Speckle aware clinical decision support system for retinal diseases
OwnerVedaant Agarwalinventor
Life of this patent
Estimating the term…
Abstract
A clinical decision support system for retinal diseases processes OCT retinal scan images to enhance certain aspects of speckle, which has been found to include relevant information that can be used to enhance the diagnosis and classification of the severity and progression of any disease found in the retina scan image. A copy of the original OCT retinal scan image is processed using a shifted window transformer block to enhance the relevant speckle information, and combine it with the original image, which is then evaluated by a trained system to classify the retinal features, including the speckle.
Claims18
An unsupervised computer-implemented method of estimating retinal-disease severity from an OCT scan comprising:
- generating, with a trained neural network, a fixed-length feature embedding of the OCT scan;
- storing embeddings of a plurality of scans that are labeled as typical;
- determining a centroid of the stored embeddings;
- computing a Euclidean distance between the fixed-length feature embedding of the scan and the centroid;
- partitioning a distribution of Euclidean distances into quartile ranges;
- and
- assigning to the scan a severity label selected from mild, moderate I, moderate II, and severe according to a quartile range in which the Euclidean distance falls.
An unsupervised computer-implemented method of estimating retinal-disease severity and progression from an OCT scan comprising:
- generating, with a trained neural network, a fixed-length feature embedding of the OCT scan;
- storing embeddings of a plurality of scans that are labeled as normal;
- determining a centroid of the stored embeddings by computing an arithmetic mean thereof,
- computing a Euclidean distance between the fixed-length feature embedding of the scan and the centroid;
- partitioning a distribution of Euclidean distances, computed from embeddings of scans labeled with a disease, into quartile ranges;
- assigning to the scan a severity label selected from mild, moderate I, moderate II, and severe according to a quartile range in which the Euclidean distance falls;
- and
- calculating a percent-deviation value for the OCT scan by dividing the Euclidean distance by an average of Euclidean distances for scans labeled with a disease and multiplying the quotient by one hundred.
An integrated optical coherence tomography (OCT) analytics system for estimating retinal-disease severity, comprising:
- a memory that stores computer-executable instructions defining a severity-estimation module;
- and
- at least one processor operatively coupled to the memory, the at least one processor being configured, upon execution of the computer-executable instructions, to:
- generate, with a trained neural network, a fixed-length feature embedding of an acquired OCT scan of a retina;
- access stored embeddings of a plurality of OCT scans labeled as typical;
- determine a centroid of the stored embeddings of the plurality of OCT scans labeled as typical;
- compute a Euclidean distance between the fixed-length feature embedding of the acquired OCT scan and the centroid;
- partition a distribution of Euclidean distances associated with OCT scans not labeled as typical into quartile ranges;
- and
- assign to the acquired OCT scan a severity label selected from mild, moderate I, moderate II, and severe according to a quartile range in which the computed Euclidean distance falls.
Dependencies are inferred from the claim text.