Automated disease identification based on ophthalmic images
An example method includes identifying at least one image of an eye of a patient. The method further includes detecting, by a first computing model, at least one first feature in the at least one image and detecting, by a second computing model, at least one second feature in the at least one image. Further, using a third computing model that is different than the first computing model or the second computing model, the method includes identifying a likelihood that the patient has one or more diseases consistent with the at least one feature and the at least one second feature. A recommendation for care of the patient is generated based on the likelihood.
1. A system comprising:
a medical imaging device configured to generate at least one image of at least one eye of a patient;
at least one processor communicatively coupled to the medical imaging device; and
memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
detecting, by at least one trained machine learning (ML) model, a set of features in an image of an eye of a patient;
determining, based on a first subset of features included in the set of features, whether the patient exhibits a first disease of multiple diseases;
determining, based on a second subset of features included in the set of features, whether the patient exhibits a second disease of the multiple diseases; and
generating a recommendation for care of the patient based on whether the patient exhibits the first disease or the second disease; and
an output device configured to output the recommendation.
2. The system of claim 1 , wherein the at least one image comprises one or more of an optical coherence tomography (OCT) image, a slit lamp image, a fundus image, a fluorescence angiogram, a color fundus photography (CFP) image, a fluorescein angiography (FA) image, an indocyanine green (ICG) angiography image, a fundus autofluorescence (FAF) image, or a retinal image.
3. The system of claim 1 , wherein the operations further comprise:
determining at least one of a number, a size, a location, or a shape of at least one feature of the first subset of features,
the at least one processor determining whether the patient exhibits the first disease based on the at least one of the number, the size, the location, or the shape of the at least one feature.
4. The system of claim 1 , wherein the operations further comprise:
detecting a landmark depicted in the image, the landmark comprising at least one of a macula, an optic disc (OD), or a fovea of the at least one eye; and
determining a distance between at least one feature of the first subset of features and the landmark,
the at least one processor determining whether the patient exhibits the first disease based on the distance between the at least one feature and the landmark.
5. The system of claim 1 , wherein:
a first feature of the first subset of features comprises a microaneurysm, a hemorrhage, drusen, exudate, edema, a cup/disc ratio (CDR), focal arteriolar narrowing, arterio-venous nicking, a cotton wool spot, an embolus, a red spot, retinal whitening, a Hollenhorst plaque, a Roth spot, a microinfarct, coagulated fibrin, new vessels elsewhere (NVE), a vitreous hemorrhage (VH), a pre-retinal hemorrhage (PRH), new vessels on a disc (NVD), venous beading, or an intraretinal microvascular abnormality (IRMA), and
a second feature of the second subset of features comprises the microaneurysm, the hemorrhage, the drusen, the exudate, the edema, the CDR, the focal arteriolar narrowing, the arterio-venous nicking, the cotton wool spot, the embolus, the red spot, the retinal whitening, the Hollenhorst plaque, the Roth spot, the microinfarct, the coagulated fibrin, the NVE, the VH, the PRH, the NVD, the venous beading, or the IRMA,
wherein the first feature is different than the second feature.
6. The system of claim 1 , wherein:
the at least one trained ML model comprises a set of ML models including at least one of: a first supervised learning model, a first unsupervised learning model, or a first semi-supervised learning model, and
each ML model of the set of ML models is characterized by a same ML architecture.
7. The system of claim 1 , wherein:
determining whether the patient exhibits the first disease or the second disease comprises identifying a severity associated with the first disease or the second disease, and
the recommendation is based at least in part on the severity.
8. The system of claim 1 , wherein training a ML model, included in the at least one trained ML model, to detect a particular feature is based on a training dataset identifying a set of images depicting the particular feature.
9. The system of claim 1 , the operations further comprising:
identifying, based on a region associated with the patient, a standard of practice,
the at least one processor determining whether the patient exhibits the first disease or the second disease based on the standard of practice.
10. The system of claim 9 , wherein the recommendation is based on the standard of practice.
11. A system comprising:
at least one processor communicatively coupled to a medical imaging device; and
memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
identifying at least one image of an eye of a patient;
detecting, by at least one trained computing model, a set of features in an image of the at least one image;
determining, by a disease computing model separate from the at least one computing models and based on the set of features, one or more likelihoods, each likelihood associated with whether the patient has a disease of multiple diseases; and
generating a recommendation for care of the patient based on the one or more likelihoods.
12. The system of claim 11 , wherein the at least one image comprises one or more of an optical coherence tomography (OCT) image, a slit lamp image, a fundus image, a fluorescence angiogram, a color fundus photography (CFP) image, a fluorescein angiography (FA) image, an indocyanine green (ICG) angiography image, a fundus autofluorescence (FAF) image, or a retinal image.
13. The system of claim 11 , wherein:
the operations further comprise:
detecting, by a first computing model included in the at least one trained computing model, a landmark depicted in the image, the landmark comprising at least one of a macula, an optic disc (OD), or a fovea the at least one eye; and
determining, by the first computing model, a distance between a respective feature corresponding to the first computing model feature and the landmark,
wherein the one or more likelihoods are determined based on the distance.
14. The system of claim 11 , wherein:
a feature of the set of features comprises a microaneurysm, a hemorrhage, drusen, exudate, edema, a cup/disc ratio (CDR), focal arteriolar narrowing, arterio-venous nicking, a cotton wool spot, an embolus, a red spot, retinal whitening, a Hollenhorst plaque, a Roth spot, a microinfarct, coagulated fibrin, new vessels elsewhere (NVE), a vitreous hemorrhage (VH), a pre-retinal hemorrhage (PRH), new vessels on a disc (NVD), venous beading, or an intraretinal microvascular abnormality (IRMA).
15. The system of claim 11 , wherein:
the at least one trained computing model comprises a set of ML models including at least one of a supervised learning model, a unsupervised learning model, or a semi-supervised learning model, and
the disease computing model comprises a look-up table.
16. The system of claim 11 , wherein:
the operations further comprise identifying data indicative of an electronic medical record (EMR) of the patient,
the at least one processor determining the one or more likelihoods based on the data indicative of the EMR of the patient.
17. The system of claim 11 , wherein the multiple diseases include at least one of diabetic retinopathy (DR), age-related macular degeneration (AMD), diabetic macula edema (DME), retinal vein occlusion (RVO), retinopathy of prematurity (ROP), coronary microvascular dysfunction, hypertensive retinopathy, ischemic optic neuropathy, papilledema, retinal artery occlusion, carotid artery occlusion, human immunodeficiency virus (HIV), acquired immunodeficiency syndrome (AIDS), syphilis, malaria, chicken pox, Lyme disease, leukemia, subacute bacterial endocarditis, sepsis, or anemia.
18. A method, comprising:
generating an image of at least one eye of a patient;
detecting, by at least one trained machine learning (ML) model, a set of features in the image;
determining, by a processor, based on a first subset of features included in the set of features, whether the patient exhibits a first disease of multiple diseases;
determining, by the processor and based on a second subset of features included in the set of features, whether the patient exhibits a second disease of the multiple diseases;
generating, by the processor, a recommendation for care of the patient based on whether the patient exhibits the first disease or the second disease; and
providing, by the processor, an output to an output device, the output indicating the recommendation.
19. The method of claim 18 , further comprising:
determining at least one of a number, a size, a location, or a shape of at least one feature of the first set of features,
the processor determining the first disease is based on the at least one of the number, the size, the location, or the shape of the at least one first feature.
20. The method of claim 18 , further comprising:
detecting a landmark depicted in the image, the landmark comprising at least one of a macula, an optic disc (OD), or a fovea of the at least one eye; and
determining a distance between at least one feature of the first subset of features and the landmark,
the processor determining whether the patient exhibits the first disease based on the distance between the at least one feature and the landmark.
21. A non-transitory computer-readable medium storing instructions which, when executed by a processor, cause the processor to:
receive an image of at least one eye of a patient;
determine, by providing the image as an input to at least one trained machine learning (ML) model, a set of features in the image;
determine, based on a first subset of features included in the set of features, whether the patient exhibits a first disease of multiple diseases;
determine, based on a second subset of features included in the set of features, whether the patient exhibits a second disease of the multiple diseases;
generate a recommendation for care of the patient based on whether the patient exhibits the first disease or the second disease; and
provide an output to an output device, the output indicating the recommendation.
22. The non-transitory computer-readable medium of claim 21 , wherein the instructions further cause the processor to:
detect a landmark in the image, the landmark comprising at least one of a macula, an optic disc (OD), or a fovea of the eye;
determine a distance between at least one feature of the first subset of features and the landmark,
the processor determining whether the patient exhibits the first disease based on the distance between the at least one feature and the landmark.
23. The non-transitory computer-readable medium of claim 21 , wherein training a first ML model included in the at least one trained ML model comprises:
identifying first images of eyes of multiple first patients;
identifying first indications of a first feature depicted in a first subset of the first images; and
providing, as a training dataset, the first subset of the first image and the first indications of the first feature,
wherein the first ML model is trained to detect the first feature using the training dataset.