Core 4: BioInstrumatics
We continue to work towards the goal of reducing the intellectual effort required to carry out the mechanics of an experiment, thereby increasing the intellectual energy available to the scientist to plan, evaluate, and analyze. We also provide integrated visualization and analysis tools within experiment-control systems to further enable the scientist to study results in context as the data is being collected. In the coming years, this core will be increasingly engaged in the integration of individual tools into a data collection/analysis pipeline for individual experiments and for our developing high throughput systems. As we move to performing one hundred experiments in one hour, we will be faced with the extraordinary demands of processing terabytes of data, including data storage, automated analysis and user interface design. Here we report principally on our development of individual tools during the past grant year, with a look ahead to how these tools will contribute to our integrated systems.
New FluoroSim Capability: Analytical PSF Optimization
Scanned-Probe and Fluorescence microscopy are powerful tools for localization of structures in biological specimens. However, aspects of the image formation process such as noise and blur from the microscope’s point-spread function (PSF) (or dilation by the SPM tip) combine to produce unintuitive image transformations on the true structure of the specimen, hindering qualitative and quantitative analysis of even simple structures in unprocessed images. We have developed an interactive combined AFM/Fluorescence simulator to train scientists who use scanned-probe and fluorescence microscopy to understand the artifacts that arise from the image formation process, to determine the appropriateness of each technique as an imaging modality in a particular experiment, and to test and refine hypotheses of model specimens by comparing the output of the simulator to experimental data.
It is difficult to extract high-quality point-spread functions from images of fluorescent beads. Noise within the images, if left unfiltered, produces aliased patterns in the resulting simulation volume that impedes understanding. Filtering this noise by averaging multiple images produces a blurred point-spread function that does not precisely represent the actual function. To address these issues, Resource graduate student Cory Quammen has implemented a model-based point-spread-function optimization tool that provides a best-fit analytical form to the measured PSF. It adjusts more than a dozen system parameters (lens aperture, cover-slip thickness, media indices of refraction, bead radius, bead center, etc) to provide the minimum-error fit. This enables the construction of subvoxel-accurate localization of the PSF for use in simulation and optimization.
Moving forward, this capability will be critical for experiment planning where the fluorescence signals appropriate for specific data analyses will need to be optimized. The Fluorescence simulator will be a tool for this activity, either as a user guided optimization – or as an engine for an automated routine.
High-Throughput Magnetics: Intelligent Storage Reduction
We have augmented our Video Spot Tracker application so that it can export full-rate video in regions surrounding tracked beads along with reduced-rate video of the entire scene. This enables a scientist to reduce the stored file size by about a factor of 100 without losing any information at or near the beads that were studied. This enables re-running of the tracking with different parameters for those beads that were part of an experimental analysis without requiring the storage of the entire video. Coupled with automatic-bead-finding capabilities, this enables automatic reduction of storage.
Structured Substrates for Tracking
The calibration of the 3DFM pole pieces and magnetic response requires the viewing of a region that is larger than the field of view of the video camera and also larger than the range of motion of the computer-controlled stage. Following beads under cilia-driven flow requires the motion of the stage in XY beyond what the computer-controlled stage can provide. Studies of tissue heterogeneity require high-resolution, wide-field-of-view images to be collected. Each of these requires us to follow objects of interest across multiple fields of view, as obtained by hand-moved stages, creating a virtual wide-field-of-view image on which the object of interest can be overlaid. This tool will be much more widely applicable than our full 3DFM user interface, providing wide-FOV microscopy to microscopists studying tissues of all kinds.
A multitude of approaches have been investigated in the context of creating panoramic images from individual photographic images (also known as image stitching or image mosaicing). The problem is one of tracking via image registration.
Our experimental setting allows a significant simplification of the tracking problem: we can restrict estimation to translational motion. Fourier-based phase correlation methods are ideally suited for image alignment based on structured substrates with known spatial frequencies. We can simplify the general phase correlation methods because we know the spatial frequency of interest a priori (by design of the structured substrate), which we can probe directly. We can compute image alignments with subpixel accuracy for purely translational motion with phase correlation. We already have designed and implemented:
- A customized FFT-based orientationdetermination algorithm that provides a robust estimate of pattern orientation based on the fundamental pattern frequency and its harmonics.
- A novel windowless discrete sine/cosine transformation algorithm that uses all image pixels to determine the relative phase of the pattern being used for tracking between pairs of images.
- A two-stage alignment procedure (fine alignment based on the pattern, coarse alignment based on cross-correlation at fixed pattern-width offsets) that enables tracking even when the stage is moved more than half a pattern width between successive images so long as there are specimen features present in the image. This procedure makes structured-substrate tracking practical for use with handcontrolled microscope stages.
We have also validated that the algorithms can align images to within a small fraction of a pixel; this validation has been done on synthetic images (whose offsets are known because they are computed by us) that include standard microscope images modulated by a pattern with Poisson noise added. We are currently studying how the defocus and opacity of the pattern affects tracking accuracy, both in simulated images and using mounted slides of octopus muscles and frog brains.
We have submitted a provisional patent application on this design and are seeking commercial partners to broadly deploy this technology.