Feature based neural network regression for feature suppression
View Patent ↗A method of obtaining one or more components from an image may include normalizing and pre-processing the image to obtain a processed image. Features may be extracted from the processed image. Neural-network-based regression may then be performed on the set of extracted features to predict the one or more components. These techniques may be applied, for example, to the problem of extracting and removing bone components from radiographic images, which may be thoracic (lung) images.
1. A method of obtaining one or more image components from an image, the method comprising:
normalizing and pre-processing the image to obtain a processed image;
extracting features from the processed image to obtain a set of extracted features; and
performing neural-network-based regression based on the set of extracted features to predict the one or more image components.
2. The method of claim 1 , further comprising:
subtracting the one or more image components from the image to obtain an image with the one or more image components removed.
3. The method of claim 2 , further comprising performing a computer-aided detection (CAD) operation on the image with the one or more image components removed.
4. The method of claim 1 , wherein the image comprises a radiographic image, and wherein the one or more image components comprise one or more bone structures.
5. The method of claim 1 , wherein said normalizing and pre-processing includes:
performing a wavelet decomposition on a normalized image.
6. The method of claim 5 , wherein said normalizing and pre-processing includes:
image resizing to obtain a resized image; and
bit depth adjustment of the resized image to obtain said normalized image.
7. The method of claim 5 , wherein said normalizing and pre-processing further comprises:
performing at least one operation on at least one result of said wavelet decomposition, wherein the at least one operation is selected from the group consisting of: gray scale registration; and noise removal and enhancement.
8. The method of claim 5 , wherein said performing a wavelet decomposition results in a set of wavelet components that includes a coarse residual component comprising low-frequency information.
9. The method of claim 8 , wherein said normalization and pre-processing further comprises:
performing, on said set of wavelet components, except for the coarse residual component, at least one operation selected from the group consisting of: gray scale registration; and noise removal and enhancement.
10. The method of claim 8 , wherein said normalization and pre-processing further comprises:
dynamically weighting the coarse residual component.
11. The method of claim 5 , wherein said wavelet decomposition comprises a five-scale wavelet decomposition using a third-order basic spline wavelet.
12. The method of claim 11 , further comprising:
obtaining shape-index images for each of the five scales.
13. The method of claim 11 , wherein said extracting features includes:
obtaining Gaussian 4-jets at the five scales.
14. The method of claim 1 , wherein said performing neural-network-based regression comprises:
applying multiple regression models to obtain multiple bone value predictions for one or more image pixels; and
combining the multiple bone value predictions to obtain a combined bone value.
15. The method of claim 14 , wherein said combined bone value is obtained by averaging the multiple bone value predictions.
16. The method of claim 1 , wherein said performing neural-network-based regression comprises:
applying multiple regression models corresponding to multiple image zones to obtain bone value predictions for the pixels of the multiple image zones.
17. The method of claim 1 , further comprising:
downloading software instructions to implement said normalizing and pre-processing, said extracting features, and said performing neural-network-based regression.
18. A machine-readable storage medium containing instructions that, upon execution by a processor, cause the processor to implement a method of obtaining one or more image components from an image, the method comprising:
normalizing and pre-processing the image to obtain a processed image;
extracting features from the processed image to obtain a set of extracted features; and
performing neural-network-based regression based on the set of extracted features to predict the one or more image components.
19. The medium of claim 18 , wherein the method further comprises:
subtracting the one or more image components from the image to obtain an image with the one or more image components removed.
20. The medium of claim 19 , wherein the method further comprises performing a computer-aided detection (CAD) operation on the image with the one or more image components removed.
21. The medium of claim 18 , wherein the image comprises a radiographic image, and wherein the one or more image components comprise one or more bone structures.
22. The medium of claim 18 , wherein said normalizing and pre-processing includes:
performing a wavelet decomposition on a normalized image.
23. The medium of claim 22 , wherein said normalizing and pre-processing includes:
image resizing to obtain a resized image; and
bit depth adjustment of the resized image to obtain said normalized image.
24. The medium of claim 22 , wherein said normalizing and pre-processing further comprises:
performing at least one operation on at least one result of said wavelet decomposition, wherein the at least one operation is selected from the group consisting of: gray scale registration; and noise removal and enhancement.
25. The medium of claim 22 , wherein said performing a wavelet decomposition results in a set of wavelet components that includes a coarse residual component comprising low-frequency information.
26. The medium of claim 25 , wherein said normalization and pre-processing further comprises:
performing, on said set of wavelet components, except for the coarse residual component, at least one operation selected from the group consisting of: gray scale registration; and noise removal and enhancement.
27. The medium of claim 25 , wherein said normalization and pre-processing further comprises:
dynamically weighting the coarse residual component.
28. The medium of claim 22 , wherein said wavelet decomposition comprises a five-scale wavelet decomposition using a third-order basic spline wavelet.
29. The medium of claim 28 , wherein the method further comprises:
obtaining shape-index images for each of the five scales.
30. The medium of claim 28 , wherein said extracting features includes:
obtaining Gaussian 4-jets at the five scales.
31. The medium of claim 18 , wherein said performing neural-network-based regression comprises:
applying multiple regression models to obtain multiple bone value predictions for one or more image pixels; and
combining the multiple bone value predictions to obtain a combined bone value.
32. The medium of claim 31 , wherein said combined bone value is obtained by averaging the multiple bone value predictions.
33. The medium of claim 18 , wherein said performing neural-network-based regression comprises:
applying multiple regression models corresponding to multiple image zones to obtain bone value predictions for the pixels of the multiple image zones.
34. An apparatus for obtaining one or more image components from an image, comprising:
means for normalizing and pre-processing the image to obtain a processed image;
means for extracting features from the processed image to obtain a set of extracted features; and
means for performing neural-network-based regression based on the set of extracted features to predict the one or more image components.
35. The apparatus of claim 34 , further comprising:
means for subtracting the one or more image components from the image to obtain an image with the one or more components removed.
36. The apparatus of claim 35 , further comprising:
means for performing a computer-aided detection (CAD) operation on the image with the one or more image components removed.
37. The apparatus of claim 34 , wherein the image comprises a radiographic image, and wherein the one or more image components comprise one or more bone structures.
38. The apparatus of claim 34 , wherein said means for normalizing and pre-processing includes:
means for performing a wavelet decomposition on a normalized image.
39. The apparatus of claim 38 , wherein said means for normalizing and pre-processing includes:
means for image resizing to obtain a resized image; and
means for bit depth adjustment of the resized image to obtain said normalized image.
40. The apparatus of claim 38 , wherein said means for normalizing and pre-processing further comprises:
means for performing at least one operation on at least one result of said wavelet decomposition, wherein the at least one operation is selected from the group consisting of: gray scale registration; and noise removal and enhancement.
41. The apparatus of claim 38 , wherein said means for performing a wavelet decomposition results in a set of wavelet components that includes a coarse residual component comprising low-frequency information.
42. The apparatus of claim 41 , wherein said means for normalization and pre-processing further comprises:
means for performing, on said set of wavelet components, except for the coarse residual component, at least one operation selected from the group consisting of: gray scale registration;
and noise removal and enhancement.
43. The apparatus of claim 41 , wherein said means for normalization and pre-processing further comprising:
means for dynamically weighting the coarse residual component.
44. The apparatus of claim 38 , wherein said wavelet decomposition comprises a five-scale wavelet decomposition using a third-order basic spline wavelet.
45. The apparatus of claim 44 , further comprising:
means for obtaining shape-index images for each of the five scales.
46. The apparatus of claim 44 , wherein said means for extracting features includes:
means for obtaining Gaussian 4-jets at the five scales.
47. The apparatus of claim 34 , wherein said means for performing neural-network-based regression comprises:
means for applying multiple regression models to obtain multiple bone value predictions for one or more image pixels; and
means for combining the multiple bone value predictions to obtain a combined bone value.
48. The apparatus of claim 47 , wherein said combined bone value is obtained by averaging the multiple bone value predictions.
49. The apparatus of claim 34 , wherein said means for performing neural-network-based regression comprises:
means for applying multiple regression models corresponding to multiple image zones to obtain bone value predictions for the pixels of the multiple image zones.