Systems and methods for detecting likelihood of malignancy for a patient using a pair of medical images
There is provided a computer implemented method for detection of likelihood of malignancy in an anatomical image of a patient for planning treatment thereof, comprising: receiving a pair of images that are either of a same patient or from two different patients, wherein the pair of images comprises anatomically corresponding anatomical images each depicting internal anatomical structures of the patient, feeding the pair of images into a model, outputting by the model, an indication of whether the pair of images are of a same patient or not, and generating an indication of likelihood of malignancy when the model wrongly outputs that the pair of images are not of the same patient, when in fact the pair of images are of the same patient, wherein treatment of the patient is planned according to the indication of likelihood of malignancy.
1. A computer implemented method for detection of likelihood of malignancy in an anatomical image of a patient for planning treatment thereof, comprising:
determining a likelihood of malignancy when a model fed by a pair of images of a same patient wrongly identifies the pair of images as not belonging to the same patient, when in fact the pair of images are of the same patient, by:
receiving the pair of images of the same patient, wherein the pair of images comprises anatomically corresponding anatomical images each depicting internal anatomical structures of the patient;
feeding the pair of images into the model,
outputting by the model, an indication of whether the pair of images are of the same patient or not; and
generating an indication of likelihood of malignancy when the model wrongly outputs that the pair of images are not of the same patient, wherein treatment of the patient is planned according to the indication of likelihood of malignancy.
2. The computer implemented method of claim 1 , wherein the indication of whether the pair of images are of the same patient is generated when a statistical distance metric between the pair of images is below a threshold, and the indication of likelihood of malignancy denoting that the pair of images are not of the same patient is generated when the statistical distance metric is above the threshold.
3. The computer implemented method of claim 1 , wherein the model is implemented based on a Siamese network architecture, wherein each of the pair of images is fed into a respective neural network component that outputs a respective set of features for each anatomical image, and wherein the indication of likelihood of malignancy is generated when a statistical distance between the sets of features of the pair of images is greater than a threshold indicating the pair of images are of the same patient.
4. The computer implemented method of claim 1 , wherein the model is trained using training images including only normal images, and excluding malignant images and suspicious images, wherein the training inputs are non-labeled indicative of no malignancy information being provided.
5. The computer implemented method of claim 1 , wherein the model is trained using training images of a screening population including low a rate of malignancy below a threshold.
6. The computer implemented method of claim 1 , wherein the pair of images are of different symmetric anatomical regions of the patient.
7. The computer implemented method of claim 1 , wherein the pair of images are mammographic images of two breasts of the patient.
8. The computer implemented method of claim 1 , wherein the pair of images are of a same anatomical region of the patient taken at different times separated by a clinically significant time interval long enough for malignancy to grow.
9. The computer implemented method of claim 1 , wherein the pair of images are of a same sensor view.
10. A computer implemented method for training a model for detection of likelihood of malignancy in an anatomical image of a patient for planning treatment thereof, comprising:
providing a set of training images including pairs of anatomically corresponding anatomical images of a plurality of sample patients, wherein each pair of images are either of a same patient or from two different patients;
training a model using the training images to identify whether or not a pair of anatomically corresponding anatomical images are of a same target patient; and
providing the model, wherein when a target pair of anatomically corresponding anatomical images of a same patient are fed into a model, an indication of malignancy is generated when the model outputs an indication that the pair of images are not of the same patient.
11. The computer implemented method of claim 10 , wherein the model learns a threshold for a statistical distance metric between the pair of anatomically corresponding anatomical images that separates between the pair of images being of the same person and not being of the same person.
12. The computer implemented method of claim 10 , wherein the model is implemented based on a Siamese network architecture, wherein each of the pair of images is fed into a respective neural network component that outputs a respective set of features for each anatomical image, and wherein the indication of likelihood of malignancy is generated when a statistical distance between the sets of features of the pair of images is greater than a threshold indicating the pair of images are of the same patient.
13. The computer implemented method of claim 10 , wherein the model is trained using training images including only normal images, and excluding malignant images and suspicious images, wherein the training are non-labeled.
14. The computer implemented method of claim 10 , wherein the model is trained using training images of a screening population including low a rate of malignancy below a threshold.
15. The computer implemented method of claim 10 , wherein the pair of images are of different symmetric anatomical regions of the patient.
16. The computer implemented method of claim 10 , wherein the pair of images are mammographic images of two breasts of the patient.
17. The computer implemented method of claim 10 , wherein the pair of images are of a same anatomical region of the patient taken at different times separated by a clinically significant time interval long enough for malignancy to grow.
18. The computer implemented method of claim 10 , wherein the pair of images are of a same sensor view.
19. A system for detection of likelihood of malignancy in an anatomical image of a patient for planning treatment thereof, comprising:
at least one hardware processor executing a code for:
determining a likelihood of malignancy when a model fed by a pair of images of a same patient wrongly identifies the pair of images as not belonging to the same patient, when in fact the pair of images are of the same patient, by:
receiving the pair of images of the same patient, wherein the pair of images comprises anatomically corresponding anatomical images each depicting internal anatomical structures of the patient;
feeding the pair of images into the model,
outputting by the model, an indication of whether the pair of images are of same patient or not; and
generating an indication of likelihood of malignancy when the model wrongly outputs that the pair of images are not of the same patient, wherein treatment of the patient is planned according to the indication of likelihood of malignancy.
20. The system of claim 19 , wherein the at least one hardware processor further executes a code for:
providing a set of training images including pairs of anatomically corresponding anatomical images of a plurality of sample patients, wherein each pair of images are either of a same patient or from two different patients; and
training the model using the training images to identify whether or not a pair of anatomically corresponding anatomical images are of a same target patient.