In response to almost any stimulus, discrete regions of the brain will experience an increase in blood flow. It is this hemodynamic response that provides a signal that can be measured via functional magnetic resonance imaging (fMRI) [1
]. Mapping these changes in blood flow has significantly improved our understanding of the way the brain organizes and processes information [2
]. In 1986 Grinvald et al. demonstrated that these changes in blood flow could be detected by simply imaging the exposed cortex under optical illumination using a photodiode array [6
]. Since then, improvements in digital camera technology, light sources, and optical filters have led to widespread use of so-called ‘optical intrinsic signal imaging’ (OISI) for neuroscience research [5
]. These intrinsic signals, corresponding simply to changes in reflected light intensity, are now understood to primarily originate from variations in the concentration of oxy- and deoxyhemoglobin that occur as a result of changes in vessel diameter, oxygen delivery, and oxygen extraction [11
Blood flow changes can be detected using a wide range of optical illumination wavelengths, yet the wavelength chosen can dramatically affect the conclusions that are drawn about the spatial extent and temporal evolution of the hemodynamic response [7
]. This is because oxy- and deoxyhemoglobin (HbO2
and HbR) have unique absorption spectra and exhibit different spatiotemporal responses in the vascular compartments within the cortex (arteries, veins, and capillaries). A significant advance in recent years has been the use of multiple illumination wavelengths, allowing spectroscopic analysis and therefore direct estimation of the changes occurring in HbO2
and HbR concentrations [12
]. Adding simultaneous measurements of blood flow to this approach makes it possible to infer oxygen delivery and extraction dynamics. These hold the potential to link hemodynamic changes to the metabolic demands of the cortex and are therefore closer measures of the underlying neuronal activity [15
]. To date, blood flow dynamics in the exposed cortex have been imaged using speckle-flow imaging [17
] or evaluated at discrete points using Laser Doppler [18
Even before OISI was demonstrated, voltage sensitive dyes were used for cortical mapping of changes in membrane potential [19
]. More recently, calcium sensitive dyes have provided a means to measure changes in neuronal intracellular calcium concentrations [11
]. Both of these types of dye are fluorescent and require that the cortex be stained prior to imaging. Measurements of these dyes have often been impaired by concurrent fluctuations in hemodynamics [21
]. Therefore, an ideal cortical imaging system would allow high-speed simultaneous capture of fluorescence, HbO2
and HbR absorption, and blood flow signals.
The conventional approach to acquiring multispectral imaging data is to use a white light source, such as a halogen or mercury-xenon lamp, and band-pass filters to select appropriate wavelengths [15
]. Filter wheels are typically used to provide switching of filters, with the wheel generating a series of triggers that instruct a camera when to capture each frame. This filtered light is commonly aligned into a flexible fiber-optic conduit to allow directed illumination of the cortex of the animal. The main disadvantage of this approach is that cameras are typically not able to acquire at their maximum frame rate when driven by an external trigger. In addition, a filter wheel will generally have 6 positions, requiring purchase of a filter for each (even if fewer than 6 wavelengths are required). Unless duplicate filters are purchased, this limits the time in which one multispectral frame can be acquired to 1/6th of the triggered frame rate of the camera (e.g. 30 frames per second with 6 filters provides only 5 ‘spectral frames’ per second [16
]). This can also affect the accuracy of spectral analysis since each spectral image is captured at a different time-point. Further, even the highest performing white light sources do not have uniform spectral density, such that certain wavelengths may be less powerful than others. Galvanometer-based filter switching systems have recently become available (e.g. allowing 8 ‘spectral frames’ per second with 4 multiplexed wavelengths [14
]). While faster and more versatile than filter wheels, they are more costly and still limited by the total power and spectral range of the light source. Tunable optical filters positioned in front of the camera are typically lossy, slow, and not wholly effective in producing pure wavelength bands (essential for fluorescence excitation). Detailed spectral analysis has been achieved by imaging a linear spectrometer onto the cortical surface, however alignment is challenging and the lack of a 2D image prevents detailed analysis [18
]. Another approach is to use broadband illumination and a color camera which has a suitable Bayer mask, or a spectral image-splitter, allowing two or more bands to be detected in parallel [26
]. Color cameras typically suffer from reduced signal-to-noise owing to optical filter losses, lower bit depth, and ¼ sized pixels compared to monochrome cameras. Image splitters reduce image resolution. For both, there is the danger of sample heating and phototoxicity from high intensity broadband illumination, and the inability to acquire multiplexed fluorescence data. Simultaneous imaging of speckle-flow and optical reflectance data has been demonstrated using two co-aligned cameras and a dichroic filter to spectrally discriminate light originating from a 785 nm laser diode (for speckle) and a filter-wheel with filters between 560 – 610 nm [16
]. However, the use of multiple cameras significantly increases system cost and complexity and exact pixel-by-pixel image registration is rarely possible.
Our new multispectral imaging system overcomes these challenges at a reduced cost through the use of multiple co-aligned high power, rapidly modulated light emitting diodes (LEDs), a high-speed microcontroller-based synchronization circuit, and an inexpensive, fast, monochromatic camera. The system has been configured to allow the camera to free-run at its maximum frame rate. A signal from the camera indicating real time frame-grabs is used to synchronize sequential strobing of the LEDs via the microcontroller. Not only does this circuit allow for increased frame rates over filter wheel systems, and novel programmable strobe sequences, the broad range of LED light sources available today means that high power can be obtained at almost any wavelength. Also, the power of each LED source can be tuned individually to provide optimal signal-to-noise and dynamic range, and each LED can be individually filtered to block any wavelengths that are not required. Since each LED is illuminated in turn, appropriate emission filters in front of the system’s camera can allow acquisition of rapidly multiplexed fluorescence and absorption data within a single run.
The system is routinely able to acquire single images at > 220 frames per second (fps). For biomedical applications that require two different wavelengths, our system can acquire complete multispectral image sets at > 110 fps. This increased speed means that spectral analysis more accurately reflects the instantaneous state of the tissue. In addition, the speed and spatial resolution of our system also allows the motion of red blood cells to be discerned throughout the cortical vessels, thereby allowing the velocity of blood flow to be measured and mapped across the field of view using the same data.
In this article, we describe the design and implementation of our new multispectral imaging system and demonstrate how it can be used to quantitatively study the spatiotemporal dynamics of hemoglobin oxygenation, changes in blood vessel diameter, blood flow dynamics from the motion of red blood cells in superficial vessels of the exposed cortex, and local neuronal activity from the fluorescence dynamics of a calcium sensitive dye.