Patent Application
20210132350
The disclosed embodiments relate to techniques for performing fluorescence microscopy imaging. More specifically, the disclosed embodiments relate to techniques for increasing the depth of field for images acquired through fluorescent microscopy imaging.
To facilitate high resolution in fluorescence microscopy, high numerical aperture (NA) lenses are typically used to acquire the images. Unfortunately, the use of high NA lenses significantly limits the depth of the field of the microscope, which means that only features within an extremely thin focal plane will be in focus, while other features that are not located in the focal plane will be out of focus. For thin-sectioned microscopy slides, this is typically not a problem due to the almost flat topography of the tissue mounted on the glass slide. However, this limited depth of field becomes a problem while imaging non-sectioned tissue sitting on the imaging plane of the microscope. This is due to the extent of tissue surface roughness at a microscopic scale, which makes it almost impossible to have all important features simultaneously in focus.
Researchers have attempted to solve this limited depth of field problem by varying the focus of an imaging device (or varying the distance of the sample from the imaging device) while the image is being acquired so that information is gathered from a number of different imaging planes. The resulting blurry image is then processed through deconvolution to produce a final image, which is in focus across a range of depths of field. (For example, see U.S. Pat. No. 7,444,014, entitled “Extended Depth of Focus Microscopy.” by inventors Michael E. Dresser, et al., issued 28 Oct. 2008.) Unfortunately, this technique requires a significant amount of computation to determine the location of objects in the z dimension, which makes it impractical for a wide range of applications.
Hence, what is needed is a technique for extending the depth of field during high-resolution fluorescence microscopy imaging without the performance problems of existing techniques.
The disclosed embodiments relate to a system that performs microscopy imaging with an extended depth of field. This system includes a stage for holding a sample, and a light source for illuminating the sample, wherein the light source produces ultraviolet light with a wavelength in the 230 nm to 300 nm range to facilitate microscopy with ultraviolet surface excitation (MUSE) imaging. The system also includes an imaging device, comprising an objective that magnifies the illuminated sample, and a sensor array that captures a single image of the magnified sample. The system also includes a controller, which controls the imaging device and/or the stage to scan a range of focal planes for the sample during an acquisition time for the single image. The system additionally includes an image-processing system, which processes the single image using a deconvolution technique to produce a final image with an extended depth of field.
In some embodiments, while scanning the range of focal planes for the sample, the system uses a tunable lens to vary a focus of the imaging device.
In some embodiments, scanning the range of focal planes for the sample involves moving one or more of the following: the sample; the objective; a tube lens, which is incorporated into the imaging device; and the sensor array.
In some embodiments, moving one or more of the sample, the objective, the tube lens or the sensor involves using one or more of: a piezoelectric actuator; a linear actuator, and a voice coil.
In some embodiments, capturing the single image of the sample involves: capturing multiple images of the sample; and combining the multiple images to produce the single image of the sample.
In some embodiments, processing the single image comprises: applying the deconvolution technique to multiple color planes of the single image acquired with a sensor with Bayer pattern separately to produce multiple deconvolved color planes; and combining the multiple deconvolved color planes to produce the final image with the extended depth of field.
In some embodiments, processing the single image involves using a two-dimensional (2D) deconvolution.
In some embodiments, the 2D deconvolution comprises a Fourier-transform-based deconvolution.
In some embodiments, the image-processing system additionally uses a machine-learning-based noise-reduction technique and/or resolution-enhancing technique while producing the final image.
In some embodiments, the machine-learning-based noise-reduction and/or resolution-enhancing technique involves creating mappings between deconvolved images and ground-truth images.
In some embodiments, the sample was previously stained using one or more fluorescent dyes.
The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.
The following description is presented to enable any person skilled in the art to make and use the present embodiments, and is provided in the context of a particular application and its requirements. Various modifications to the disclosed embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the present embodiments. Thus, the present embodiments are not limited to the embodiments shown, but are to be accorded the widest scope consistent with the principles and features disclosed herein.
The data structures and code described in this detailed description are typically stored on a computer-readable storage medium, which may be any device or medium that can store code and/or data for use by a computer system. The computer-readable storage medium includes, but is not limited to, volatile memory, non-volatile memory, magnetic and optical storage devices such as disk drives, magnetic tape, CDs (compact discs), DVDs (digital versatile discs or digital video discs), or other media capable of storing computer-readable media now known or later developed.
The methods and processes described in the detailed description section can be embodied as code and/or data, which can be stored in a computer-readable storage medium as described above. When a computer system reads and executes the code and/or data stored on the computer-readable storage medium, the computer system performs the methods and processes embodied as data structures and code and stored within the computer-readable storage medium. Furthermore, the methods and processes described below can be included in hardware modules. For example, the hardware modules can include, but are not limited to, application-specific integrated circuit (ASIC) chips, field-programmable gate arrays (FPGAs), and other programmable-logic devices now known or later developed. When the hardware modules are activated, the hardware modules perform the methods and processes included within the hardware modules. The methods and processes described in the detailed description section can be embodied as code and/or data, which can be stored in a computer-readable storage medium as described above.
MUSE (Microscopy with UV Surface Excitation) is a new approach to microscopy, which provides a straightforward and inexpensive imaging technique that produces diagnostic-quality images, with enhanced spatial and color information, directly and quickly from fresh or fixed tissue. The imaging process is non-destructive, permitting downstream molecular analyses. (See Farzad Fereidouni, et al., “Microscopy with UV Surface Excitation (MUSE) for slide-free histology and pathology imaging,” Proc. SPIE 9318, Optical Biopsy XIII: Toward Real-Time Spectroscopic Imaging and Diagnosis, 93180F, 11 Mar. 2015.)
To facilitate MUSE imaging, samples are briefly stained with common fluorescent dyes, followed by 280 nm UV light excitation that generates highly surface-weighted images due to the limited penetration depth of light at this wavelength. This method also takes advantage of the “USELESS” phenomenon (UV stain excitation with long emission Stokes shift) for broad-spectrum image generation in the visible range. Note that MUSE readily provides surface topography information even in single snapshots, and while not fully three-dimensional, the images are easy to acquire, and easy to interpret, providing more insight into tissue structure.
Unfortunately, working with samples with intrinsic depth information can pose problems with respect to determining appropriate focal points as well as capturing extended depth-of-field images. We have developed an accelerated and efficient technique for extending depth of field during MUSE imaging by employing swept-focus acquisition techniques. We have also developed a new method for rapid autofocus. Together, these capabilities contribute to MUSE functionality and ease of use.
Because MUSE operates by performing wide-field fluorescence imaging of tissue using short ultraviolet (UV) light (typically 280 nm) excitation. MUSE techniques can only image the surface of a thick specimen. The fact that objects inside the tissue are not visualized by MUSE allows the computational operations required to capture extended depths of focus to omit computationally intensive operations required by previous approaches to this problem. (Note that MUSE is the only wide-field UV imaging technique that captures an image instantaneously with only surface-weighted features.)
Previous techniques for capturing a single image while varying the z-axis position of focus required an object-estimation step as part of the extended depth of field (EDOF) computation, because those non-surface weighted imaging methods obtained multiple signals at different depths within the imaged volume. For example, see step 310 in the flow chart in FIG. 3 of U.S. Pat. No. 7,444,014 (cited above). In contrast, our new MUSE EDOF technique only detects emitting objects located along the specimen surface, which may or may not be flat. This allows the computationally expensive “object-estimation step” described in U.S. Pat. No. 7,444,014 to be omitted.
Our new MUSE EDOF imaging technique provides a number of advantages.
(1) The technique is compatible with conventional microscope designs. Only minor alterations of a conventional microscope design are required. These alterations can involve assembling the microscope objective on a piezoelectric actuator (or using other mechanical stages) so that a focal plane scanning operation can be synchronized with image acquisition by the camera.
(2) No extra data or image acquisition is required. Thanks to the rapid response of a piezoelectric actuator, the image acquisition time is not extended during the EDOF image-acquisition process.
(3) This technique facilitates near-real-time analysis. During operation, the piezoelectric actuator (or other mechanical device) is configured to move the microscope stage or objective into a desired range (˜100 μm) within the acquisition time of the camera to capture a single image. This technique scans the desired range in a synchronized way with the camera and collects light from all of the layers. By using this image-acquisition technique, we essentially integrate the three-dimensional (3D) profile of the object convoluted with a 3D point spread function (PSF) into a two-dimensional (2D) image. According to the convolution theorem, the integration on z-direction drops out of the convolution and it turns into an object convoluted with a 2D PSF. Hence, the method of analysis is basically a 2D deconvolution of the image. Note that there exist a large number of available methods to perform 2D convolution. However, because of time constraints we have focused on Fourier-transform-based techniques.
(4) The technique also facilitates noise reduction. Although the PSF-based deconvolutions are effective, they add variable amounts of noise to the resulting image. To reduce this noise, we can use a machine-learning-based noise reduction technique, which operates by creating mappings between deconvolved images and “ground-truth” EDOF images obtained using multiple individual planes. These can be combined into a single EDOF image with much lower noise, but at the cost of multiple image acquisitions and very long computational times. Note that using the machine-learning-based mappings (on a sample-type-to-sample-type basis) allows us to acquire rapid, single-image swept-focus images, and to compute a high-quality, low-noise resulting image. Noise reduction can also be accomplished by applying custom window functions on Fourier transformed images to suppress low signal high frequency components. (See for example “Optical Systems with Resolving Powers Exceeding the Classical Limit,” JOSA Vol. 56, Issue 11, pp. 1463-1471.1966.) This technique provides real-time noise reduction because it is simply includes multiplying the window function to the Fourier transformed image. Yet another noise-reduction technique is standard Weiner filtering. (See Wiener, Norbert, 1949, Extrapolation. Interpolation, and Smoothing of Stationary Time Series. New York: Wiley.)
A major benefit of this technique is speed. Note that the image-acquisition time is not limited by the scanning because the scanning operation is synchronized by camera exposure time. Moreover, because the collected data is limited, the associated analysis technique does not require data from multiple layers.
Because stage 108 is movable, it is possible to synchronize the movement of the focal plane through the sample 110 with the gathering of the image by sensor array 102. For example, stage 108 can be used to move the sample 110 toward the objective 104 one micron per millisecond in a 100 ms window, during which sensor array 102 gathers the image. (In an alternative embodiment, objective 104 and/or sensor array 102 can be moved instead of stage 108.) Also, we can employ any type of linear actuator, such as a piezoelectric transducer or a voice coil, to move stage 108.
Note that it is not practical to gather the image for much longer than the time which would saturate the imaging device, so we can only gather a limited amount of light for each focal plane. To remedy this problem, we can take more images and then average them.
During processing step 306, we assume that by summation of the image layer over the axial axis, the convolution on the z-axis drops out and we have the summation of the objects convoluted with sum of the PSF.
For a 3D object, the image from different axial layers is determined by convolving the 3D PSF with the object profile:
I(x,y,z)=O(x,y,z)⊗PSF(x,y,z) (1)
by defining the accumulated intensity as
and the accumulated PSF as
and assuming that the light comes from the surface layer only
we can take the integral of both sides of Eq. (1) over the axial axis:
I
a(x,y)=O(x,z)⊗PSFa(x,y) (2),
and can recover the object by applying a deconvolution filter of PSFc(x,y):
O(x,y)=Ic(x,y)⊗−1PSFc(x,y) (3).
For near-real-time processing, methods based on inverse Fourier transforms will be acceptable. In these methods, the summed image is Fourier-transformed and divided by the OTF (the Fourier transformation of the PSF) to deconvolve the summed PSF from the blurred image. While fast, these FFT-based methods are unfortunately noisy and require appropriate noise suppression, such as through Weiner filtering. By using a fast method for performing the Fourier transformation, such as the method used by the Fastest Fourier Transform in the West (FFTW) software library developed by Matteo Frigo and Steven G. Johnson at the Massachusetts Institute of Technology, an EDOF for a 9 megapixel image can be returned within a second.
As illustrated by the images that appear in
Note that the extended surface topography of the kidney sample, which comprises thick tissues, makes it hard to obtain in-focus images, even using a conventional autofocus, as illustrated in Row A, wherein a single image is taken at optimal focus. In contrast, row B illustrates a much clearer, deconvolved through-focus single image, and row C illustrates a deconvolved average of a 10-frame z-stack. Note that while the images in row C, which are constructed from multiple z-planes, display much less noisy in-focus images than in row B, it takes almost 10 times longer to acquire these images, and longer still to process the multiple z-plane images. Our goal is to achieve the quality of row C within the time it takes to acquire and process the data in row B.
Various modifications to the disclosed embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the present invention. Thus, the present invention is not limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.
The foregoing descriptions of embodiments have been presented for purposes of illustration and description only. They are not intended to be exhaustive or to limit the present description to the forms disclosed. Accordingly, many modifications and variations will be apparent to practitioners skilled in the art. Additionally, the above disclosure is not intended to limit the present description. The scope of the present description is defined by the appended claims.
This application claims priority under 35 U.S.C. § 119 to U.S. Provisional Application No. 62/623,320, entitled “Method for Extending Depth of Field for Microscopy Imaging” by the same inventors as the instant application, filed on 29 Jan. 2018, the contents of which are incorporated by reference herein.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/US19/15485 | 1/28/2019 | WO | 00 |
| Number | Date | Country | |
|---|---|---|---|
| 62623320 | Jan 2018 | US |
