Customer Solutions
Prototyping a Digital Flow Cytometer the NI way
Author(s):
Eric Nehrlich, San Fransisco Industrial Software
Industry:
Life Science
Product:
Data Acquisition, LabVIEW, PXI/CompactPCI, Signal Conditioning
The Challenge:
Becton Dickinson Biosciences (BDB) asked us to quickly and cost-effectively prototype a digital flow cytometer to determine if such a system could replace their current analog card-cage electronics.
The Solution:
We used three PCI-6110E cards and LabVIEW to acquire the analog signals directly from the photo-multiplier tubes (PMTs). After digitization, we used LabVIEW to reproduce the signal processing that had been done with the card cage, and to provide sophisticated analysis and display tools.
Abstract
Flow cytometry involves characterizing cells by analyzing their fluorescence as they flow past a laser. Existing designs use analog signal processing to manipulate the fluorescence signals. By using NI hardware and LabVIEW, we were able to quickly and cost-effectively prototype a digital version of their system. The prototype demonstrated to our client, Becton Dickinson Biosciences (BDB), the feasibility of using PC-based instrumentation and digital signal processing for cytometry. Our system replaced a card cage of several boards with three NI PCI-6110E cards, and achieved greater precision than the old system with similar usability.
Introduction
Becton Dickinson Biosystems makes flow cytometers, which analyze laser light scattered off cells to classify and sort cells. The scattered light is collected by photo-multiplier tubes (PMTs) and amplified before further processing. In the original system, the analog signals were fed into a card cage of several custom-designed boards, before being digitized and brought to a computer for display and further analysis. We were asked to develop a prototype system that would do the signal processing digitally, replacing the card cage with data acquisition cards from National Instruments.
Description of the problem
In flow cytometry, populations of cells are tagged with various fluorescence markers, and suspended in a stream that is squeezed to only a single cell width in diameter. When a laser focused on the stream illuminates a falling cell, it will fluoresce at frequencies corresponding to its markers. By passing this fluorescent light through filters, different wavelengths can be separated so that the response for each individual marker can be determined.
The user can determine what kind of cell passed through the laser by examining the response from each PMT and filter pair. The responses from two PMT-filter pairs are plotted along the axes of a dot plot, allowing multi-dimensional comparisons. Regions are drawn on the dot plot to identify populations of cells to assist in calculating counting statistics. One complication was that our system used two lasers, which illuminated the stream at different points in its descent, introducing a delay between the observed signals.
Fluorescences can also have overlapping frequency responses. To compensate for this, a mechanism must be available to subtract out frequency response that may be due to another signal. For instance, if 20% of fluorescence 2 (FL2) shows up in the frequency response range of fluorescence 1 (FL1), the user can determine the "real" amount of FL1 by subtracting 20% of the FL2 response from the measured FL1 response.
Hardware
We chose PCI-6110E cards as our data acquisition hardware, due to their high-speed capabilities. The pulses we needed to digitize were often in the 4 ì s range, so we needed a digitization rate of at least 4 MHz. We used LabVIEW to write our acquisition software due to the ease of integration with the cards, and our expertise with it. The development system used was a 450 MHz dual-Pentium II processor computer with 256 MB of RAM, running Windows NT.
Implementation
This project required a sampling rate of 4 MHz to adequately describe the pulses. Taking data at this rate continuously would mean an enormous amount of data to process. Since the actual rate of cells was on the order of a few kHz, we were only interested in approximately 1 percent of the signal. By turning off sampling when there was no pulse , we achieved a tremendous reaction in the amount of data to be processed by LabVIEW. We accomplished this by re-routing the internal signals of the PCI-6110E timing hardware onto RTSI lines to create an external clock for acquisition.
We used LabVIEW to acquire and process the data from the cards. First, pulses from the two lasers were synchronized by subtracting out the time offset. The height, area, and width of each pulse were then calculated. A compensation matrix was applied, permitting the subtraction of overlapping frequency responses as described above. The data could also be written to a datalog file at this point if desired. Each of these vi’s was heavily optimized because they had to be run for each cell acquired.
After this, the data was sent via a notifier to a display loop. This loop displayed a two-axis scatter plot, as well as a histogram of each axis’s data individually. It also calculated counting statistics for user-defined regions defined on that dot plot. We used the picture control toolkit to allow the user to select regions of any shape by pointing and clicking. From this screen, the user could call up a vi to select a compensation matrix to adjust the frequency overlap between signals. Users could also spawn off dot plots in separate child windows to allow them to examine several characteristics at once. This was accomplished by creating a template vi, which was copied to a temporary file and then launched dynamically whenever a new child window was selected.
Lastly, an oscilloscope mode was added, which allowed the user to observe the actual waveforms being collected. This proved to be exceedingly useful as it alleviated the need to bring in a separate oscilloscope for the purpose of synchronizing the signals from the two lasers and seeing pulses.
Results
By using off-the-shelf hardware and LabVIEW, we were able to look at cells on the screen in approximately three months. In three more months, we had recreated the functionality of the previous analog system, including compensation, regions, and various counting statistics. In fact, our digital system had additional capabilities, including the ability to measure pulse width and area, the ability to use a full compensation matrix, the oscilloscope display and greater precision in all measurements. Furthermore, it was enormously flexible, allowing changes to the analysis and signal processing algorithms in software, without costly redesign of circuit boards. The final system was capable of analyzing up to 5,000 cells a second, after optimizing both our data acquisition and analysis strategies.
Our development effort of six months had more accuracy and versatility than an analog system that had been evolving for 25 years. The power of computer-based instrumentation was amply demonstrated.
nehrlich.pdf
View the entire user solution in Adobe Acrobat PDF format.
