Team:ETH Zurich/Microscope
Content
Introduction
Table
Model
Human Practices
Achievements
Microscope.
Scroll
Throughout the project we pursued and experimented with multiple imaging approaches. On the final version of AROMA we used the brightfield motility imaging approach, corresponding to Approach A , presented on the left hand side. Furthermore we present a possible improvement of this brightfield approach, namely a lensless microscope which would come at a reduced height and cost but so far does not provide enough resolution yet. For our luminescence approach B we also created an imaging solutionshown on the right. Although the requirements are pretty different for all of those three imaging approaches, the hardware implementation is always based on our modular imaging casing system and the same imaging chip.
Brightfield Imaging Setup
Luminescence Imaging Setup
Lensless Imaging Setup
Figure showing our different imaging setups.
Brightfield Microscope
Introduction
Brightfield imaging is the most commonly used microscope setup. One illuminates the sample from one side and images from the other. The attenuation of light at the sample is then visible as a dark spot on an otherwise bright background. The benefit of using this approach is the relatively simple design and the fact that the concepts are well-known and have been applied for a very long time. Additionally, having a small focus plane results in fewer artifacts from the microfluidic chip and dirt and allows to precisely focus on the plane of E. coli. The only downside of this approach is the limited contrast. Biological samples often do not provide a strong contrast, which is also the case for E. coli. This can be explained with the attenuation not being strong enough. Nevertheless, we were keen to test this method and optimize it to image E. coli in a quality that makes it suitable for our image processing algorithm. Meaningly, to resolve them on a single-cell level.
Theory
Illumination
The illumination is a crucial part when it comes to microscopy. In brightfield microscopy a LED is used as a light source which should illuminate the source as evenly as possible. Not only the intensity but also the direction of the light rays are important. Ideally they should be perfectly parallel. We looked into different techniques for achieving such a lighting, such as the Koehler illumination. During our experiments we found out, that due to our special evenly illuminating LED we did not need to add additional lenses. Therefore we ended up placing the LED at a distance of around 10 cm from the sample. Without adding complexity to our setup this ensures even lighting and satisfies our requirements.
Magnification
Brightfield Microscope Setup
Implementation
Set-Up
To build the brightfield microscope we started to design and develop all the necessary parts (LCCSM system) which were then 3D printed. The LCCSM is the basis for our brightfield microscope and we customized and optimized every single part. We started by mounting the camera sensor inside the base plate. Then we stacked a height segment of 5 cm and then mounted the 5 cm Thorlabs mount part.
In that mount we can fit the rail system of Thorlabs. Both of our lenses are mounted to the rails which allows an easy adjustment of their height. Ideally one places the lenses in a way such that the they face each other with the flat side in order to reduce aberrations. As already described in the theory section the less curved, tube lens is placed facing the sensor and the objective lens facing the sample. We used a 250 mm and 24 mm lens so the distance in between the sensor and the tube lens should be exactly 250 mm and the objective lens should be 24 mm away from the sample, while the distance in between the lenses does not have any effect. This results in an exact magnification of 10.4x.
We would not recommend using a higher magnification as the setup becomes even taller (when using another tube lens) or the sharpness is reduced due to aberrations (when using another objective lens). The rail system of Thorlabs is then enclosed in our casing system which protects the apparatus from external spill light.
Approximately 24mm behind the objective lens, the microfluidic chip is mounted. The mount allows to move the chip horizontally, making it possible to image different channels. On top of that we put the lid, another 10 cm part and finally the part containing the LED lightsource. The total setup can reach a height of up to 45 cm, but depending on the magnification and light source system this is usually lower.
Illustration Video
Parts Used in our brightfield setup.
| 3D Parts Used | Optical Elements Used | |
|---|---|---|
| 2x 10cm, 5cm, 1cm section | Thorlabs 8x 10cm rods | |
| Thorlabs top and bottom adapter | Thorlabs Rod Connector CP02/M | |
| Microfluidic chip case | Tube Lens - f=200mm plano convex LA1708 | |
| Filter section | Object Lens - f=25.4mm plano convex LA 1951 | |
| Top LED section | Bright LED (LED-CO) Distrelec |
Focusing
Focusing the two lenses ended up being one of the hardest challenges at first. It is extremely difficult to measure the exact distances from the lens to the sensor or sample and one therefore needs different, more precise techniques. The tube lens, with the long focal length, which sits above the sensor, can be calibrated in the following way: One only mounts this single lens in front of the camera. One then points the whole thing towards a distant object, 50m or more away. You then adjust the distance of the lens to the camera sensor until you see a sharp image of the distant object. The theory behind this is simple. The distant object is so far away that focusing to it essentially means focusing on infinite rays coming into the lens. And as the basics of lens theory tell us infinite rays are directed through the focal point of a lens resulting in a sharp image.
In order to focus the objective lens we needed to assemble the whole microscope as explained above. We then have two possibilities. We can either move the sample up and down or the objective lens. We tested both strategies. First, we built a height adjustable sample tray, which worked, but ended up being not very practical, as one ideally wants the sample to stay at a place and additionally it makes it harder to add a function to move the sample from left to right in order to swap the microfluidic channel. [TRALALA Microfluidic Link] Due to those reasons we stuck to the second possibility: moving the objective lens. We implemented this by simply adding a longer thread to the lens which is anyways screwed into the Thorlab mount. This way we don’t screw the lens thightly in, but adjust its focus by screwing it further in or out. Sometimes it is these small things making a self built microscope setup work.
Robert focusing the tube lens to infinity
Optimization
The bandpass filter mounted on the modular microscope
Comparison
In order to better classify our brightfield imaging solution, we compare our results to two commercially available solutions. In terms of compactness our solution is roughly comparable to USB, handheld microscopes. In particular, we compared our own setup with the DinoLite high magnification device, the AM4515T8, which is to date, one of Dino Lite’s highest magnification devices and costs around 700 Euros. While the AM4515T8 is more compact than our device, in our experiments we could not resolve E.coli at all making it unsuitable for our approach. Bigger, high-end lab microscopes obviously achieve a way better resolution and overall imaging quality. However, their price is significantly higher and they cannot be packed onto a small robot like AROMA.
Results
We achieved great results with our brightfield microscope. We were able to clearly resolve E.coli with our own LCCSM system and a cheap monochromatic imaging sensor. This allowed us to further process the images and determine the motility of the tiny bacteria. See motility image analysis. We would like to particularly highlight that we do not know any commercially available platform at a comparable price that achieves the same resolution capabilities. Furthermore we want to point out that with our LCCSM system even higher magnifications are possible, when sacrificing the compactness.
E. Coli imaged with our brightfield imaging setup after adding contrast
Lensless Microscope
Introduction
While our brightfield imaging approach generated good and reliable results, the underlying concepts are already well known and have been applied fora very long time. As we want our biosensor to be as cheap and compact as possible, we also did some research on how the brightfield approach could be further improved or expanded. Throughout this research we came across the field of lensless microscopy, which to us seems the perfect evolution in combination with the motility imaging problem.
However, as there are only a few groups in the world using lensless devices we also knew that it would be extremely difficult to get such a device running. Our first experiments on the lensless approach went very well and we were able to resolve 8 µm small yeast cells. However, as the follow-up experiments and improvements did not lead to the expected results, we decided to create a model of this lensless system. This model actually perfectly confirmed all our experimental data but unfortunately also predicted difficulties to actually achieve the resolution necessary to resolve single E. coli cells using our available hardware.
Therefore, we put most of our effort in developing our brightfield microscope presented above. Nevertheless, in the following, we will show you some of our setups, results and how our own lensless imaging system evolved over time. We do hope that other iGEM teams in the future jump on this track and overcome the difficulties that come with this approach.
The evolution of our lensless microscope
Theory
Unlike the brightfield imaging approach shown above, a lensless microscope is not in need of lenses at all. This on the one hand reduces the cost of the overall setup but more importantly also shrinks the size of the microscope significantly
as shown on top.
The first iteration
Our first prototype of a lensless imaging system simply consisted of a few LEGO bricks, a mobile phone as the lightsource and a raspberry pi camera chip with the lens being removed as explained here.
The Figure below shows this overall setup. Notice that for the very first experiments the pinhole is not really needed.
This classical shadow imaging setup, in which the incoherent LED lightsource is placed approximately 4-5 cm away from the sample and the sample as close as possible to the imaging sensor already leads to impressive results when imaging a sample of approximately 50 µm or bigger with strong contrast. The figure below for instance shows the result of imaging a hair in this setup. However, the limited coherence and the unrestricted large size of the lightsource (~0.5 cm radius) makes it impossible to resolve smaller structures with less contrast like for instance yeast cells. This is due to the fact that the shadows created from each “ray” of light simply overlap such that in the end everything blurs. This problem can be overcome by placing a pinhole in front of the light source as shown in the figure 2.3.
The raspberry pi camera sensor with a removed lens
Very first lensless measurement: A hair
The overall setup after adding a pinhole
The second iteration - adding a pinhole
From a theoretical perspective, the pinhole together with the incident bright light approximately leads to plane waves and a holographic image with a fringe magnification of 1. The holographic image is created through the interference of the incoming wave which passes the sample unscattered and the secondary waves which are emitted by the objects as illustrated in figure 3. By only adding a pinhole to the setup above we were able to successfully image yeast cells (see picture BELOW)
Illustration of the lensless setup by [Göröcs, Z., Ozcan, A., & Member, S. (2013). On-Chip Biomedical Imaging, 6, 29–46.]
Illustration how the interference pattern forms by Jericho, M. H., & Kreuzer, H. J. (2011). Point Source Digital In-Line Holographic Microscopy, 3–31. https://doi.org/10.1007/978-3-642-15813-1
Measuring 8um large yeast cells with our own setup seen in figure 2
The final iteration
The final Lensless Setup realised with our LCCSM
1um beads
2um beads
3um beads
5um beads
E. coli in a high density visible in the background
Imaging of a microfluidic chip
Different measurements accomplished with the Lensless microscope
Nevertheless, it was not possible for us to resolve single E. coli in this setup. The reason for this lies in the fact that, in line with the results of our simulation (HERE LINK TO SIMULATION PAGE), the bacteria simply do not have enough contrast for our limited pixel size. What we did see though is a dense population of bacteria.
Outlook
Possible future improvements to our final setup
Chulwoo et. al [https://doi.org/10.1364/OE.18.004717] propose using a briefringent crystal to introduce phase contrast to overcome this problem of limited contrast, however we did not have enough time and resources to implement this approach. Another idea to further improve the resolution and to achieve magnification even without lenses would be to switched to a monochromatic, coherent laser diode as lightsource (405 nm,150 mW from Thorlabs) and a very tiny pinhole (1 µm) to create a spherical wavefront. In such a setup, an additional magnification can be achieved through placing the sample way closer to the pinhole than to the sample as shown in the figure below. However, it is important to bear in mind that for this approach the monochromatic light source as well as the tiny pinhole are absolutely vital, as otherwise there would be too many interference patterns that overlap. In addition, an advanced image processing pipeline (like for instance holopy) is needed to infer the real objects from the hologram.
Illustration how an additional magnification can be achieved in a lensless setup (left) compared with the standard setup (right) by Sang Bok Kim, Hojae Bae, Kyo-in Koo, Mehmet R. Dokmeci, Aydogan Ozcan, A. K. (2012). Lens-free Imaging for Biological Applications. J Lab Autom, 17(1), 43–49. https://doi.org/10.1177/2211068211426695.Lens-Free
Conclusion
With the lensless imaging setup presented above, we were able to distinguish structures as small as 1 µm beads. However, due to the lack of contrast, it was not possible to resolve individual E. coli cells. We still think that a lensless microscope would be the ideal evolution of our brightfield microscope, although also further issues such as the lack of a focal plane in a lensless microscope would have to be addressed.
Luminescence Imaging
Introduction
Above, we presented two approaches on how to read out the biological signal corresponding to Approach A . We also invested time and effort to create a proper readout solution for Approach B . In particular, we added parts to our LCCSM system in order to satisfy the requirements in luminescence imaging and created an algorithm which allows us to read out a luminescence signal.
Design
The goal of the luminescence imaging solution is to detect a very low intensity of light. Inspired by TRALALA REF 1 which shows a luminescence detection in combination with smartphone cameras, we decided to build our own luminescence imaging solution, based on our LCCSM system. Since the light intensity of a luminescence signal is very low, it is crucial to ensure an efficient capture of photons and to optimally protect the imaging chamber from external spill light (vgl TRALALA REF1). In order to increase the number of photons actually reaching our chip, we decided to put reflective aluminum foil in the lid of our imaging chamber. Ideally, this should double the intensity at the chip as the light emitted to the top is then also reflected down onto our sensor. The resulting system is shown below if figure TRALALA:
Sample holder
Imaging Camera
Reflective Cover
The Luciferase Imaging setup
Implementation
Imaging Chip and Camera
In order to read out such a low intensity signal, it is crucial to use a very sensitive imaging chip. In first experiments we were using the standard Raspberry Pi Camera
module v2 which is based on the Sony IMX219 sensor. However with this chip it was not possible to resolve the luminescence signal. Therefore we decided to switch to the MT9J001
camera module which is based on the On-Semi MT9J001 monochrome raw sensor and provides a low-noise CMOS imaging technology that achieves near CCD quality in terms of sensitivity and
signal to noise ratio. We hera have a SNR of 34dB vs 36dB of the RaspberryPi Camera Sensor. Currently this chip is rather designed for developers use, that’s why we needed to implement the integration into
our software pipeline on our own. (see Software Overview)
Experiments
During multiple testing runs we verified our hardware as well as the corresponding software readout (see Software Overview). In all our experiments, the goal was to detect a luminescence signal generated by the enzymatic conversion of luciferin to oxyluciferin by the firefly luciferase. Since we expected the signal level to be rather low, all of the experiments included many negative controls. Furthermore, in the initial testing phase plate reader measurements were conducted in parallel to verify our own measurement data.
In the graph below, we present the result of a series of measurements taken with our own device. The x-axis, depicts the different samples, while the y-axis shows the intensity of the luminescence signal measured by our device. The figure clearly shows that we can resolve the luminescence signal, as only the samples that are supposed to emit photons (i.e. induced cells in buffer with luciferase, 1:1 induced cells together with uninduced cells, 1:3 induced cells together with uninduced cells and notably also uninduced cells in buffer with luciferase) lead to a signal of 0.2 or above. The fact that also the uninduced cells in the buffer with luciferase create a signal can be explained by the effect of leaky protein expression.
All of the negative controls led to a signal around zero. This variation in the signal is solely due to noise (e.g. the thermal noise of the sensor which varies over time).
On the whole, this figure shows that we can clearly resolve and detect the luminescence signal. We further verified our design through repeated executions of this kind of experiment which always led to similar results.
We were very satisfied with the results achieved by our low-cost device as we are able to reliably detect the luminescence signal generated by induced E. coli and can even detect the signal generated by uninduced cells through leakage.
Data generated with our own luminescence soft- and hardware. The sample was always a drop of 40uL and in all measurements with cells, a constant OD of 0.4 was ensured.
Results
We were very satisfied with the results achieved by our low-cost device as we are able to reliably detect the luminescence signal generated by induced E. coli and can even detect the signal generated by uninduced cells through leakage.
References
- Kim, H., Jung, Y., Doh, I. J., Lozano-Mahecha, R. A., Applegate, B., & Bae, E. (2017). Smartphone-based low light detection for bioluminescence application. Scientific Reports, 7(June 2016), 1–11. https://doi.org/10.1038/srep40203
- https://www.onsemi.com/pub/Collateral/MT9J003-D.PDF
Modular Casing System
Introduction
Overview of the LCCSM
The Parts
The main component
This is essentially a box with a hole in the top and the bottom. Where the hole sits we created an angled border, where the next box can be clicked in from the top. These boxes vary in height and enclose the microscope. This box connector is present on all parts, making it extremely flexible to connect all of them together.
The base
Here we mount the sensor. We created custom mounts for the Raspberry Pi camera and the arducam. It also acts as a stand or a mount to the robot.
The sample holder
This is where one introduces the sample into the microscope. We created a sliding drawer that can be used to push the sample into the imaging chamber. It is completely enclosed, so it does not let any spill light in.
The microfluidic chip holder
For the microfluidic chip we custom-built a case that allows to slide the chip horizontally enabling to image the different channels of the microfluidic chip. It also holds the microfluidic chip tightly in place and is fully enclosed.
The Lens or Pinhole or Filter mount
In order to mount pinholes, filters or lenses we created segments with a small hole in the middle. This allows the part to be mounted in the middle. Additionally we add small screws to each of the four sides to be able to tighten the part in the middle and position it to our desire.
The Thorlabs mount
As we decided to use a 10x magnification lens setup for our brightfield approach we realised it would be difficult to adjust the distances between the lenses, the chip and the sample precisely without having rods that the parts are sliding on. This is why we decided to integrate the Thorlabs rod system into our modular microscope setup. By making them fit tightly inside our 3D printed part we can adjust the height of each lens very precisely and individually. This part also allows to add other Thorlabs components in the microscopy setup such as for instance filters for fluorescence imaging.
The Light mount
At the very top of the microscope one often has a light source. We created a variety of different mounts that allow to place differently sized laser diodes (for the lensless system) or LEDs on top of our microscope.
Conclusion
The common, very simple clicking mechanism makes every part completely modular. This allows great flexibility and the design of complex setups by combining the simple parts. As the system already went through numerous iterations it is optimized to include all necessary functionality. It has proven to be extremely useful in the process of iterating through different setups and we would not have been able to implement all of these approaches without the LCCSM system in such a small amount of time. To enable everyone to make use of it we provide further instructions and downloads to the 3D files at the link below: