3D-printable portable open-source platform for low-cost lens-less holographic cellular imaging

Digital holographic microscopy is an emerging, potentially low-cost alternative to conventional light microscopy for micro-object imaging on earth, underwater and in space. Immediate access to micron-scale objects however requires a well-balanced system design and sophisticated reconstruction algorithms, that are commercially available, however not accessible cost-efficiently. Here, we present an open-source implementation of a lens-less digital inline holographic microscope platform, based on off-the-shelf optical, electronic and mechanical components, costing less than $190. It employs a Blu-Ray semiconductor-laser-pickup or a light-emitting-diode, a pinhole, a 3D-printed housing consisting of 3 parts and a single-board portable computer and camera with an open-source implementation of the Fresnel-Kirchhoff routine. We demonstrate 1.55 μm spatial resolution by laser-pickup and 3.91 μm by the light-emitting-diode source. The housing and mechanical components are 3D printed. Both printer and reconstruction software source codes are open. The light-weight microscope allows to image label-free micro-spheres of 6.5 μm diameter, human red-blood-cells of about 8 μm diameter as well as fast-growing plant Nicotiana-tabacum-BY-2 suspension cells with 50 μm sizes. The imaging capability is validated by imaging-contrast quantification involving a standardized test target. The presented 3D-printable portable open-source platform represents a fully-open design, low-cost modular and versatile imaging-solution for use in high- and low-resource areas of the world.


Laboratory and 3D Printed Opto-Mechanical DIHM Setups
Both developed DIHM implementations, described in the following, aim at achieving and validating the maximum achievable spatial resolution when an temporally and spatially coherent source as well as a temporally and spatially incoherent semiconductor light source at equivalent emission wavelengths are considered. First, a LD-based lens-less DIHM platform is developed that aims at demonstrating single-digit micrometer spatial resolution cellular imaging by a temporally coherent source and is schematically depicted in Fig. 1(a). It employs a 405 nm emitting LD in a laser-pickup which has been dismounted from a commercially available standard Blu-ray disc drive. The laser emission is coupled into a standard single-mode fibre. The diverging fundamental Gaussian mode beam is directed towards the object glass plate at a distance of 24.09 mm from the fibre exit facet. The CMOS camera is positioned at a distance of f = 30 mm. Second, an LED-based lens-less DIHM platform, depicted schematically in Fig. 1(c), is designed and constructed by 3D printable parts, with the over all system costs amounting to less than $190. Compared to Fig. 1(a), in Fig. 1(b) a 430 nm emitting high-power LED is employed as a temporally and spatially incoherent semiconductor light source. A fraction of the emitted light is passed through a high-precision pinhole where 1.1 μW of optical power are emitted at the pinhole for an injection current of 125 mA and impinge on objects to be imaged (z = 5.91 mm). For live cell imaging, such ultra-low www.nature.com/scientificreports www.nature.com/scientificreports/ optical power is of critical importance, as cell damage by light exposure needs to be minimised. To construct the LED-based platform in Fig. 1(b), first three mechanical parts in Fig. 1(c) are 3D-printed, see section "Methods", and assembled as sketched in Fig. 1(d,e) where also specific dimensions of the platform are depicted. For both experiments, equal spatial separations between emission facet, microscope glass plate, carrying the micro-objects under investigation, and CMOS detector are chosen. A distance of 30 mm between light source and detector is selected in order to reach a compact experimental set-up, while still maintaining illumination of the hole detector area. The position of the object emerges from resolution optimization, see section "Resolution". A Raspberry Pi single-board portable computer and a Raspberry Pi CMOS detector camera with a pixel size of (1.12 × 1.12) μm 2 serve as light source current injection driver and hologram acquisition, see section "Methods". A constant injection current of 28 mA for the LD were provided by a commercial LD driver, while a current of 125 mA for the LED was provided by a simple electrical circuit made of off the shelf components. A conventional rechargeable power bank battery pack provides electrical energy for the computer. Estimated theoretical spatial resolutions of δ LD = 0.87 μm with a LD source and δ LED = 0.92 μm with a LED light source at an available CMOS pixel size of 1.12 μm and at a distance of f = 30 mm between fibre facet and CMOS detector are expected for the LD and LED light source, respectively.

Image Reconstruction
Following the hologram acquisition, information retrieval of the cellular objects deposited on the object glass plate is performed numerically. The propagation of light fields is completely described by diffraction theory. Hence it is possible to reconstruct amplitude and phase information of the objects from their interference patterns generated on the camera. At the object location the pattern is focused and reveals the shape and morphology of micro-objects. Numerically, arbitrary planes can be re-focused in retrospect yielding access to volumes with a large number of objects in different heights in z-direction which can be studied with acquiring a single image. The propagation of a wave front towards the detector is described by the Fresnel-Kirchhoff diffraction integral det denotes the wave field on the detector, U x y ( , ) in is the incident wave, t x y ( , ) the transmission function of the object, and → = r x y z ( , , ) and → = R X Y ( , , 0) are two points in the object plane respectively detector plane 54 . The amplitude and phase distribution of the object can be obtained via an inverse integral  open-source microscope software Fiji, see "Methods" section, that implements the angular spectrum estimation for small distances in the order of micrometers up to several centimeters, is employed. The amplitude distribution in the reconstructed plane is calculated using two Fourier transforms where z is the height of the reconstructed plane, k the wave vector, n the index of refraction, N the number of pixels and p the pixel size of the CMOS detector.  and −1  denote the Fourier Transform and the inverse Fourier Transform. The formalism described above is implemented in two open-source reconstruction software packages, see "Methods" section. In the following, we elaborate and identify two easy to implement reconstruction software packages on a standard desktop computer or potentially also on a mobile phone. HoloPy, a software package for python, allows for hologram reconstruction, but also hologram simulation and scattering calculations. The algorithm to reconstruct point source holograms is based on Fresnel-Kirchhoff diffraction 1 . It considers a background subtracted hologram, experimental parameters including distances and light source wavelengths and then reconstructs the hologram using two Fourier transforms. Alternatively, hologram fitting is provided by HoloPy where the position of a scatterer is simulated in order to produce the same interference pattern instead of image back-propagation 55 . For the case of a known number of scatterers, this method can be recommended as it allows to reconstruct spherical or cylindrical object shapes. It is less practicable, however, when arbitrarily shaped micro-objects are of interest, as for example folded RBCs. For the latter case, an open-source plugin 56 for Fiji, see "Methods" section, is a possible solution with a user-friendly graphical-user-interface, implemented in Java. Phase, amplitude and intensity distribution of micro-objects can be reconstructed at arbitrary heights in z-direction.

Micro Particle and Cellular Imaging
In the following, we first capture and image standardized PMSs of diameter (6.5 ± 0.2) μm by both the LD and LED-based lens-less DIHM platform and reconstruct the resulting object properties by the Fiji plugin. Second, we investigate anonymized mature human RBCs as micro-objects in the same manner. Third, we image cell suspension culture TBY2s. The recorded holograms and subsequently reconstructed object planes for multiple PMSs and RBCs are depicted in Fig. 2. Laser-based DIHM hologram (a) and reconstruction (b) is presented next to LED-based DIHM hologram (c) and reconstruction results (d). Both insets depict an isolated PMS or RBCs enlarged to five times its original size. It becomes evident that the LD-based platform provides sharper images where a larger number of interference fringes can be captured per object. These fringes overlap within the hologram resulting in a hologram with more grainy texture as compared to the LED-based results in Fig. 2(a,c). In contrast, the LED-based reconstructed image is considerably more washed out resulting in comparably extended objects. Accordingly, for human RBCs imaged by the LD-based platform, the oval disk shape can clearly be resolved as depicted in Fig. 2(e,f). Several RBCs appear to be tilted in their spatial position, resulting in an elliptical shape. This is in stark contrast to the results obtained by the LED-based platform where information retrieval, for example on the cell morphology, are scarce. However, by the LED-based platform, individual cells can clearly be distinguished, thus exemplifying its potential for individual cell counting or tracking. In order to validate both platform's imaging capabilities also for extended cellular objects, fast growing plant tobacco TBY2s have been prepared and imaged. TBY2s are employed in various fields of plant biology as a model material and are ideally suited for cellular and molecular analyses 57 . Corresponding results are depicted in Fig. 3. The recorded holograms and reconstructed object planes for an isolated TBY2s are depicted for LD-based DIHM hologram (a) and reconstruction (b) is presented whereas the hologram, obtained by the LED-based DIHM, is depicted in (c) and the corresponding reconstruction in (d). Both platforms allow to successfully access individual cell segments with a length of 50 μm as well as internal structures including cell nuclei and vacuoles. In Fig. 3(b), a dividing cell undergoing mitosis can be observed. For LED illumination, individual vacuoles are not distinguishable. This is not surprising, as the vacuole membrane thickness is around one order of magnitude smaller than the cell wall thickness of (7-10) nm for plant vacuoles 58 as compared to (71-87) nm for tobacco leaf cells walls 59 . However it is possible to identify the nuclei of several cells. Interestingly, TBY2s infer a more complex interference pattern as compared to both PMSs and RBCs, indicating a stronger absorption and thus increased hologram contrast. We found that Fiji revealed a substantially faster reconstruction as compared to HoloPy. The reconstruction of 10 planes of a digital hologram by the Fiji plugin demands 30 seconds computational time on a regular consumer PC as compared to several minutes by HoloPy. Towards larger volumes, the reconstruction time can theoretically be improved by performing computations on a graphics processing unit 11 as demonstrated for live imaging 60 . In the following section, we aim to quantify the theoretical lateral resolution as well as the spatial resolution experimentally achieved by both the LD-based and LED-based DIHM platforms.

Resolution
In DIHM, the lateral resolution is bounded by the optical assembly numerical aperture (NA) and the illumination wavelength λ 1 : between object and detector, with the distance f between light source and detector. This originates from the consideration, that a higher resolution is possible, the closer the object is placed to the detector, however at the same time the interference fringes move closer. Here, s opt denotes the distance at which different interference rings are still resolved by different camera pixels. The axial resolution of the system can be calculated according to   Fig. 4. With LED illumination, element 1 of group 7 is the last resolvable element corresponding to a resolution of 128 line pairs/mm and a line width of 3.91 μm. For LD illumination, element 3 of group 8 is still resolvable, leading to a resolution of 322.5 line pairs/mm and a line width of 1.55 μm. This is mostly a result of the higher temporal and spatial resolution of the laser in comparison to the used LED, as well as the smaller wavelength. The resolving power with LD illumination is thus significantly higher.
Finally, to quantify the achieved image contrast and thus evaluate the capability of the developed DIHM setups, in the following the intensity of a single USAF element is averaged along its axis and plotted (see red and blue rectangle in Fig. 4(a,b). Then, the contrast of consecutive extreme points is calculated by =  Fig. 4(c,d). Results obtained by both light sources indicate decreasing contrast values with increasing line pairs. The maximum resolvable elements are as shown by the the intensity profile plots for horizontal and vertical elements 3 of group 8 in the inset of Fig. 4(c). For LED, the maximum resolvable element is element 1 of group 7 as depicted in the inset of Fig. 4(d). We compared our contrast values to reported values 62 , where an unspecified laser or a LED with a central wavelength of 470 nm (25 μm pinhole) was used for image acquisition. The reported value for laser illumination coincides with our results. With LED illumination, the reported value reaches a higher resolution, group 7 element 3 is still resolvable. This could be due to a higher signal-to-noise ratio of the employed camera, the smaller bandwidth of only 10 nm as well as the use of advanced reconstruction algorithms. The achieved spatial resolution of both setups is suited to image and detect individual micro-particles and cellular objects including ensembles of microscopic biological samples. For the LD-based DIHM, sharper interference pattern and a higher number of interference fringes could be captured. Thus, more object information is retrieved yielding a crisper reconstructed image including more details. We attribute this to the LD's higher second-order temporal coherence as well as spatial coherence. The LED's spatial coherence could be increased by reducing the pinhole diameter which would require increased LED biasing currents which then require efficient cooling of the LED and thereby increasing the physical dimensions of the setup. Our DIHMs allow to image cellular objects with dimensions smaller than 10 μm as well as microscopy of larger objects with a spatial resolution of 3.91 μm. To www.nature.com/scientificreports www.nature.com/scientificreports/ ensure maximum possible resolution, a LD-based DIHMs is recommended, whereas for applications where high temporal stability, compact set-up, ruggedness and reduced costs a required, LED-based DIHMs is recommended.

Conclusion
A 3D printable platform for lens-less holographic cellular imaging with open accessible software solutions has been developed delivering spatial resolutions of 1.55 μm by LD or 3.91 μm by LED illumination. A 405 nm Blu-ray semiconductor laser-pickup coupled to an optical fibre and a 430 nm high power LED in combination with a 15 μm pinhole have been successfully employed as DIHM light sources. Despite its lower degree of temporal coherence, the LED proved to be of advance in terms of implementation, price and lower safety concerns. A single-board portable Raspberry Pi computer and camera operate the light sources as well as perform the image acquisition. By an open-source software implementation of the Fresnel-Kirchhoff algorithm, we imaged and successfully reconstructed 6.5 μm PMS and human RBCs with a diameter of about 8 μm, as well as TBY2s with an individual size of about 50 μm. Less than 1.1 μW of optical power were sufficient for holographic imaging microscopy. Such ultra-low optical power can be of critical importance for live cell imaging where light exposure of the cells needs to be as low as possible. Equally compact setups could be envisioned for the LD when fibre-coupled LDs are available, which are however considerably expensive. The DIHM setup presented here may serve as a reliable, easy to implement and flexible to extend solution for student an early researcher education and for different demands in microscopic imaging. The total costs for the LED setup amount to $190 ($3 LED, $75 Pinhole, $35 Raspberry Pi 3, $25 Raspberry Pi Cam v2, $25 3D print, $27 power bank) and thus enables a convenient entry into the wide field of digital holography. Future work could include the integration of the DIHM with micro-fluidic channels 48,63,64 or considering machine learning algorithms to automatically count and identify particles 65 , as well as diagnose illnesses such as meningitis 66 , iron-deficiency anemia or diabetes mellitus 67  μm for a Gaussian intensity distribution, with Δλ the full width at half-maximum spectral line width 70 . A high-precision stainless steel pinhole with a diameter of (15 ± 1.5) μm ($75, P15D, Thorlabs Inc.) has been employed as all 3D printing attempts yet did not yield a necessary sufficiently high mechanical grade circular diameter. All neccessary parts for the LED setup can be found at https://github.com/teph12/DIHM. All parts and Raspberry Pi housing were printed with a commercial 3D-printer ($880, Prusa i3 MK3, Prusa Research s.r.o.) using standard polylactide synthetic polymer filament with 1.75 mm diameter. A spatial printing resolution of 0.4 mm, parallel to the optical table, and 0.05 mm, in vertical or z-direction is available. Within the printed DIHM housing, LED and pinhole were fixed with tape ($9, 3M-ID 70005241826 Scotch Magic Tape, 3M Inc.). The case with VESA mounts for Raspberry Pi 3 (B/B+), Pi 2 B, and Pi 1 B+ can be accessed by 71 . The CMOS camera module is glued to the upper part of the box after removing the lens mounted in front of the module. We observed that otherwise strong hologram distortions appeared. In order to electrically bias the LD, a commercial LD driver was used to provide a constant output current. However, an open-source driver is under construction while all parts are available for in total $20 72 . For the LED, a custom soldered circuit has been developed accessing Raspberry Pi's general purpose input/output (GPIO). The assembled system is depicted in 5b). The driver circuits can well be integrated into the 3D printed Raspberry Pi housing. A diagram of the circuit can be seen in Fig. 5(a). The portable computer, camera and DIHM light sources are supplied with electrical power by a conventional power bank battery pack with maximum output power of 10 W ($27, Aukey PB-N36). A computer monitor and computer mouse are required for the hologram acquisition. V_BATTERY is the voltage provided by the power bank, whereas V_ TRIGGER is the voltage between the Raspberry Pi GPIO and ground. It triggers a transistor (BD135), which lets a current of 125 mA flow through the LED. . They are diluted in saline solution to a high degree (up to 0.02%). The employed polystyrene micro-spheres ($129, BS-Partikel GmbH, Germany) have a diameter of (6.5 ± 0.2) μm. They are equally high diluted in distilled water. Furthermore a few ml of ordinary dish detergent are added to prevent aggregation and adhesion of the micro-spheres. After diluting the particular object a drop of the solution is placed by a standard plastic pasteur pipette on a high transmission flat glass microscope slide (75 mm × 25 mm × 1 mm, B270 I, SCHOTT AG) and then covered with a glass cover plate (22 mm × 22 mm × 0.15 mm, B270 I, SCHOTT AG). Nicotiana tabacum cv. BY-2 suspension plant cell cultures 73 were grown in liquid saline medium based on a modified Linsmaier and Skoog medium with agitation on a incubator shaker. The cells have been grown in 50 ml medium within a 250 ml glass flask. Every 7 days, 5% inoculum had been transferred into a fresh medium 74,75 and stored permanently on an incubator shaker. Image acquisition, reconstruction and resolution validation. The pictures are acquired with a fixed white balance. For details see the camera file on https://github.com/teph12/DIHM. The image is then transferred to a PC with Windows 10 operating system and equipped with an Intel i3 processor and 8 GB of RAM, with Fiji installed. In Fiji the image is converted to a 32-bit black and white picture, which is then loaded in the Numerical Reconstruction plugin. Here, using the parameters of image acquisition (distance between camera and object, wavelength, image size) the image is reconstructed. Subsequently the image contrast is normalized and enhanced by 0.2%. For the quantification of the experimentally achieved spatial resolution, a commercially available standardized 1951 USAF positive high-contrast chrome on quartz glass microscopic imaging test target created by photo lithography on a glass microscopic slide serves as a reference object ($900, Ready Optics, US). It consists of groups of horizontal and vertical lines with standardized spatial frequencies starting at group 4, element 1 with 31 μm spacing and ending at group 11, element 6 with 137 nm spacing.