Improving model performance by artificial blending of healthy tissue
View Patent ↗A method for improving machine learning algorithm performance is described. The method may comprise receiving a first constituent image of human tissue; receiving a second constituent image of human tissue; overlapping a portion of the second constituent image on a portion of the first constituent image to create an augmented image; and training a model using a dataset comprising at least the augmented image.
1. A method for improving machine learning algorithm performance comprising:
receiving a first constituent image comprising healthy human tissue;
receiving a second constituent image comprising unhealthy human tissue;
overlapping a portion of the second constituent image on a portion of the first constituent image to create an augmented image;
generating, using a function, a value of a pixel of the augmented image, wherein the pixel is at a grid location in the overlapping portions of the second and the first constituent images within the augmented image, and wherein the function receives as a first input a first value of a pixel of the first constituent image at a grid location within the first constituent image which corresponds to the grid location of the pixel of the augmented image and as a second input a second value of a pixel of the second constituent image at a grid location within the second constituent image which corresponds to the grid location of the pixel of the augmented image; and
training a model using a dataset comprising at least the augmented image.
2. The method of claim 1 , wherein training the model further comprises classifying the augmented image based on a classifier of the first constituent image and on a classifier of the second constituent image.
3. The method of claim 1 , wherein the function may comprise a maximum value of any of the pixels of the constituent images at the location.
4. The method of claim 1 , wherein the function may comprise a weighted sum of the values of the pixels of the constituent images at the location.
5. The method of claim 4 , wherein weights of the weighted sum may be based on at least one of the following: a first image classification, a second image classification, a position in the first constituent image, or a position in the second constituent image.
6. A system for improving machine learning algorithm performance comprising:
a processor;
memory accessible by the processor; and
computer program instructions stored in the memory and executable by the processor to perform the steps of:
receiving a first constituent image comprising healthy human tissue;
receiving a second constituent image comprising unhealthy human tissue:
overlapping a portion of the second constituent image on a portion of the first constituent image to create an augmented image;
generating, using a function, a value of a pixel of the augmented image, wherein the pixel is at a grid location in the overlapping portions of the second and the first constituent images within the augmented image, and wherein the function receives as a first input a first value of a pixel of the first constituent image at a grid location within the first constituent image which corresponds to the grid location of the pixel of the augmented image and as a second input a second value of a pixel of the second constituent image at a grid location within the second constituent image which corresponds to the grid location of the pixel of the augmented image; and
training a model using a dataset comprising at least the augmented image.
7. The system of claim 6 , wherein training the model further comprises classifying the augmented image based on a classifier of the first constituent image and a classifier of the second constituent image.
8. The system of claim 6 , wherein the function may comprise a maximum value of any of the pixels of the constituent images at the location.
9. The system of claim 6 , wherein the function may comprise a weighted sum of the values of the pixels of the constituent images at the location.
10. The system of claim 9 , wherein weights of the weighted sum may be based on at least one of the following: a first image classification, a second image classification, a position in the first constituent image, or a position in the second constituent image.
11. A computer program product for improving machine learning algorithm performance, the computer program product comprising a non-transitory computer readable storage having program instructions embodied therewith, the program instructions executable by a computer, to cause the computer to perform a method comprising:
receiving a first constituent image comprising healthy human tissue;
receiving a second constituent image comprising unhealthy human tissue:
overlapping a portion of the second constituent image on a portion of the first constituent image to create an augmented image;
generating, using a function, a value of a pixel of the augmented image, wherein the pixel is at a grid location in the overlapping portions of the second and the first constituent images within the augmented image, and wherein the function receives as a first input a first value of a pixel of the first constituent image at a grid location within the first constituent image which corresponds to the grid location of the pixel of the augmented image and as a second input a second value of a pixel of the second constituent image at a grid location within the second constituent image which corresponds to the grid location of the pixel of the augmented image; and
training a model using a dataset comprising at least the augmented image.
12. The computer program product of claim 11 , wherein training the model further comprises classifying the augmented image based on a classifier of the first constituent image and a classifier of the second constituent image.
13. The computer program product of claim 11 , wherein the function may comprise a weighted sum of the values of the pixels of the constituent images at the location and wherein the weights of the weighted sum may be based on at least one of the following: a first image classification, a second image classification, a position in the first constituent image, or a position in the second constituent image.