Retinal disease biomarker prediction by stacking slices of 3D OCT image to reshape into 2D image and applying trained feature extractor and CNN

Patent metadata

Field	Value
Publication number	US-12499535-B2
Application number	US-202117758625-A
Country	US
Kind code	B2
Filing date	Jan 25, 2021
Priority date	Jan 24, 2020
Publication date	Dec 16, 2025
Grant date	Dec 16, 2025

Abstract

Deep learning methods and systems for detecting biomarkers within optical coherence tomography volumes using such deep learning methods and systems are provided. Embodiments predict the presence or absence of clinically useful biomarkers in OCT images using deep neural networks. The lack of available training data for canonical deep learning approaches is overcome in embodiments by leveraging a large external dataset consisting of foveal scans using transfer learning. Embodiments represent the three-dimensional OCT volume by “tiling” each slice into a single two dimensional image, and adding an additional component to encourage the network to consider local spatial structure. Methods and systems, according to embodiments are able to identify the presence or absence of AMD-related biomarkers on par with clinicians. Beyond identifying biomarkers, additional models could be trained, according to embodiments, to predict the progression of these biomarkers over time.

First claim

The invention claimed is:

1 . A method to predict retinal disease biomarkers comprising:
obtaining at least one three-dimensional optical coherent tomography image using a computer system; reshaping the three-dimensional optical coherent tomography image into a two-dimensional image by stacking a plurality of slices of said three-dimensional image on top of one another using the computer system; applying a pre-trained feature extractor, wherein the pre-trained feature extractor independently operates on each of the plurality of slices; applying a convolutional neural network operating across the plurality of slices, wherein the convolutional neural network produces a feature vector; and generating an output of biomarker prediction, wherein the prediction is a transformation of the feature vector.

2 . The method of claim 1 , wherein the final output is at least one biomarker predictive of age-related macular degeneration (AMD), AMD subtype, AMD progression, or AMD deterioration.

3 . The method of claim 2 , wherein the AMD subtype is selected from the group consisting of early AMD, intermediate AMD, wet AMD, and geographic atrophy.

4 . The method of claim 2 , wherein the at least one biomarker is intraretinal hyperreflective feature over drusen, intraretinal hyperreflective feature over non drusen, intraretinal cystoid spaces, drusenoid pigment epithelial detachment, subretinal tissue, subretinal drusenoid deposits, incomplete retinal pigment epithelial and outer retinal atrophy (IRORA), complete retinal pigment epithelial and outer retinal atrophy (cRORA), and any combinations thereof.

5 . The method of claim 1 , further comprising slice pooling by sampling a mean value and a maximum value from every slice.

6 . The method of claim 1 , wherein the convolutional neural network is a 1D convolutional neural network.

7 . The method of claim 1 , wherein feature vector transformation is a decision layer comprising at least two fully connected layers.

8 . The method of claim 7 , wherein the layers are connected with a rectified linear unit between each layer.

Assignees

• Univ California
• Doheny Eye Inst

Inventors

• Halperin Eran
• Rakocz Nadav
• Chiang Jeffrey
• Gupta Muneeswar
• Sadda Srinivas

Classifications

• A61B5/6821

  Eye · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/30041

  Eye; Retina; Ophthalmic · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/10101

  Optical tomography; Optical coherence tomography [OCT] · CPC title