Ai-driven signal enhancement of low-resolution images
AssigneeIllumina, Inc.
InventorsYina Wang, Merek Siu, Rishi Verma, Jeffrey Gau, Stanley Hong, Anindita Dutta, Aathavan Karunakaran, Simon Prince
Life of this patent
Estimating the term…
Abstract
AI-driven enhancement of low-resolution images enables sequencing of a nucleic acid material under low-resolution conditions with requisite accuracy and improved cost and operational efficiency. A training set of images obtained under actual or simulated high-resolution and low-resolution conditions is used to train a model to recover enhanced images, as if taken under high resolution conditions, from unenhanced images taken under low resolution conditions.
Claims26
A method of training a model to produce an enhanced resolution image from a low-resolution image in a sequencing operation, the method including:
- obtaining high-resolution training image data from one or more images of a target for each of a plurality of targets, wherein the high-resolution target image data yields a predefined set of high-resolution images for a number of cycles of a sequencing run, and wherein the target comprises a biological sample;
- generating low-resolution training image data from the one or more images captured of the target during the number of cycles of the sequencing run, wherein the one or more images are synthetically degraded to simulate one or more low-resolution conditions, and wherein the low-resolution training image data yields a predefined set of low-resolution images corresponding pairwise to at least a portion of the predefined set of high-resolution images for the number of cycles of the sequencing run;
- training parameters of the model using the low-resolution training image data and the high-resolution target image data,
- wherein the parameters are trained to extract features of the low-resolution training image data for predictively generating a set of enhanced images of comparable resolution to corresponding high-resolution images of the set of high-resolution images, and
- outputting the trained parameters of the model,
- wherein the trained parameters are configurable into one or more trained filters for recovering an enhanced production image from one or more low-resolution production images of a production target, and
- wherein the enhanced production image has an optical resolution above a prescribed threshold required in a sequencing operation.
A method of producing an enhanced resolution images from low-resolution images captured in a sequencing operation in production, the method including:
- obtaining low-resolution target image data from one or more low-resolution images of a target captured by one or more sensors of an optic imaging system;
- inputting the low-resolution target image data into a model, wherein the model comprises one or more trained filters for recovering an enhanced resolution image from the one or more low-resolution images;
- wherein the one or more trained filters are configured from parameters refined from training data comprising:
- high-resolution target image data from one or more images of a target for each of a plurality of targets captured under high resolution, wherein the high-resolution target image data yields a set of high-resolution training images;
- low-resolution training image data processed from the set of high-resolution training images such that the set of high-resolution training images are synthetically degraded to simulate one or more low-resolution conditions, wherein the low-resolution training image data yields a set of low-resolution training images corresponding pairwise to the set of high-resolution training images;
- outputting the enhanced resolution image for base calling.
Dependencies are inferred from the claim text.