System and method for dense-stochastic-sampling imaging
View Patent ↗Embodiments of the present invention are directed to imaging technologies, and, in particular, to an imaging system that detects relatively weak signals, over time, and that uses the detected signals to determine the positions of signal emitters. Particular embodiments of the present invention are directed to methods and systems for imaging fluorophore-labeled samples in order to produce images of the sample at resolutions significantly greater than the diffraction-limited resolution associated with optical microscopy. Embodiments of the present invention employ overlapping-emitter-image disambiguation to allow data to be collected from densely arranged emitters, which significantly decreases the data-collection time for producing intermediate images as well as the number of intermediate images needed to computationally construct high-resolution final images. Additional embodiments of the present invention employ hierarchical image-processing techniques to further resolve and interpret disambiguated images.
1. An imaging system comprising:
a light-detection subsystem that illuminates a sample or a portion of the sample to activate light emitters within the sample, excites light emission from the light emitters, and records, over a period of time, intermediate output images of light emitters within the sample or the portion of the sample;
an image-generation subsystem that processes the intermediate output images to identify and record initially determined positions of the light emitters within the sample or the portion of the sample and that iteratively computes computed output images from light-emitter-position models in order to disambiguate overlapping light-emitter images in the intermediate output images to produce a composite image that includes indications of the positions of the light emitters within the sample or the portion of the sample; and
an image-interpretation subsystem that iteratively fits geometric primitives to the composite image, produced by the image-generation subsystem, in order to produce one or more refined and interpreted images of the sample or the portion of the sample.
2. The imaging system of claim 1 wherein the light-detection subsystem comprises a fluorescence-microscopy optical subsystem.
3. The imaging system of claim 2 wherein the light emitters are fluorophores which are activated by illumination by exposure to light of a first frequency and which are excited to emit fluorescent light by exposure to light of a second frequency.
4. The imaging system of claim 3 wherein the light emitters are fluorophores which are activated exposure to light of a first frequency and which are excited to emit fluorescent light by exposure to light of a second frequency.
5. The imaging system of claim 1 wherein the image-generation subsystem processes each intermediate image by:
filtering noise from the intermediate image to produce a noise-filtered intermediate image;
selecting local maxima from the noise-filtered intermediate image;
selecting light emitters from the selected local maxima; and
disambiguating overlapping light-emitter images in order to add any additional, detected light emitters to the selected light-emitter positions.
6. The imaging system of claim 5 wherein the image-generation subsystem filters noise from the intermediate image by applying a Gaussian filter to the intermediate images.
7. The imaging system of claim 5 wherein the image-generation subsystem selects local maxima from the noise-filtered intermediate image by selecting pixels or voxels that have higher associated intensity values than all adjacent pixels or voxels in the intermediate image.
8. The imaging system of claim 5 wherein the image-generation subsystem selects light emitters from the selected local maxima by:
computing a spot metric for each local maximum as the product of a curvature computed over a neighborhood of the local maximum in the image and the mean intensity of a neighborhood of the local maximum; and
selecting, as light emitters, those local maxima with computed spot metrics greater than a threshold value.
9. The imaging system of claim 5 wherein the image-generation subsystem disambiguates overlapping light-emitter images in order to add any additional, detected light emitters to the selected light emitters by:
partitioning the light emitters into clusters that each comprises one or more initial light emitters; and
for each cluster,
iteratively
inserting a partner to a selected initial light emitter into the cluster,
fitting initial light-emitter positions and the partner position to observed intensities for the cluster,
computing a computed image for the cluster,
computing a residual statistic from the computed image and observed image, and
adding the partner to the cluster when a statistical test indicates that the residual statistic computed for the cluster including the partner is better than the residual statistic computed for the cluster prior to inserting the partner.
10. The imaging system of claim 9 wherein computing a computed image for the cluster further includes modeling each light-emitter and the partner as parameterized Gaussian intensity distributions.
11. The imaging system of claim 1 wherein each intermediate image comprises a number of two-dimensional images obtained over a range of sample position relative to the optical axis of the imaging system.
12. The imaging system of claim 1 wherein the image-interpretation subsystem iteratively fits geometric primitives to the composite image, produced by the image-generation subsystem, in order to produce one or more refined and interpreted images of the sample or the portion of the sample by:
iteratively
selecting a set of geometrical elements for a currently considered hierarchical processing level;
recursively fitting geometrical elements from the set of geometrical elements to an intermediate, refined image from a preceding iteration or, in a first iteration, to the composite image produced by the image-generation subsystem to produce one or more candidate refined images; and
selecting candidate refined images for processing in a subsequent iteration or, in a final iteration, as the one or more refined and interpreted images.
13. The imaging system of claim 12 wherein the image-interpretation subsystem selects a set of geometrical elements for a currently considered hierarchical processing level by:
selecting geometrical elements with sizes and complexities appropriate to the currently considered hierarchical processing level; and
selecting, from among those geometrical elements with sizes and complexities appropriate to the currently considered hierarchical processing level, geometrical elements compatible with structures and features known to occur within the sample.
14. The imaging system of claim 12 wherein the image-interpretation subsystem carries out a geometrical-element-fitting search in order to consider a large number of possible mappings of geometrical elements to the intermediate, refined image from a preceding iteration or, in the first iteration, to the composite image produced by the image-generation subsystem.
15. The imaging system of claim 14 wherein candidate images are evaluated by the extent to which features of the intermediate, refined image from a preceding iteration or, in the first iteration, of the composite image produced by the image-generation subsystem, are covered, or contained within, the geometrical elements fitted to them in the current iteration as well as by the likelihood that features in the candidate images correspond to structures and features known to occur in the sample.
16. The imaging system of claim 14 wherein a fit of a geometrical element to one or more features of the intermediate, refined image from a preceding iteration or, in the first iteration, to the composite image produced by the image-generation subsystem, is evaluated by the extent to which features of the intermediate, refined image from a preceding iteration or, in the first iteration, of the composite image produced by the image-generation subsystem, are covered, or contained within, the geometrical element as well as by the likelihood that any intermediate feature in the candidate image produced by fitting the geometrical element to the one or more features corresponds to structures and features known to occur in the sample.
17. A method for producing an image, the method comprising:
repeatedly
illuminating a sample or a portion of the sample to activate light emitters within the sample,
exciting light emission from the light emitters, and
recording, over a period of time, intermediate output images of light emitters within the sample or the portion of the sample;
processing the intermediate output images to identify and record initially determined positions of the light emitters within the sample or the portion of the sample;
iteratively computing computed output images from light-emitter-position models in order to disambiguate overlapping light-emitter images in the intermediate output images to produce a composite image that includes indications of the positions of the light emitters within the sample or the portion of the sample; and
iteratively fitting geometric primitives to the composite image, produced by the image-generation subsystem, in order to produce one or more refined and interpreted images of the sample or the portion of the sample.
18. The method of claim 17 wherein processing the intermediate output images to identify and record initially determined positions of the light emitters within the sample or the portion of the sample further includes:
filtering noise from the intermediate image to produce a noise-filtered intermediate image;
selecting local maxima from the noise-filtered intermediate image;
selecting light emitters from the selected local maxima; and
disambiguating overlapping light-emitter images in order to add any additional, detected light emitters to the selected light-emitter positions.
19. The method of claim 17 wherein iteratively computing computed output images from light-emitter-position models in order to disambiguate overlapping light-emitter images in the intermediate output images to produce a composite image that includes indications of the positions of the light emitters within the sample or the portion of the sample further includes:
partitioning the light emitters into clusters that each comprises one or more initial light emitters; and
for each cluster,
iteratively
inserting a partner to a selected initial light emitter into the cluster,
fitting initial light-emitter positions and the partner position to observed intensities for the cluster,
computing a computed image for the cluster,
computing a residual statistic from the computed image and observed image, and
adding the partner to the cluster when a statistical test indicates that the residual statistic computed for the cluster including the partner is better than the residual statistic computed for the cluster prior to inserting the partner.
20. The method of claim 17 wherein iteratively fitting geometric primitives to the composite image, produced by the image-generation subsystem, in order to produce one or more refined and interpreted images of the sample or the portion of the sample further includes:
iteratively
selecting a set of geometrical elements for a currently considered hierarchical processing level;
recursively fitting geometrical elements from the set of geometrical elements to an intermediate, refined image from a preceding iteration or, in a first iteration, to the composite image produced by the image-generation subsystem to produce one or more candidate refined images; and
selecting candidate refined images for processing in a subsequent iteration or, in a final iteration, as the one or more refined and interpreted images.