Fluorescence imaging under background light with a self-reset complementary metal– oxide–semiconductor image sensor

The authors propose and demonstrate the fluorescence imaging of green fluorescence protein (GFP) expressed in a brain slice with a self-reset complementary metal–oxide–semiconductor image sensor under background light. By using a self-reset function to avoid pixel saturation, the weak fluorescence of GFP was successfully observed, even under background light. The result suggests that the sensor can be applied to in vivo imaging of laboratory animals during light–dark cycles, so that they can observe the different responses to bright and dark environments.


Introduction
An implantable complementary metal-oxide-semiconductor (CMOS) image sensor enables in vivo imaging even if a laboratory animal implanted with the sensor is moving freely.In previous works, we demonstrated that the sensors can image signals of blood flow correlated with neural activities on a brain surface under freely moving conditions [1].Functional brain measurements with a freely moving animal can provide insight into the natural behaviour of animals [1][2][3][4][5][6][7].
Fluorescence imaging is a widely used imaging method for studying brain functionality because many fluorescence indicators or florescent proteins have been developed [8].Implantable sensors with optical filters have been applied to fluorescence imaging [9,10].
One of the remaining problems is that fluorescence imaging must be performed in the dark because the fluorescence is usually weaker than ambient light.In the field of physiology, much attention has been paid to the difference in responses in bright and dark environments [11][12][13].Thus, fluorescence imaging of neural activity in a bright environment is a key challenge.A high-dynamic-range image sensor realises imaging under high-intensity-exposure light.However, it exhibits a restricted peak signal-to-noise ratio (SNR) based on the capacity of the sensor's pixel capacitance [14].
In this paper, we propose to use an implantable CMOS image sensor with a self-reset function for detecting fluorescence.The sensor realises high SNR imaging in a bright environment by avoiding pixel saturation.Moreover, to demonstrate fluorescence imaging in a bright environment, we practically imaged green fluorescence protein (GFP) expressed in a brain slice under white background light which is intense enough to saturate a normal image sensor's pixels.

Self-reset image sensor
A self-reset function avoids pixel saturation with overexposure [15][16][17].In this work, we focused on reducing pixel size and improving the SNR of the self-reset image sensor.Resolution of the implantable image sensor relies on the pixel size of the image sensor.Figs.1a, b and Table 1 show a schematic diagram of the pixel circuit, a chip photograph and specifications of the self-reset CMOS image sensor, respectively.To reduce the pixel size, we designed the self-reset image sensor without a counter for the number of self-resets in the pixel circuit.The transistor number per pixel is ten, and the pixel size is 15 × 15 µm 2 , which is much smaller than that of other self-reset sensors [15][16][17].The pixel size of the developed sensor is small enough to be used for measuring brain function.
The image sensor was designed to be inserted onto the surface of rodent's brain and its size is 1050 × 3000 µm 2 .The brain tissue damage of the device implantation was negligible for functional brain measurements [18].The entire image sensor will be coated with a parylene film for waterproofing to implant into living animals [1].
Figs. 2a and b show the output and noise performances of the self-reset image sensor as functions of light intensity.The result shown in Fig. 2a indicates that the pixel outputs are reset as the light intensity exceeds its threshold value.This function avoids saturation of the pixels.
To avoid saturation with a normal image sensor, the number of incident photons to the pixels should be decreased by decreasing the exposure time or light intensity.The noise level can be reduced by frame averaging.However, noises from digital circuits and a quantisation noise generated by an analogue-to-digital converter are added at each frame.With the self-reset sensor, these noises are reduced by the square root of the number of frames because frame averaging is not required.In other words, the effective signal level can be amplified by increasing exposure time, and the noises are reduced.In particular, quantisation noise reduction is effective for fluorescence imaging under background illumination because the signal level of the fluorescence is usually much smaller than its offset level originated by the background.

Fluorescence imaging under background illumination
We performed an experiment of fluorescence imaging with and without background light.Fig. 3 shows the schematic diagram of the experimental system.As a fluorescent sample, we prepared a mouse brain slice that expressed GFP in neural cells.
A pair of objective lenses was installed between a sample and the sensor to magnify the image by 3.3 times, because the size of the neural cells of the sample was almost the same size as the pixels.This large optical set up was used only for the evaluation experiment and is not necessary in in vivo imaging.The sample was a preparation of a 40 µm thick brain slice of GFP-expressed mouse.The light sources of excitation and background illumination were a blue lightemitting diode (M470L2, Thorlabs) and a halogen-tungsten lamp (CLS 150 XD, Leica), respectively.The excitation light was focused on the observation area by a plano-convex lens to increase light density, and the background illumination was diffused by an opal diffuser (DFO-50C03-1, Sigma koki) to provide uniform illumination.
The pixel array of the sensor was coated with an absorption fluorescence filter composed of yellow dye (VALIFAST YELLOW 3150, Orient chemical).The sides of the sensor were shielded with a black resist (CFPR BK-4611, Tokyo Ohka) in order to suppress unexpected photoelectric effects of the incident light.
Fig. 4 shows the imaging results of the experiment.The exposure time was 75 ms.The intensities of the excitation and background lights were set to 18 mW/cm 2 and 16 lx, respectively.The 16 lx correspond to an output signal intensity of 38,000, and self-resets occur four times in each frame.Fig. 4a  The noise in Fig. 4d is greater than the noise in Fig. 4a.The noise originates from the shot noise of background light and from selfreset noise.Since these are random noises, the noise level can be reduced by frame averaging.
The black points in Fig. 4d result from the different number of self-resets that occur with and without exposing the excitation light.The black points can be easily identified from other significant signals because they have negative values.The actual subtraction values can be calculated by adding the maximum values of the pixels.

Conclusion
We proposed an imaging technique using a CMOS image sensor with a self-reset function, and demonstrated that the fluorescence of brain tissue is detectable under background light by subtracting the images with and without exposing the excitation light.The size of the self-reset image sensor is 1050 × 3000 µm 2 and the   pixel size is 15 × 15 µm 2 .The size of the self-reset image sensor is as small as the implantable image sensor we have used in in vivo imaging [1].Using the self-reset sensor, the fluorescence was successfully observed under the background light by avoiding pixel saturation.Thus, this method is expected to continuously measure brain activity of freely moving laboratory animals under lightdark cycling conditions.

Acknowledgment
This work was supported by the Semiconductor Technology Academic Research Centre (STARC).
is a control fluorescence image obtained without background illumination.The GFP-positive cells are clearly observed as white points.Figs.4b and c are background and fluorescence images, respectively, under the background illumination.Fig. 4d is the subtraction image of Fig. 4b from Fig. 4c to extract the fluorescence components.If we compare Figs.4a and d, the fluorescence of GFP-positive cells is also clearly observed in Fig. 4d.

Fig. 4 Fig. 1
Fig. 4 Brain slice images taken by the self-reset image sensor a Excitation light (control image without using self-reset function) b Background light c Excitation and background light d Subtraction image of (b) from (c)

Fig. 3 Fig. 2
Fig. 3 Experimental system for fluorescence imaging with background illumination

Table 1
Specifications of self-reset image sensor