Wide field-of-view lensless fluorescence imaging device with hybrid bandpass emission filter

Lensless microscopy makes it possible to build wide field-of-view devices with small dimensions due to its inherently simple optics.^1–3 Thus, it is expected to be applied to various areas, for example, time-lapse imaging of culturing cells in the small space of a CO₂ incubator^4,5 or point-of-care devices with high portability. For the purpose of specific cell observations or molecular activity measurements, fluorescent labels are very powerful tools. However, in the field of lensless imaging, few studies have examined fluorescence imaging. One of the reasons is poor performance of emission filters for lensless imaging devices.

In lensless fluorescence imaging devices, it is typical to use an interference filter^6,7 or absorption filter,^5,8–15 as well as conventional lens-based fluorescence microscopes. However, both of them have problems in a lensless setup. Recently, we proposed a novel “hybrid” emission filter that uses the advantages of the interference and absorption filters complementarily.¹⁶ This filter shows excitation light transmittance of approximately 10⁻⁸ with a lensless setup. Thus, highly sensitive fluorescence detection performance similar to that for a lens-based fluorescent microscope has been achieved.

The hybrid filter shows high excitation-light rejection performance. In some cases, the limit of detection is not determined by the leakage of the excitation light but by other causes. For example, a mouse brain shows some auto-fluorescence over a wide emission spectrum with light excitation. If the observation target is green fluorescent protein (GFP), the auto-fluorescence from the yellow to red band reduces the fluorescence detection performance.

In this study, we propose to improve the filter performance using the advantages of interference filters, which can realize a high wavelength selectivity by designing the effective thickness of the layers. Another filter to reject the unwanted fluorescence components was added to improve green fluorescence detection. The proposed filter was fabricated on a fiber optic plate (FOP) as a substrate and was easy to handle. Thus, it is possible to combine the filter with an image sensor assembled as a package. We demonstrated a wide field-of-view lensless fluorescence imaging device based on the commercially available image sensor shown in Fig. 1(a). The imaging area of the new device is 67 mm², which is approximately 37 times wider than that in our previous works.^15,16 As a demonstration, a GFP-developed mouse brain slice was observed. The wide field-of-view feature enabled us to image almost the whole brain slice. Also, we demonstrated the improvement of GFP imaging performance by the effect of the bandpass characteristics of the new hybrid filter.

Figure 1(a) is a photograph of the fabricated device. The dimensions of this device are 60 mm × 30 mm × 17 mm. The image sensor is a commercially available product (CMV2000, AMS, Austria) with a pixel count of 2,048 × 1,088 and a pixel size of 5.5 μm. The glass lid of the sensor was removed to add the bandpass filters.

Figure 1(b) shows a schematic of the device cross-section. The sensor chip is connected to the electrodes on the package, with the wires placed above the chip surface. To shift the focal plane, we used an FOP (J5734, Hamamatsu Photonics, Japan), which is composed of a bundle of optical fibers and transmits images from one side to the other. The FOP is 2.54 mm thick and has a core pitch of approximately 3 μm, which is about half of the image sensor pixel size. It was also used as the substrate for the interference filters. A high-quality interference filter can be fabricated on the FOP because it is based on glass and has high stability.

The filter is composed of three parts. A longpass interference filter is deposited on the top to reflect most of the normal incident excitation light. This filter is placed on the top because the FOP emits auto-fluorescence that can decrease performance. Also, the reflected excitation light irradiates the observation target again, which almost doubles the fluorescence intensity.

The second part is on the bottom of the FOP, where a shortpass interference filter is deposited. The expected incident light reflected by this filter is the auto-fluorescence from the target, which is much weaker than the excitation light. Thus, the FOP fluorescence with the reflection band can be negligible.

The third part is an absorption filter. In a lensless imaging device, the excitation light is scattered by an observation target and irradiates the filter at various incidence angles. Because the transmission spectrum of an interference filter shifts with incidence angle, high-excitation light rejection, as accomplished in lens-based fluorescent microscopy, cannot be achieved. Thus, in the hybrid filter, the scattered component is rejected by an absorption filter. Here, it should be noted that absorption filters also emit fluorescence. The excitation light rejection performance can be improved with a thicker absorption filter, but there is an effective limitation due to auto-fluorescence from the absorption filter. Thus, a hybrid filter structure is required to achieve ultra-high excitation light rejection. The fabrication method is similar to that discussed in our previous paper,¹⁶ which contains a detailed description.

In this structure, stray light from the sides of the FOP or the image sensor would create artifacts. To avoid them, a shield coating of black paint (CS-37, Canon Chemical, Japan) was applied.

Figures 2(a) and 2(b) show linear and logarithmic plots for each filter component. The expected wavelength of the excitation light is 448±10 nm. The longpass interference filter exhibits a transmittance of approximately 10⁻⁴ in this wavelength range. The designed filter thickness is approximately 2.3 μm.

The absorption filter is composed of yellow dye (Valifast Yellow 3150, Orient Chemical, Japan) and a polymer matrix (NOA63, Norland, USA). The weight ratio of the dye to the matrix is 1:1. The thickness is 4.1 μm, at which the auto-fluorescence of the filter is comparable to that of the FOP. The transmittance of the absorption filter is smaller than 10⁻³ in the excitation band.

The cut-on wavelength of the shortpass filter is 560 nm. Its designed thickness is approximately 4.2 μm, which is intended to transmit the GFP emission band and reject the auto-fluorescence components from an observation target. The rejection band is wide and up to around 850 nm. Thus, angled red light is still rejected with this filter.

Figure 3(a) shows a sensitivity spectrum comparison of the lensless imaging devices with the previous (longpass) and the present (bandpass) hybrid filters. The curve with green shading under it is the GFP emission spectrum.

The result shows that the previous filter rejects the excitation band. However, Si-based image sensors are sensitive from the visible to near-infrared bands, and the sensitivity for the red band is higher than that for the green band. With the proposed hybrid bandpass filter, the sensitivity to the red band is successfully reduced. Thus, the sensitive band is matched to the GFP emission band.

Figure 3(b) shows the sensitivity spectra for light with different incidence angles. The spectrum of the interference filter shifts shorter by declining the angle, but the spectrum of the absorption filter does not change. As a result, the transmission band of the proposed filter is narrowed for larger angles of incidence. Although this effect also reduces the sensitivity to the target fluorescence, the spatial resolution is improved a little by reducing the effective numerical aperture.

Commonly, an interference filter provides a rejection band. Thus, a shortpass filter transmits light at wavelengths longer than the rejection band. It creates this shift by decreasing the incidence angle. However, even with an incidence angle of 45 degrees, a longer side edge of the rejection band was not observed in this experiment. Thus, the fabricated hybrid filter has sufficiently high rejection performance in the red band.

We checked the spatial resolution of the fabricated device by using a USAF 1951 test target (R1DS1N, Thorlabs, USA). To prepare conditions similar to those for green fluorescence imaging, a green LED (NSPG300D, Nichia, Japan) with a peak wavelength of 525 nm was used as the light source. To reduce the directivity of the incident light, an opal optical diffuser (83-389, Edmund Optics, USA) was inserted between the light and the device. A magnified captured image is shown in Fig. 4. The result shows that the resolution of the present device is 36 line pairs/mm for omnidirectional illumination. This value is similar to that for our previous device.¹⁶ 

Figure 5 shows a fluorescence image of a mouse brain slice obtained with the fabricated device. The specimen is from an adult GAD67-GFP mouse. The thickness of the slice is 100 μm. All procedures for preparation of the animal tissue conformed to the animal care and experimentation guidelines of the Nara Institute of Science and Technology (Approval ID: 1532). The slice was placed directly on the surface of the device. The excitation light was irradiated from above the device. Thus, the system is a type of transmission microscope. The imaging area is 11.26 mm × 5.98 mm (= 67.40 mm²). Almost the whole slice can be observed in a single shot. The image was processed to improve its quality.

The magnified images for each step of the processing procedure are shown in Fig. 6. Figure 6(a) is the raw image of a single shot taken with an exposure time of 15 ms. Random temporal noise is apparent. Figure 6(b) is an image averaged over 100 sequential frames, which drastically reduced the random noise. The image sensors and the FOPs have pixel or core arrays with different pitches. Also, we stacked two FOPs with their optical cores unaligned. Thus, a fixed hexagonal pattern of noise appears in the images. Figure 6(c) is a two-dimensional fast Fourier transform of the whole image. It should be noted that the intensity is on a common logarithm scale. This image contains some bright points corresponding to the hexagonal pattern in Fig. 6(b). Thus, the peaks were filtered out with the mask indicated as small red squares in the image. Fig. 6(d) is the final processed image obtained by inverse Fourier transformation. In this image, the periodic fixed noise pattern is apparently reduced, and small features, such as fluorescent cell bodies, can be observed.

Figure 7 compares the imaging results with the two types of lensless devices and a lens-based fluorescent microscope. The images obtained with the lensless device with the hybrid longpass filter and a 2-ms exposure time are in Figs. 7(a, b). Figs. 7(c, d) are those obtained with the proposed hybrid bandpass filter, with an exposure time of 15 ms. The image processing described above was applied to these lensless images. Figs. 7(e, f) show the reference image obtained with an epifluorescence microscope. The imaged areas are the hippocampus and reticular thalamic nuclei. In both of the hippocampus images obtained by the lensless imaging device, bright GFP fluorescent cell bodies can be observed. However, with the bandpass filter, the image is clearer and the number of detectable cell bodies is higher. Comparison of the images around the thalamic nuclei (Figs. 7(b, d, f)) shows that the image contrast with the bandpass filter device is similar to that with the microscope. These results indicate that the auto-fluorescence components were reduced and GFP signals are detected more clearly than with the longpass filter device.

In the proposed device, interference filters are used to realize the bandpass characteristics. One of the advantages of interference filters is that they have a highly flexible design capability for changing the transmission spectrum. It is difficult to realize bandpass transmission characteristics using an absorption shortpass filter because of its relatively low wavelength selectivity. This is also an advantage with respect to a lensless fluorescent imaging setup using a prism.^17–20 Although the total internal reflection by the prism can reduce the excitation light transmission, it has no spectral selectivity. In our proposed filter structure, the red band is removed by the interference filter with a high selectivity that cannot be achieved with absorption dyes. Thus, the higher fluorescence sensitivity at the green band was realized. Also, the hybrid filter structure is expected to be applicable to multicolor fluorescent imaging by incorporating a multi-spectrum transmission filter with a special interference filter design.

One of the advantages of lensless imaging is that a wide field-of-view can be achieved with a relatively small device. The imaging area is basically limited only by the size of the image sensor. The present filter is physically stable because an FOP is used as the substrate. Thus, imaging devices can be fabricated by simply affixing the filter device to a commercial image sensor after its glass lid is removed. The area of the present device is 67.4 mm², which is 27 times larger than that in our previous work.¹⁶ It is not difficult to enlarge the device by using, for example, the APS-C (23.6 mm × 15.8 mm) or 35-mm (36 mm × 24 mm) format image sensors that are widely used in commercially available digital cameras.

The imaging results for the test target shows that the resolution is approximately 36 line pairs/mm. This value is acceptable in some biochip applications because the required resolution can be controlled by designing the array pitch of the small fluorescent chambers.^14,21–24 However, it is almost the lower limit of the resolution needed to observe cell bodies. Indeed, although fluorescent cell bodies were successfully observed in Fig. 7(c), they were sparse. It is difficult to observe cells with a high density.

Usually, fluorescence is emitted omnidirectionally. Thus, the resolution decreases with the distance from the target or the thickness of the filter. Also, the contrast of fine structures is decreased as a result of blurring. In the present device, the thickness of the absorption filter was optimized. However, by inserting the shortpass filter, the total thickness is similar to that of our previous hybrid filter. The resolution can be improved by reducing the numerical aperture for fluorescence wavelengths.²⁵ The disadvantage of this method is its low transmittance.

Other possible ways to improve the spatial resolution are to use a patterned excitation light source^5,9,26 or a custom shield mask.²⁷ The proposed filter is compatible with these techniques because it is assumed that the excitation light is normally incident. This is also one of its advantages over prism-based devices,^17–20 which require a large prism for large imaging area. Also, as demonstrated in our previous work, imaging in bright field mode is useful for observations at high spatial resolution.^16,28

In this work, we intended to obtain high excitation light rejection performance. Thus, we adopted an absorption dye that has relatively low auto-fluorescence. However, the transmission at the wavelength of the GFP emission peak is not as high as that shown in Fig. 3(a). In some cases, the light emission from the target is sufficiently intense and low excitation light power is more important than the wavelength selectivity. By using a dye with shorter absorption band, it is possible to improve the light sensitivity of the device.

The hybrid filter structure realizes high fluorescence sensitivity even in lensless imagers, and fluorescence from GFP-developed cells in a mouse brain was successfully observed. By applying this technique to implantable sensors,^29–33 it is expected that the fluorescence imaging performance can be improved drastically. The advantages of implantable devices are their small size and light weight in comparison with miniature microscopes³⁴ or fiber photometry devices.³⁵ Also, they are suitable for cross-section imaging. However, to achieve low invasiveness, the FOP should be as thin as possible. The light source has to be placed in the side or above the sensor. Thus, observation of thick samples with high resolution requires a special excitation setup.

In this work, we demonstrated a wide field-of-view lensless fluorescence imaging device with bandpass characteristics. The present hybrid filter has excitation light rejection performance as high as that achieved in our previous study. The bandpass characteristics are achieved by combining shortpass interference filter layers, which is not possible by the use of total internal reflection. The proposed filter was mounted on a large sensor with an imaging area of 67 mm². The FOP substrate shifts the focal plane from the image sensor surface and protects the shortpass absorption filter. We demonstrated fluorescence imaging of a GFP-developed brain slice, and almost the whole slice could be observed with the fabricated device. With the bandpass characteristics, the auto-fluorescence component of the slice was reduced. After image processing, the somas of neural cells were successfully observed.

The dimensions of the device were 60 mm × 30 mm × 17 mm. This size is suitable for placement on the shelf of a CO₂ incubator and would be also useful for time-lapse imaging of cells or culturing tissues. It is expected that this technique will make it possible to observe various phenomena that are overlooked by other methods.

The fabrication method of the hybrid bandpass filter is basically the same as that of a longpass filter. Details are described in our previous paper.¹⁶ In this work, a longpass interference filter was deposited on the top side of an FOP, and a shortpass filter was deposited on the bottom side. The fabricated filter was easy to handle and thick enough to place on a commercially available image sensor. The fabrication procedure is described below.

1. Interference filters were deposited on an FOP (J5734, Hamamatsu Photonics, Japan) by a filter company (Tac Coat, Japan).

2. The imaging area of the FOP was adjusted by lapping the sides to avoid interference from the wires of the image sensor.

3. A cover glass (24 mm × 24 mm) was spin-coated with CYTOP-M (AGC Chemical, Japan) at a rotation speed of 3,000 rpm. The coated film was cured by heating. CYTOP is a fluoropolymer that was used as a releasing film.

4. A mixture of Valifast yellow 3150 (Orient Chemical, Japan), cyclopentanone, and NOA63 (Norland, USA) with a weight ratio of 1:2:1 was prepared. Surfactant (R-41, DIC, Japan) was added at 0.5% by weight to increase the wettability of the CYTOP.

5. The dye mixture was spin-coated onto the CTYOP at a rotation speed of 2,000 rpm.

6. The coated film was cured by UV irradiation. This procedure reduced shrinking of the film during heating.

7. The film was cured by heating at 150 °C for 30 min.

8. The FOP with the interference filters was bonded to the film with epoxy resin. The epoxy was cured at room temperature for 24 h.

9. The edges of the cured epoxy and the dye film were cut with a knife. The film was easily removed from the CYTOP to complete preparation of an FOP with a hybrid bandpass filter.

10. An image sensor module without a glass lid was prepared. The method for removing the lid is described in Section V C. The image sensor used in this work was model CMV2000 (AMS, Austria).

11. The FOP with the filter was bonded to a sensor pixel array with epoxy resin. The bonding wires were also molded with epoxy.

12. To shield the sensor from stray light, the side of the FOP was coated by painting it with black paint (CS-37, Canon Chemical, Japan).

The experimental setup used a commercially available epifluorescence microscope (BX51WI, Olympus, Japan). The light source was a mercury lamp (U-RFL-T, Olympus, Japan). A filter set for GFP (U-MGFPHQ, Olympus, Japan) was used. The excitation filter was changed to one appropriate for our hybrid filter (FF01-448/20-25, Semrock, USA) to increase the excitation rejection performance. The lensless imaging device was set on the stage of the microscope. A 100-μm-thick perfusion-fixed brain slice from a GAD67-GFP mouse was placed on the device and covered with a slide glass to avoid drying. The light source was also used for the lensless device. The objective lens was not used for the lensless imaging experiment to illuminate a wide area. This microscope was also used to obtain the reference images.

For imaging with the microscope, the sample was flipped over, and a cover glass was placed on the slice. An objective lens (5×, NA=0.1, MPlan N, Olympus, Japan) was used.

Usually, an image sensor package is sealed with a glass lid. In the present work, we prepared a commercially available image sensor by removing the lid from its package mounted on a printed circuit board (PCB). The procedure is described below.

1. The PCB except for the image sensor was covered with Cu tape to protect the PCB surface coating and solder from flame.

2. The adhesive bonding the lid to the package was carefully heated through the lid with a pen-type burner. Interference fringes appeared as the lid moved away from the package as a result of heating.

3. After all the adhesive lost its bonding force, it was easily removed by a knife.

This work was supported by the Murata Science Foundation, the Support Center for Advanced Telecommunications Technology Research, the Japan Society for the Promotion of Science (Kakenhi Grant 18H03519), and the Japan Science and Technology Agency, Core Research for Evolutional Science and Technology Program.