Method and system for acquisition of fluorescence images of live-cell biological samples
Inventors
Jackson, Timothy • Sjögren, Rickard • Edlund, Christoffer • FORSGREN, Edvin
Assignees
Sartorius Stedim Data Analytics AB • Essen Bioscience Inc • Sartorius Bioanalytical Instruments Inc
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11422355-B2
Publication Date
2022-08-23
Expiration Date
Abstract
A method is disclosed for acquiring a single, in-focus two-dimensional projection image of a live, three-dimensional cell culture sample, with a fluorescence microscope. One or more long-exposure “Z-sweep” images are obtained, i.e. via a single or series of continuous acquisitions, while moving the Z-focal plane of a camera through the sample, to produce one or more two-dimensional images of fluorescence intensity integrated over the Z-dimension. The acquisition method is much faster than a Z-stack method, which enables higher throughput and reduces the risk of exposing the sample to too much fluorescent light. The long-exposure Z-sweep image(s) is then input into a neural network which has been trained to produce a high-quality (in-focus) two-dimensional projection image of the sample. With these high-quality projection images, biologically relevant analysis metrics can be obtained to describe the fluorescence signal using standard image analysis techniques, such as fluorescence object count and other fluorescence intensity metrics (e.g., mean intensity, texture, etc.).
Core Innovation
The invention generates an in-focus two-dimensional projection image of a fluorescence image of a three-dimensional live-cell sample. One or more long exposure images are acquired by moving a focal plane of a camera through the sample in a Z direction, so that the camera integrates fluorescence intensity from the sample over a Z-dimension. The long exposure images are supplied to a neural network model trained from a plurality of training images, and the trained model generates the in-focus two-dimensional projection image.
The invention further includes training a neural network to generate the two-dimensional projection image using a training set. The training set comprises one or more long exposure images of a three-dimensional live-cell training sample obtained by moving a focal plane of a camera through the training sample in a Z direction to integrate fluorescence intensity over the Z-dimension, together with a ground truth image obtained from a set of images obtained at a different Z focal plane position and combined using a projection algorithm into a two-dimensional projection image.
A live-cell imaging system is provided to perform the generation of the two-dimensional projection image. The system includes a fluorescence microscope with one or more excitation light sources, objective lenses, and a camera to obtain one or more fluorescence images from the three-dimensional sample held within a sample holding device, and a motor system configured to move in a Z direction so that the camera obtains one or more long exposure images by integrating fluorescence intensity over the Z-dimension as the focal plane moves through the sample. A processing unit with a trained neural network model generates the two-dimensional projection image from the one or more long exposure images.
Claims Coverage
The independent claims are 1, 12, 18, and 27, collectively covering four inventive features. They cover Z-direction focal-plane scanning with Z-integration via long exposure, neural-network generation of an in-focus 2D projection, generation of paired training sets using long-exposure Z-integrated images and 2D projection ground truth from different Z focal plane positions, and a corresponding live-cell imaging system with Z motor movement and neural-network processing.
In-focus 2D projection from Z-integrated long exposure fluorescence images using a trained neural network
A method for generating an in-focus two-dimensional projection image of a fluorescence image of a three-dimensional live-cell sample by acquiring one or more long exposure images via moving a focal plane of a camera through the sample in a Z direction to integrate fluorescence intensity over a Z-dimension, supplying the long exposure images to a neural network model trained from a plurality of training images, and generating the in-focus two-dimensional projection image using the trained neural network model.
Training a neural network using Z-integrated long exposure images paired with 2D projection ground truth
A method for training a neural network to generate a two-dimensional projection image by obtaining a training set with one or more long exposure images of a three-dimensional live-cell training sample obtained by moving a focal plane of a camera through the training sample in a Z direction to integrate fluorescence intensity over a Z-dimension and a ground truth image obtained from a set of images at a different Z focal plane position combined using a projection algorithm into a two-dimensional projection image, and conducting a model training procedure using the training set to generate a trained neural network.
Live-cell imaging system with Z-direction motor movement and neural-network generation of 2D projections
A live-cell imaging system including a fluorescence microscope with a motor system configured to move the fluorescence microscope relative to the sample holding device in a Z direction such that the camera obtains one or more long exposure images by moving a focal plane through the sample in the Z direction to integrate fluorescence intensity over a Z-dimension, and a processing unit including a trained neural network model for generating the two-dimensional projection image from the one or more long exposure images.
Training-set generation from Z-integrated long exposure images and ground-truth 2D projections
A method for generating a training set for training a neural network by acquiring one or more long exposure fluorescence images of a three-dimensional training sample by moving a focal plane of a camera through the sample in a Z direction to integrate fluorescence intensity over a Z-dimension, generating a ground truth image of the same training sample from one or more different images obtained by the camera, repeating for a multitude of different training samples, and supplying the acquired images as the training set.
Across the independent claims, the approach is to integrate fluorescence intensity over a Z-dimension using long exposure images obtained by moving a camera focal plane through a three-dimensional sample, and then generate an in-focus two-dimensional projection image using a neural network trained with ground-truth two-dimensional projections derived from different Z focal plane positions via a projection algorithm.
Stated Advantages
Documented Applications
No documented applications found
Interested in licensing this patent?