A Fast Optical Recording System

| Overview |
Introduction | Construction of the system | Fluorescence Imaging | Electronics |
Acquisition and storage of optical signals | Performance evaluation |
Linearity | Noise figure | Conclusions |
References |

| Fast Ion Imaging | System description |Brain Slices| Downloadable Movies |CCD Primer |


This document presents technical details, hardware and software of a
complete imaging system which uses a fast CCD sensor and a 41 Msample/s A/D converter to acquire full-frame 12 bit/pixel digitised images with a time resolution of 1.25 ms/image. The noise characteristics and the photon-capture capabilities of the system were analysed in detail. At the maximum readout rate, noise from the electronics amounted to ~300e-, without cooling. Correspondingly, under illumination conditions compatible with cell biology, the minimum fluorescence change signal, DF/Fo, detectable with a S/N @ 1 was of the order of 1% . Better figures, down to 0.25% DF/Fo, were obtained at lower rates while cooling the sensor to about 15oC.This apparatus permits to resolve ion gradients in the cytoplasm of individual cells as well as the spatio-temporal pattern of neural activity of cell assemblies in neural tissue.

Optical methods are interesting to physiologists and neuroscientists for several reasons
[1]. Of particular importance is the capability of recording simultaneously from multiple sites and the possibility to reach fast sampling rates. Therefore it is fundamental to develop imaging systems that, combining both features, achieve a high spatio-temporal resolution. In this respect, CCD sensors are well suited for detecting low-light level signals from fluorescent probes introduced into living cells [2]. These devices possess high quantum efficiency for converting light into electrical signals that allow monitoring variations in the intracellular concentration of Ca2+ and other fundamental ions within the cell cytoplasm. Here we provide a detailed description of an imaging system based on a sensitive CCD sensor with high readout rate (16 MHz). Sensor’s output was digitised at 12 bit/pixel by customised electronics permitting image acquisition rates up to 800 frames/s with a full-frame resolution of 128 128 pixels (click here to see a demonstrative movie).

Construction of the system
Fig.1 presents a block-diagram of the recording system that was built around an upright microscope, mounted on a vibration isolation table, and included a CCD detector unit, a digital frame-grabber and a PC. A second PC was used to drive a laboratory interface for the synchronisation of optical and electrical signals from the cell preparation placed on the microscope stage.

Figure 1. Schematics of the recording system.

Fluorescence imaging
The microscope (MI 250, Newport-Microcontrole) was equipped with infinity-corrected water-immersion objectives (
10, N.A. 0.30; 20, N.A. 0.50; 40, N.A. 0.75; 63, N.A. 0.90; Achroplan, Carl Zeiss) and a custom made epi-fluorescence illuminator incorporating an excitation interference filter and a long-pass dichroic mirror. Excitation light from a 75W stabilised Xenon arc source (CAIRN Research Ltd.) was coupled to the microscope via a liquid light guide gated by a rapid shutter (UniBlitz, Vincent Associates). Fluorescence emission was selected using a second pair of dichroic blade and interference filter (Fig.2).

Figure 2. Scheme of the optics. The set of interference filters and dichroic mirrors shown was optimised for Calcium Green-1.

Fluorescence images were formed on the fast CCD sensor using an achromatic doublet as projection lens. Distances in the image plane were calibrated by placing a 10 m-pitch graticule under the objective. A second projection lens was used to form IR images of the specimen on a conventional CCD camera operating at standard video rate, with the spatial resolution allowed by the PAL signal (768 576 pixels). The entire optical group was mounted on a system of X-Y-Z motorised translators that allowed it to be moved with respect to the imaged sample. For each image pixel, fluorescence signals were computed as ratios DF(t) / F(0) = [F(t) - F(0)] / F(0), where t is time, F(t) is fluorescence following a stimulus that causes calcium elevation within the cell and F(0) is pre-stimulus fluorescence computed by averaging 10-20 images. Both F(t) and F(0) were corrected for mean auto-fluorescence computed from a 20 20 pixel rectangle devoid of obvious cellular structures. Ratio magnitude was encoded by 8 bit pseudo-colour look-up tables to produce false-colour images. This local ratio computation is expected to provide correction for time-independent non-uniformity in optical path-length and dye concentration [3]. Light was attenuated with a diaphragm to avoid phototoxicity by reducing photo-bleaching rates to 0.5%/s. Images were analyzed using routines developed from Matlab 5.1 Image Processing Toolbox (The Mathworks Inc.). Images were generally smoothed with a 33 two-dimensional median filter [4] and fluorescence traces were smoothed with a three-point zero-phase digital filter [5].

The core of the optical recording section was a fast CCD sensor with 128 128 pixels (DALSA IA-D1). The sensor was attached to a copper plate thermo-electrically cooled by a 50 Watt peltier device (Marlow Industries) driven by a variable power supply. The CCD output was digitised at 12 bit/pixel using a specialised A/D conversion circuit, whose block diagram is shown in Fig.3. The circuit consisted of surface-mount electronic devices (SMD) placed on a 6-layer printed circuit board (PCB) containing separate layers that were routed using low-noise arrangement techniques. Low-noise FET-input operational amplifiers (OPA655, Burr Brown) having large bandwidth, low distortion and low bias current were selected for this design.

Figure 3. (A) block diagram of the 12 bit A/D image digitisation board. Typical waveforms, sampled at 1 GHz using a digital oscilloscope (LC574A , LeCroy) from the crucial points of the circuits, are reproduced for clarity. The CCD output video signal is formed by an alternation of internal reference (porch) and pixel voltage: each pixel signal in an image is separated from the next one by the porch voltage. Pixel frequency is under control from an external clock and can be increased up to 16 MHz. (B) Idealised waveforms to show how the CDS circuit measures the difference between porch and pixel value at each tick of the CCD clock. Top: CDS input. Bottom: output. (C) Software flow-chart, using the image-acquisition library (ITEX-IC Core) supplied by the manufacturer of the frame grabber. First the number of frames to be acquired was specified. Then host memory was allocated using host memory allocation functions. Finally all frames in a sequence were acquired with a single call to the high-level function icp_start_VCR_record.

The board’s input was amplified in voltage by the inverting buffer and sent to the correlated double sampling (CDS) section whose function was to reduce the detrimental effects of low frequency noise (low relatively to the pixel frequency) and slow power supply and temperature drifts. This was achieved by taking the pixel-by-pixel difference between CCD internal reference (porch) and pixel level (Fig.3B). The underlying assumption is that porch and pixel values are equally affected by low-frequency noise, coherent over a time-scale comparable to the clock period. For each period of the CCD clock, a capacitor in the CDS circuit is charged to the porch voltage during phase (a). The switch is opened during the successive phase (b) allowing the difference between pixel and porch voltage to be measured by the A/D converter at the times marked by vertical arrows. The lower limit of the capacitor’s value was determined by the desire to achieve a settling time shorter than the shortest clock semi-period (about 30 ns), taking into account the finite output impedance of the operational amplifiers. The upper bound was set by the constraint that the capacitor should not discharge appreciably (i.e. less than 1 part in 4096) through the parasitic resistance during one semi-period. A compromise value was found to be 50 pF. Output from the CDS entered a summing block where the black-level reference was displaced by adding a reference voltage and the result was sent to the clip op-amp. This was used to increase the dynamic range of the A/D converter by the removal of the signal fraction below the black-level reference and to protect the converter’s sensitive input from out-of-range signals. The A/D converter adopted for this design (AD9042, Analog Devices) performed nominal 12 bit conversions at a maximum rate of 41 million of samples per second. Digital output was converted to standard RS422 differential signal by a set of line drivers (AM26C31, Texas Instruments).

Acquisition and storage of optical signals
The CCD detector unit operated in frame-transfer mode, with photo-charge integration time amounting to 96% of the inter-frame interval. During the remaining 4%, pixel data were shifted from the light-exposed to the masked region of the CCD. Digitised pixel values were output from the 12 pairs of RS422 differential lines and fed to a 16 bit digital frame grabber (IC-PCI/AM-DIG16, Imaging Technology) controlled by dedicated software. The software permitted the sequential capture of frames at the maximum rate compatible with the sensor’s electronics (800 frames/sec) without frame loss and up to the capacity of the PC’s RAM. In a typical experiment, sequences of 100 to 1000 frames were stored in RAM and saved off-line on a UW SCSI hard drive either as raw binary data or as Audio Video Interlaced files (AVI, Microsoft Video for Windows RIFF format). Computer programs used to drive the frame grabber were written in Visual C++ using object-level real-time library code (ITEX-IC Core) supplied by the manufacturer and were executed on a 133 MHz Pentium PC operating under Windows 95 (Fig.3C). The control signals required to trigger image acquisition were generated by a 12 bit laboratory interface (1401plus, Cambridge Electronic Design) using customised control software running on a second PC (see Fig.1). The frame-valid (FVAL) pulses, produced by the CCD sensor whenever a frame-transfer operation was completed, were sampled by the same interface in order to trace image timing.

Performance evaluation
To assess the capability of the recording system to acquire high-speed sequences of frames, a LED digital counter was built and clocked by FVAL pulses, each FVAL pulse incrementing the counter by one digital unit. The framegrabber driver allowed the user to allocate up to 232 Bytes of physical RAM (not virtual memory). It follows that, in principle, the maximum amount of PC memory that could be reserved for acquisition was limited only by the available RAM, although in the present configuration only RAM sizes up to 128 MBytes were tested. Careful analysis of images generated by focusing the sensor onto the counter (Fig.4A) revealed that frame storage to PC RAM and subsequent dump to the hard drive occurred with no frame loss for up to 3500 consecutive images (each stored image occupied 32 kBytes and the operating system required some 10 Mbytes to run efficiently).

Figure 4. System validation. (A) Consecutive digital images of a LED counter clocked by the FVAL signal of the CCD. Images show no lag in response, indicating that the effective speed of the optical recording system is that of the FVAL signal (Max. 800 frames/s). (B) Image of fluorescence beads clustered on a microscope coverslip. Fluorescence intensity was measured from the four regions of interest (ROIs), labeled (a) to (d), for a set of 100 frames acquired at 800 frames/s. (C) Bead fluorescence, averaged within each ROI and over the whole 100 frame sequence, is plotted against normalised illumination intensity for the four ROIs shown in (B). Solid lines are linear least-square fits to data. Error bars are smaller than plot symbols and do not show. Extrapolation to zero input gives non-zero output because of residual offset, affecting the CCD signal prior to digitisation.

Linearity was checked by illuminating a sample of 10 M diameter fluorescent beads (FluoSperes; F-8830, Molecular Probes) placed on a microscope cover slip under the objective (Fig.4B). A sequence of 100 images was acquired at the rate of 800 frames/s and bead fluorescence was measured from several points of the view field and plotted vs. illumination intensity level (Fig.4C). Illumination was varied by changing the aperture of the diaphragm in front of the fibre optics and its intensity was measured with a photodiode placed on the illumination light path after the dichroic blade. The input/output function was linear over a wide range of intensities and irrespective of the position of the sample in the view field, indicating that the system was well suited for dynamic fluorescence measurements.

Noise figure
The overall noise associated with the acquisition of an image by a CCD device is where NR and ND are the readout noise and the dark-charge noise of the CCD, respectively, and NPh is the photon shot-noise associated with the fluorescence signal. The sensor’s full-well capacitance was about 3.8
105 e-. When it was illuminated to half-saturation under constant and uniform conditions, without cooling, the single-pixel noise, measured during sequence acquisition at the maximum frame rate, was approximately 2.5 bits. As NPh is the square root of the photoelectron number, the contribution to the overall noise from the electronics was of the order of 300 e-. This figure, which at 800 frames/s is due almost exclusively to NR, can be assumed as an upper bound for the electronics noise at all frame rates. In fact, NR decreased whereas ND increased when reducing the speed of acquisition, but ND was kept efficiently under control by cooling the sensor. The contribution from ND to the total noise was negligible for frame rates 200 frames/s, even without cooling, and was rendered negligible for frame rates down to 50 frames/s by keeping the sensor at 15 oC. Maximum background fluorescence intensities F(0) of the order of 4 104 photoelectrons/ms/pixel were consistently achieved by loading cells with Calcium Green-1 or Oregon Green-1 at concentrations of 75-100 M, while using illumination levels that did not cause either photodamage or appreciable bleaching over repeated exposure periods of 100-200 ms. Therefore, the minimum signal detectable with a S/N @ 1, was of the order of 1% at the minimum inter-frame interval of 1.25 ms. In practice, from experiments conducted on living cells within brain slices, we determined that 1.5 ms should be taken as the shortest viable inter-frame interval for biological fluorescence imaging applications. The S/N could be increased by a factor 2.5 by operating the sensor in 22 binning mode, i.e. by collecting 4 times as many photons per pixel at the same rate (the frame rate can be doubled in binning mode, if desired, and memory requirements are obviously reduced to 1/4 under these conditions).

The purpose of this presentation was to illustrate the performance of an optical recording system developed for fast biological imaging. It was constructed by assembling a range of different optical and electronic components and optimised for high-speed detection of fluorescence emissions from long-wavelength Ca2+-selective probes. At the core of the optical detection section we placed a low-cost commercial CCD sensor, originally designed for industrial applications, whose output was digitised by a specialised A/D board developed for this project. This allowed image capture at rates up to of 800 frames/s, with the possibility of acquiring over 3500 full-frame images in sequence with 12 bit/pixel accuracy. Such high rates and sensitivity result from the good quantum efficiency of the sensor (40%) and the extremely high speed at which frames were transferred from the unmasked to masked regions of the CCD. As the overhead for frame-transfer time was reduced to a mere 4% of the inter-frame interval, photo-charge integration was particularly efficient. The recording apparatus was tested by monitoring the dynamics of intracellular Ca2+ in experiments conducted on cells within highly organised sensory and
neural tissues. The results demonstrated its versatility and potentiality, especially in cases in which high rates of acquisition were required. CCD-based cameras previously described and utilised for applications in the field of biological imaging achieved rates of the order of 100 frame/s only at the expenses of spatial resolution.

In principle, the recording system described here can be made to operate also in ratiometric mode using Fura-2 in combination with a fast-switching wave-length selector (e.g. Polychrome II; TILL Photonics) with an expected maximum rate of 329 frame-pairs/s, although this was not tested. The major drawbacks of this design is the lack of confocality. However this may not necessarily constitute a serious limitation for several interesting experimental situations where 2D signal propagation patterns are the main focus. In this respect, it is worth mentioning that, if the investigated sample possess near-radial symmetry, the optical sectioning power of the fluorescence microscope can be improved substantially by restricting the illumination field [6].

1. Cohen, L. Special topic: Optical approaches to neural function. Ann Rev Physiol 1989; 51: 487-490.

2. Lasser-Ross, N., Miyakawa, H., Lev-Ram, W., Young S.R., Ross, W.N. High time resolution fluorescence imaging with a CCD camera. J Neurosci Meth 1991; 36: 253-261.

3. Neher, E., Augustine, G.J. Calcium gradients and buffers in bovine chromaffin cells. J Physiol 1992; 450: 273-301.

4. Lim, J.S. Two Dimensional Signal and Image Processing. Prentice Hall, Englewood Cliffs 1990; NJ. 469-476.

5. Oppenheim, A.V., Shaffer, R.W. Discrete-Time Signal Processing. Prentice Hall, Englewood, Cliffs 1989; NJ. 311-312.

6. Hiraoka, Y., Sedat, J.W., Agard, D.A. Determination of three-dimensional imaging properties of a light microscope system. Biophysical J 1990; 57: 325-333.

| Fast Ion Imaging | System description |Brain Slices| Downloadable Movies |CCD Primer |