Framework for detecting discrepancies between images and image interpretations

1 . A method comprising:

receiving, at a discrepancy determination system, source data comprising an image representing a portion of a human subject and textual data associated with the image;

executing, by the discrepancy determination system and using the image as first input, a machine-learned image processing model to determine first values for a set of medical conditions represented in the image;

executing, by the discrepancy determination system and using the textual data as second input, a machine-learned text processing model to determine second values for the set of medical conditions represented in the image;

determining, by the discrepancy determination system, a first discrepancy value based on a difference between a first value of the first values associated with a medical condition of the set of medical conditions and a corresponding second value of the second values associated with the medical condition of the set of medical conditions;

determining, by the discrepancy determination system, that the discrepancy value meets or exceeds a first discrepancy threshold;

executing, by the discrepancy determination system and using the image data as the first input, the machine-learned image processing model to determine third values for the set of medical conditions represented in the image;

executing, by the discrepancy determination system and using the image data as the first input, the machine-learned image processing model to determine fourth values for the set of medical conditions represented in the image;

determining, by the discrepancy determination system, a second discrepancy value based on a difference between a third value of the third values associated with the medical condition of the set of medical conditions and a corresponding fourth value of the fourth values associated with the medical condition of the set of medical conditions;

determining, by the discrepancy determination system, that the second discrepancy value is below a second discrepancy threshold; and

transmitting, by the discrepancy determination system to a computing device for presentation on an interface and based on determining that the first discrepancy value meets or exceeds the first discrepancy threshold, an indication of the medical condition of the set of medical conditions and the first discrepancy value but excluding the second discrepancy value from the indication based at least in part on determining that the second discrepancy is below the second discrepancy threshold.

2 . The method of claim 1 , wherein the image comprises one of an X-ray, a magnetic resonance imaging (MRI) image, a computed tomography (CT) scan image, a computed axial tomography (CAT) scan image, an ultrasound image, a histology image, optical coherence tomography (OCT) image, or a fundus image.

3 . The method of claim 1 wherein the first value and the second value each indicate one of:

a likelihood of a presence of the medical condition; or

a risk associated with the medical condition.

4 . A method comprising:

receiving, by one or more computing devices, first data of a first modality associated with an image and second data of a second modality associated with the image;

executing, by the one or more computing devices and using the first data as first input, at least one of a first prediction generation component or a second prediction generation component to determine a first value for a first parameter of the image;

executing, by the one or more computing devices and using the second data as second input, at least one of the first prediction generation component or the second prediction generation component to determine a second value for the first parameter of the image;

determining, by the one or more computing devices, a first discrepancy between the first value and the second value that exceeds a first discrepancy threshold;

executing, by the one or more computing devices and using the first data as the first input, at least one of the first prediction generation component or the second prediction generation component to determine a third value for a second parameter of the image;

executing, by the one or more computing devices and using the second data as the second input, at least one of the first prediction generation component or the second prediction generation component to determine a fourth value for the second parameter of the image;

determining, by the one or more computing devices, a second discrepancy between the third value and the fourth value is below a second discrepancy threshold; and

transmitting, by the one or more computing devices to a computing device for presentation on an interface and based at least in part on determining the first discrepancy and the second discrepancy, an indication of the first discrepancy by excluding the second discrepancy from the indication based at least in part on determining that the second discrepancy is below the second discrepancy threshold.

5 . The method of claim 4 , wherein the first data comprises an image of a portion of a human subject.

6 . The method of claim 4 , wherein the first value comprises one of a representation of a probability of a medical condition or a representation of a level of risk of the medical condition.

7 . The method of claim 4 , wherein at least one of the first prediction generation component or the second prediction generation component is one or more of:

a machine-learned image processing model, or

a machine-learned text processing model.

8 . The method of claim 4 , wherein at least one of the first prediction generation component or the second prediction generation component is one or more of:

a machine-learned image processing model, or

an image metadata processing component.

9 . The method of claim 4 , wherein the second data comprises one or more medical codes indicating one or more medical conditions.

10 . The method of claim 4 , wherein the first parameter comprises a fault in a physical component represented in the image.

11 . The method of claim 4 , wherein the first parameter comprises one of:

a demographic attribute of a subject of the image; or

a position of the subject of the image relative to an imaging system that captured the image.

12 . The method of claim 4 , wherein the first data comprises data representing one of an X-ray, a magnetic resonance imaging (MRI) image, a computed tomography (CT) scan image, a computed axial tomography (CAT) scan image, an ultrasound image, a histology image, optical coherence tomography (OCT) image, or a fundus image.

13 . A system comprising:

one or more processors; and

one or more computer-readable media storing computer-executable instructions that, when executed by the one or more processors, cause a discrepancy determination system to perform operations comprising:

receiving first data of a first modality associated with a human subject and second data of a second modality associated with the human subject;

executing, using the first data as first input, at least one of a first prediction generation component or a second prediction generation component to determine a first value for a first parameter associated with the human subject;

executing, using the second data as second input, at least one of the first prediction generation component or the second prediction generation component to determine a second value for the first parameter associated with the human subject;

determining a first difference between the first value and the second value that satisfies a first discrepancy criterion;

executing, using the first data as the first input, at least one of the first prediction generation component or the second prediction generation component to determine a third value for a second parameter associated with the human subject;

executing, using the second data as the second input, at least one of the first prediction generation component or the second prediction generation component to determine a fourth value for the second parameter associated with the human subject;

determining a second difference between the third value and the fourth value does not satisfy a second discrepancy criterion; and

transmitting, to a computing device for presentation on an interface and based at least in part on determining that the first difference between the first value and the second value satisfies the discrepancy criterion, an indication of the first difference by excluding the second difference from the indication based at least in part on determining that the second difference between the third value and the fourth value does not satisfy the second discrepancy criterion.

14 . The system of claim 13 , wherein the first parameter comprises a medical condition represented in an image associated with the human subject.

15 . The system of claim 14 , wherein:

the first value is a first representation of a first risk of the human subject having the medical condition; and

the second value is a second representation of a second risk of the human subject having the medical condition.

16 . The system of claim 15 , wherein transmitting the indication of the first difference comprises transmitting instructions to present a graphical representation of a risk scale comprising a first indication of the first risk and a second indication of the second risk.

17 . The system of claim 13 , wherein transmitting the indication of the first difference comprises determining that the first parameter is represented in a parameter relevance filter.

18 . The system of claim 13 , wherein one of the first data or the second data comprises digital imaging and communications in medicine (DICOM) data associated with an image associated with the human subject.

19 . The system of claim 13 , wherein the first data comprises data representing one of an X-ray, a magnetic resonance imaging (MRI) image, a computed tomography (CT) scan image, a computed axial tomography (CAT) scan image, an ultrasound image, a histology image, optical coherence tomography (OCT) image, or a fundus image.

20 . The system of claim 13 , wherein at least one of the first prediction generation component or the second prediction generation component is one or more of:

a machine-learned image processing model, or

a machine-learned text processing model.