High-throughput Automated Tracking (HAT)
Feature List
What is HAT?
High-throughput Automated Tracking (HAT) is a set of routines and software objects specialized for tracking and analyzing the movement and appearance of large numbers of particles in time-lapse movies. It can be used as a stand-alone image processing library, or integrated into an existing image processing application as a module. It is designed for ease of use for the novice user, speed and data support for the high-throughput user, and flexibility and programmability for the advanced researcher.
What has HAT been used for?
HAT was first developed for tracking bacteria (Listeria monocytogenes) undergoing actin-based propulsion in cellular extract. This application needed two features that were not included in today's microscopy image processing modules. First, the tracking algorithm needed to be fast and automated, as researchers anticipated tracking thousands of bacteria in hundreds of individual movies. Second, the program needed to keep track of the entire region around the tracked object, in multiple image channels, for later shape and intensity distribution analysis. Since neither of these features are in any existing microscopy image processing program, researchers set out to develop their own, automated image processing system.
HAT has successfully been used to track large populations (>5000) bacteria, at rates of 100 bacteria/hour. The bottleneck is the data collection phase (not the tracking or analysis phase). The data collected on each bacterium exceeds 1MB, including x-y trajectory, heading, movies, and links to all original files. This system allowed researchers to develop a suite of measurements which statistically describe the movement of Listeria over time, including measures of linear speed, rotational yaw angle, spectral analysis of fluctuations, and identification of collisions and other anomalous events, as well as how the movement of the bacteria is coupled to changes in the intensity of actin near its surface.
Essential to the use of HAT was its speed and flexibility. As the research unfolded, new statistics and calculations were done simply by reprogramming parts of HAT and re-analyzing the entire data set. This was possible because HAT takes at most a few hours to calculate complete statistics for an entire data set of several thousand bacteria, and is designed to scale linearly with data set size, keeping the development cycle short and predictable. More importantly, because data is kept by HAT in a relational database, it was possible to automatically keep track of how data was spread out over the hundreds of data and calculation files. Because of this feature it was possible to do multiple calculations to cross-check and verify results. Finally, when producing the final results for publication, graphs and tables could be produced directly from SQL queries of the HAT database, speeding production, final tweaks and verification, and ensuring the accuracy of published results.
As initial results using HAT were reviewed, it became clear that HAT would be useful for more than just tracking bacteria. With this in mind, HAT was developed as a general purpose tracking tool. Although optimized for tracking bacteria, it contains many general programming features so that new algorithms could be developed and tested without significant overhead. For instance, HAT implements several commonly used algorithms for tracking analysis, such as two dimensional spline fitting and power spectral analysis, which are not included in the typical image processing suite or scientific graphing program. These algorithms can be used as building blocks for more advanced analyses, saving development time.
What are possible scientific applications of HAT?
Although HAT was originally designed to track bacteria, it can in principle be used in any system where robust tracking algorithms can be employed, and is ideal for projects in which high throughput and advanced image processing is desired.
Historically, many projects in cell biology have historically relied on images with marginal signal-to-noise ratios, limiting tracking techniques to manual or semi-automated at best. However, with the advent of cooled CCD cameras, advanced labeling and illumination techniques, many areas of cell biology and neuroscience have been revolutionized by vast improvements in signal-to-noise ratios. In many cases, the limit has become the rate and accuracy with which large numbers of particles or image features can be tracked over time; these are projects well suited for HAT.
For instance, cellular motility is often studied using keratocyte models, in which individual cells move around on a two dimensional substrate. How cells move, how they change shape and how this is accomplished by the underlying molecular machinery has interested researchers for decades. With the development of labeled fluorescent proteins and new imaging techniques, it has become possible to visualize the distribution of actin and other cytoskeletal proteins as the cell crawls. Of great interest now is how the dynamics of these proteins are coupled with the movement of cells over time, and why some individual cells move faster or slower than others. This kind of problem is well suited for analysis with HAT. Hundreds or thousands of individual cells can be tracked in detail to identify features of individuals and population trends.