Stephen Anthony

While fluorescently stained cells may look nice, unless you can quantify where the different fluorophores are within the cell, all you have is a pretty picture. Particle detection has been a recurrent challenge throughout Anthony’s research career, where often existing solutions were insufficient. The nature of the challenge, though, has depended upon what is being imaged. When imaging micron-scale Janus particles, detection of the Janus particles was not an issue, but determining their orientation was. On the other hand, when imaging fluorophores in cells, the number of photons available is limited and low signal-to-noise ratio is a much bigger concern. Research in this area included developing a wavelet-based spot detection algorithm capable of detecting particles in lower signal-to-noise ratio images. Alternatively, simple detection may not be difficult but quantification may be challenging due to the need to distinguish whether an observation corresponds to one particle or multiple particles. In this case, compressive sensing algorithms were adapted to his purposes, demonstrating the capability to resolve a pair of point spread functions (highlighted by the larger, white circle) separated by less than half the Abbe limit.

Tracking Moving from spatial information to spatio-temporal dynamics generally requires following of particles (e.g. fluorophores) though time, typically requiring single particle tracking algorithms. Unfortunately, multiple hypothesis tracking is known to be NP-hard, so except for small problems the optimum solution cannot be found, only approximations. Anthony’s research has included the development of a quasilinear time N-dimensional particle tracking algorithm. In particular, this algorithm was designed to support arbitrary dimensions beyond standard space and time dimensions. For instance, particle intensity or ellipticity could be incorporated as dimensions.

Following the theme of exploring additional dimensions, some of Anthony’s research has focused on multispectral and hyperspectral imaging, adding an additional spectral dimension to the image data. While the resulting image data can extend to 5D data sets (3 spatial dimensions, 1 temporal, and 1 spectral), at its simplest level hyperspectral imaging consists of collecting the intensity at multiple wavelengths for every pixel in a standard 2D image. This ranges from multi-color systems such as Jeri Timlin’s dual-color, video-rate, total internal reflection (TIRF) and stochastic optical reconstruction (STORM) microscope to Michael Sinclair’s hyperspectral confocal microscope which can resolve complete fluorescence spectra. Recent work (along with Jeri Timlin) has focused on combining hyperspectral imaging and super-resolution microscopy, specifically stimulated emission depletion (STED) microscopy, as well as incorporating a second spectral dimension, the excitation spectrum.

When working with higher dimensional data sets like hyperspectral images, our ability as humans to visualize all the data is limited. Moreover, quantitative analysis is much more reliable than subjective human impressions. Therefore, sophisticated multivariate analysis tools are needed to extract the underlying spectral signatures and spatio-temporal distributions. However, while much of Anthony’s research has focused on applying multivariate curve resolution (MCR) to hyperspectral images, MCR decompose other signals into components, including gas chromatography-mass spectroscopy (GC-MS).