Skip to main content

The control system for the high-throughput magnetic system described in our first Core research project.

Subproject 3.1 Synchronous Capture and Magnet Control

The twelve specimen wells of the Array Microscope are recorded on video simultaneously. They each have associated computer-controlled magnetic drive units, driven from a dedicated computer. The twelve video streams and associated magnetic drive signals must be synchronized to enable appropriate analysis, and they must also be synchronized with stage motion.

Aim 1: Develop synchronous Video, Magnet, and Stage Control and Data Storage.

Research Methods: We will use a commercial software-based clock-synchronization tool that uses Ethernet to keep the clocks on the systems synchronized to within 10ms. All data storage will occur via our public-domain VRPN protocol, which stores timing information with each message.

Subproject 3.2 Full-Rate Capture, Background Compute

It would require several days to copy the data for one day’s experiments off of the cluster of twelve camera computers. This means that all data analysis must be performed on the cluster itself. For experiments requiring more analysis than can be provided while the data is being collected, we propose to develop real-time capture routines that can operate during offline calculation on stored data from the previous experiment to enable data collection for a new experiment to overlap with data analysis for an earlier experiment.

Aim 2: Full-Rate Capture during Background Computation.

Research Methods: We intend to purchase computers with two SATA RAID controllers and to populate each with four drives (two per controller, in RAID 0 configuration). Each computer will have 8 processor cores and a CUDA-enabled image-processing accelerator card. We will ping-pong experiments between sets of drives, so that analysis proceeds on the most-recently-written set and data collection occurs on the other set.

Subproject 3.3 Automatic Motion Control

Aim 3: Locate sufficient beads at an appropriate distance from each of the twelve poles.

Research Methods: The same methods applied in System Parameter Estimation (project 2.2.5, aim 9) will be used to identify the location, top and bottom of the pole tip and pole flat, and to position the camera at the appropriate distance from the tip. In this case, the stage motion will be controlled automatically in all three axes, removing the need for either patterned substrates or image stabilization. The mean of the best positions for each of the twelve wells will be used as the initial estimate for the best location.

The initial estimate will be refined by seeking the Z position that ensures there are at least a minimum number of beads in focus in each of the twelve images (the minimum will depend on the experiment, but will be at least one). The number of beads will be determined using the automatic bead-finding routines in the Video Spot Tracker (project 3.2.5).

Subproject 3.4 Intelligent Storage Reduction

The 12-computer cluster will not have enough bandwidth coming out of it to store the entire video sequence for a day’s experiments. Instead we envision storing the position vs time data for each bead instead of each pixel of every video frame from each camera, greatly reducing the volume of stored data. For quality-control purposes, scientists will want to have at least some of the video data from an experiment to verify that analysis was done appropriately.

Aim 4: Dramatically reduce storage space while enabling quality control.

Research Methods: The simplest approach is to only store a subset of the video frames (storing one per second will produce an 80-fold reduction in space) and to compress these video frames using a lossless compressor (a further factor of 4 is typical for our existing experiments). This will enable validation of analysis, but not re-running analysis in case of subtle tracking errors.

The second approach will be to use the built-in ability of the VRPN imager server to send subsets of a full image. In this case, the regions around each tracked bead in the image in every frame. For a 640×480 video stream with ten tracked beads of diameter 10 pixels, this provides a factor-of-300 reduction in storage while enabling full-frame-rate, full-accuracy tracking of the beads later if needed.

The third approach is to adopt a “security camera” storage mode, where only the portions of the experiment that included interesting or surprising results are stored, while the rest are discarded after analysis reveals nothing interesting.