Automated fish reader using learning machines
View Patent ↗An automated reader for reading fluorescence in-situ hybridization signals includes one or more computer processors for receiving a digitized FISH image and executing the steps of converting colors within the image to a hue value, separately for each color extracting quantitative values to detect the presence of signals corresponding to spots and applying a plurality of algorithms to extract features from the signals to determine cell shapes and segment cells within the FISH image. After recombining the signals, the extracted features for the colors learning machines are used to classify the spots according to the color and separate merged signals of classified spots that are in close proximity to each other within the image. The classified spots are counted to determine relative frequency of colors and a report is generated providing the number of classified spots of each color.
1. An automated reader for reading fluorescence in-situ hybridization signals, comprising:
one or more computer processors for receiving a digitized FISH image comprising a plurality of spots having colors, wherein the one or more processors being are programmed to execute the steps of:
converting colors within the FISH image to a hue value;
separately for each color:
extracting a plurality of quantitative values to detect the presence of signals corresponding to spots; and
applying a plurality of algorithms to extract features from the signals to determine cell shapes and segment cells within the FISH image;
recombining the extracted features for the colors;
using a first learning machine and a first subset of the extracted features, classifying the plurality of spots according to the color of each spot to generate spot classifications, wherein at least of portion of said spot classifications are associated with close proximity or partially overlapping spots within the FISH image and comprise merged signals;
using a second learning machine and a second subset of the extracted features, separating said merged signals to generate separated spot classifications;
counting said spot classifications and said separated spot classifications to determine relative frequency of colors among the plurality of spots; and
generating a report of the number of classified spots of each color.
2. The automated reader of claim 1 , wherein the first learning machine is a support vector machine.
3. The automated reader of claim 1 , wherein the second learning machine is a support vector machine.
4. The automated reader of claim 1 , wherein the plurality of algorithms to extract features comprises a Hough transform.
5. The automated reader of claim 1 , wherein the plurality of algorithms to extract features comprises a Markov random field model.
6. The automated reader of claim 1 , wherein the plurality of algorithms to extract features comprises a Gaussian mixture model.
7. The automated reader of claim 1 , wherein the one or more processor is further programmed to execute a third learning machine to associate signals with cells within the FISH image.
8. The automated reader of claim 1 , further comprising a user interface comprising a masking tool for manually selecting one or more regions within the FISH image for analysis.
9. The automated reader of claim 8 , wherein the masking tool is further operable for excluding one or more regions within the FISH images from analysis.
10. A method for reading fluorescence in-situ hybridization signals, comprising:
receiving in one or more computer processors a digitized FISH image comprising a plurality of spots having colors;
converting colors within the FISH image to a hue value;
separately for each color:
extracting a plurality of quantitative values to detect the presence of signals corresponding to spots; and
applying a plurality of algorithms to extract features from the signals to determine cell shapes and segment cells within the FISH image; recombining the extracted features for the colors;
using a first learning machine and a first subset of the extracted features, classifying the plurality of spots according to the color of each spot to generate spot classifications, wherein at least of portion of said spot classifications are associated with close proximity or partially overlapping spots within the FISH image and comprise merged signals;
using a second learning machine and a second subset of the extracted features, separating said merged signals to generate separated spot classifications;
counting said spot classifications and said separated spot classifications to determine relative frequency of colors among the plurality of spots; and
generating a report of the number of classified spots of each color.
11. The method of claim 10 , wherein the first learning machine is a support vector machine.
12. The method of claim 10 , wherein the second learning machine is a support vector machine.
13. The method of claim 10 , wherein the plurality of algorithms to extract features comprises a Hough transform.
14. The method of claim 10 , wherein the plurality of algorithms to extract features comprises a Markov random field model.
15. The method of claim 10 , wherein the plurality of algorithms to extract features comprises a Gaussian mixture model.
16. The method of claim 10 , wherein the one or more processor is further programmed to execute a third learning machine to associate signals with cells within the FISH image.
17. The method of claim 10 , further comprising a user interface comprising a masking tool for manually selecting one or more regions within the FISH image for analysis.
18. The method of claim 17 , wherein the masking tool is further operable for excluding one or more regions within the FISH images from analysis.
19. The method of claim 1 , wherein the first subset of the extracted features comprises an intensity value for each color, a preliminary color assignment, a spot perimeter, a spot area, and a spot eccentricity.
20. The method of claim 1 , wherein the second subset of the extracted features comprises a distance between two signals, current color assignments, areas of the two signals, and signal intensity for the two signals.
21. The method of claim 10 , wherein the first subset of the extracted features comprises an intensity value for each color, a preliminary color assignment, a spot perimeter, a spot area, and a spot eccentricity.
22. The method of claim 10 , wherein the second subset of the extracted features comprises a distance between two signals, current color assignments, areas of the two signals, and signal intensity for the two signals.