Granted Patent B2

US 12,685,437 · App. 17/604,251 · Granted Jul 21, 2026

Inventors: Shelley Boyd (Toronto, CA); Natalie Pankova (Ottawa, CA); Nehad Hirmiz (Toronto, CA); Huiyuan Liang (Toronto, CA)

Assignee: Tracery Ophthalmics Inc.

A61B3/0025A61B3/12G06T7/0012G06V10/771G06V10/7715G06V10/774G06V10/82G06V40/19G06V40/193G06V40/197G06T2207/10048G06T2207/10101G06T2207/20081G06T2207/20084G06T2207/30041G06V2201/032

Abstract

Claims (34)

1 . A computer system comprising:

a processor;

a memory in communication with the processor, the memory storing instructions that, when executed by the processor cause the processor to:

at a training phase,

receive training data corresponding to a plurality of ocular images, wherein the plurality of ocular images of the training data correspond to a plurality of imaging modalities, said plurality of imaging modalities including delayed near-infrared analysis (DNIRA) and at least one imaging modality other than DNIRA;

generate delta image data based on a difference between said DNIRA images and images of said at least one imaging modality other than DNIRA;

perform feature extraction and feature selection to generate features based on the training data and the delta image data to build a pattern recognition model;

at a classification phase,

receive a plurality of ocular images corresponding to a plurality of imaging modalities;

classify features of the plurality of ocular images using the pattern recognition model.

2 . The computer system of claim 1 , wherein the pattern recognition model is at least one of a convolutional neural network, machine learning, decision trees, logistic regression, principal components analysis, naive Bayes model, support vector machine model, and nearest neighbor model.

3 . The computer system of claim 1 , wherein the feature extraction generates a masked image of defined shapes.

4 . The computer system of claim 3 , wherein the feature selection is based on at least one of focality of the defined shapes, a number of focal points per unit area of the defined shapes, and a square root of an area of the defined shapes.

5 . The computer system of claim 1 , wherein the features are defined by areas of hypofluorescence.

6 . The computer system of claim 1 , wherein the features are defined by areas of hyperfluorescence.

7 . The computer system of claim 1 , wherein the training phase further comprises building a pattern recognition model for each of the plurality of imaging modalities.

8 . The computer system of claim 1 , wherein the plurality of ocular images comprises a cross-section image.

9 . The computer system of claim 1 , wherein the plurality of ocular images comprises an en face image.

10 . The computer system of claim 1 , wherein the training phase further comprises registering the plurality of ocular images to a common coordinate system.

11 . The computer system of claim 1 , wherein the training phase further comprises cross-modal fusion of the plurality of ocular images to a common coordinate system.

12 . The computer system of claim 1 , wherein the plurality of imaging modalities other than DNIRA comprise at least one of infra-red reflectance (IR), confocal scanning laser ophthalmoscopy (cSLO), fundus autofluorescence (FAF), color fundus photography (CFP), optical coherence tomography (OCT), OCT-angiography, fluorescence lifetime imaging (FLI).

13 . The computer system of claim 1 , wherein the memory stores further instructions that, when executed by the processor cause the processor to:

generate a cross-section segmentation map corresponding to an en face region of an eye, each segment of the cross-section segmentation map corresponding to a cross-section image at that region of the eye;

classify each segment of the cross-section segmentation map as a phenotype of one of normal, drusen, retinal pigment epithelium detachments (RPEDs), pseudodrusen geographic atrophy, macular atrophy, or neovascularization based at least in part on classification of the cross-section image corresponding to that segment using the pattern recognition model.

14 . The computer system of claim 1 , wherein the plurality of ocular images comprises multiple cross-section images corresponding to multiple time points and the memory stores further instructions that, when executed by the processor cause the processor to:

generate, for each of the multiple time points, a cross-section segmentation map corresponding to an en face region of an eye, each segment of the cross-section segmentation map corresponding to a cross-section image at that region of the eye;

classify each segment of each cross-section segmentation map as a phenotype of tissue state of one of normal, drusen, retinal pigment epithelium detachments (RPEDs), pseudodrusen geographic atrophy, macular atrophy, or neovascularization, based at least in part on classification of the cross-section image corresponding to that segment using the pattern recognition model; and

generate a time series data model based on the cross-section segmentation map at each of the multiple time points.

15 . The computer system of claim 14 , wherein the time series data model is based at least in part on identified changes in the cross-section segmentation maps over time.

16 . The computer system of claim 14 , wherein the time series data model is used to generate a visual representation of disease progression.

17 . The computer system of claim 14 , wherein the time series data model is based at least in part on elapsed time between the multiple time points.

18 . The computer system of claim 1 , wherein the features selected comprise phenotypes of a user associated with the plurality of ocular images.

19 . The computer system of claim 1 , wherein the memory stores further instructions that, when executed by the processor cause the processor to: correlate the features with stage or grade variants of blinding eye disease including Age Related Macular Degeneration (AMD), monogenic eye disease, inherited eye disease and inflammatory eye disease.

20 . The computer system of claim 1 , wherein said at least one imaging modality other than DNIRA comprises fundus autofluorescence (FAF), and wherein generating said delta image data comprises a delta analysis comprising DNIRA subtracting FAF.

Assignments (1)

Continuity (1)

References Cited (147)