. The system according to claim 28 , wherein the 2-D image of the patient is a 2-D x-ray image.
36. The system according to claim 28 , wherein each quantification comprises a percentage of lung voxels with a Hounsfield (HU) value below the predetermined Hounsfield unit threshold.
37. The system according to claim 28 , wherein each of the 2-D images is associated with but not derived form a corresponding one of the 3-D images and is captured by an imaging apparatus at the same time as the corresponding one of the 3-D images or within a certain predetermined time frame of capturing the corresponding one of the 3-D images.
38. The system according to claim 28 , wherein the predetermined Hounsfield unit threshold is a −950 Hounsfield unit threshold.
39. The system according to claim 28 , wherein the lung disease is emphysema.