Systems and methods for automatically diagnosing X-ray images
Various systems, methods, and non-transitory computer readable mediums for diagnosing a query X-ray image are disclosed. Example embodiments relate to operating the system to apply an artificial neural network model to the query X-ray image to extract a query feature vector representative of image characteristics of the query X-ray image; compare the query feature vector with one or more annotated feature vectors associated with respective one or more annotated X-ray images stored in an annotated image database to determine a similarity level between the query X-ray image and each annotated image of the one or more annotated images; associate the query X-ray image with a set of annotated images based at least on the similarity level and a similarity threshold; and assign the query X-ray image with a disease classification based at least on the disease classification of one or more annotated images of the set of annotated images.
1. A method for diagnosing a query X-ray image, the method comprising:
applying an artificial neural network model to the query X-ray image to extract a query feature vector representative of image characteristics of the query X-ray image, wherein applying the artificial neural network model to the query X-ray image comprises:
dividing the query X-ray image into one or more query image portions;
generating a value to represent the image characteristics within each query image portions of the one or more query image portions; and
generating the query feature vector to include the value for each query image portion;
comparing the query feature vector with one or more annotated feature vectors associated with respective one or more annotated X-ray images stored in an annotated image database to determine a similarity level between the query X-ray image and each annotated X-ray image of the one or more annotated X-ray images;
associating the query X-ray image with a set of annotated X-ray images based at least on the similarity level and a similarity threshold; and
assigning the query X-ray image with a disease classification based at least on the disease classification of one or more annotated X-ray images of the set of annotated X-ray images.
2. The method of claim 1 further comprises preprocessing the query X-ray image prior to applying the artificial neural network model to the query X-ray image.
3. The method of claim 2 , wherein preprocessing the query X-ray image comprises:
dividing the query X-ray image into one or more query image portions.
4. The method of claim 3 , wherein dividing the query X-ray image into the one or more query image portions comprises:
dividing the query X-ray image substantially equally to generate a first and a second query image portion.
5. The method of claim 4 , wherein the query X-ray image comprises an X-ray image of a lung, and the first and the second query image portions each corresponding to a left-side of the lung and a right-side of the lung, respectively.
6. The method of claim 5 further comprises horizontally flipping the second query image portion corresponding to the right-side of the lung.
7. The method of claim 1 further comprises:
comparing the query feature vector with the one or more annotated feature vectors to determine the similarity level between each query image portion and a related annotated image portion of the one or more annotated X-ray images;
associating each query image region with a set of annotated image portions based at least on the similarity level and the similarity threshold; and
assigning each query image portion with the disease classification based at least on the disease classification of one or more annotated image portions of the set of annotated image portions.
8. The method of claim 7 further comprises:
defining a set of related query image portions for each query image portion assigned the disease classification associated with a disease present identifier; and
for each related query image portion:
applying the artificial neural network model to extract a related query image region feature vector representative of image characteristics of that related query image portion;
comparing the related query image region feature vector with one or more annotated image region feature vectors associated with respective one or more annotated image portions to determine the similarity level between that related query image portion and each annotated image portion;
associating that related query image portion with a set of annotated image portions based at least on the similarity level and the similarity threshold; and
assigning that related query image portion with the disease classification based at least on the disease classification of one or more annotated image portions.
9. The method of claim 8 , wherein defining the set of related query image portions for each query image portion assigned the disease classification associated with the disease present identifier comprises:
dividing the query image portion into one or more related query image portions.
10. The method of claim 8 , wherein defining the set of related query image portions for each query image portion assigned the disease classification associated with the disease present identifier comprises:
defining one or more related query image portions based on a predefined distance from a center of that query image portion.
11. The method of claim 1 , wherein comparing the query feature vector with the one or more annotated feature vectors to determining the similarity level comprises:
applying a distance transform to the query feature vector and each annotated feature vector; and
assigning the similarity level based on a distance generated from applying the distance transform, wherein the similarity level increases with shorter distances.
12. The method of claim 1 , wherein assigning the query X-ray image with the disease classification based at least on the disease classification of one or more annotated X-ray images of the set of annotated X-ray images comprises:
determining the disease classification of a majority of annotated X-ray images within the set of annotated X-ray images; and
assigning the query X-ray image with the determined disease classification.
13. A system for diagnosing a query X-ray image, the system comprising:
a communication component to provide access to one or more X-ray images via a network; and
a processor in communication with the communication component, the processor being operable to:
apply an artificial neural network model to the query X-ray image to extract a query feature vector representative of image characteristics of the query X-ray image, wherein the processor is operable to:
divide the query X-ray image into one or more query image portions;
generate a value to represent the image characteristics within each query image portions of the one or more query image portions; and
generate the query feature vector to include the value for each query image portion;
compare the query feature vector with one or more annotated feature vectors associated with respective one or more annotated X-ray images stored in an annotated image database to determine a similarity level between the query X-ray image and each annotated X-ray image of the one or more annotated X-ray images;
associate the query X-ray image with a set of annotated X-ray images based at least on the similarity level and a similarity threshold; and
assign the query X-ray image with a disease classification based at least on the disease classification of one or more annotated X-ray images of the set of annotated X-ray images.
14. The system of claim 13 , wherein the processor is further operable to preprocess the query X-ray image prior to applying the artificial neural network model to the query X-ray image.
15. The system of claim 14 , wherein the processor is operable to divide the query X-ray image into one or more query image portions.
16. The system of claim 15 , wherein the processor is operable to divide the query X-ray image substantially equally to generate a first and a second query image portion.
17. The system of claim 16 , wherein the query X-ray image comprises an X-ray image of a lung, and the first and the second query image portions each corresponding to a left-side of the lung and a right-side of the lung, respectively.
18. The system of claim 17 , wherein the processor is further operable to horizontally flip the second query image portion corresponding to the right-side of the lung.
19. The system of claim 13 , wherein the processor is further operable to:
compare the query feature vector with the one or more annotated feature vectors to determine the similarity level between each query image portion and a related annotated image portion of the one or more annotated X-ray images;
associate each query image region with a set of annotated image portions based at least on the similarity level and the similarity threshold; and
assign each query image portion with the disease classification based at least on the disease classification of one or more annotated image portions of the set of annotated image portions.
20. The system of claim 19 , wherein the processor is further operable to:
define a set of related query image portions for each query image portion assigned the disease classification associated with a disease present identifier; and
for each related query image portion:
apply the artificial neural network model to extract a related query image region feature vector representative of image characteristics of that related query image portion;
compare the related query image region feature vector with one or more annotated image region feature vectors associated with respective one or more annotated image portions to determine the similarity level between that related query image portion and each annotated image portion;
associate that related query image portion with a set of annotated image portions based at least on the similarity level and the similarity threshold; and
assign that related query image portion with the disease classification based at least on the disease classification of one or more annotated image portions.
21. The system of claim 20 , wherein the processor is operable to divide the query image portion into one or more related query image portions.
22. The system of claim 20 , wherein the processor is operable to define one or more related query image portions based on a predefined distance from a center of that query image portion.
23. The system of claim 13 , wherein the processor is operable to:
apply a distance transform to the query feature vector and each annotated feature vector; and
assign the similarity level based on a distance generated from applying the distance transform, wherein the similarity level increases with shorter distances.
24. The system of claim 13 , wherein the processor is operable to:
determine the disease classification of a majority of annotated X-ray images within the set of annotated X-ray images; and
assign the query X-ray image with the determined disease classification.