This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2017.2749040, IEEE
Transactions on Biomedical Engineering
IEEE Transactions on Biomedical Engineering
A high-resolution mini-microscope system for
wireless real-time monitoring
Zongjie Wang, Student Member, IEEE, Akash Boddeda, Benjamin Parker, Roya Samanipour, Sanjoy
Ghosh, Frederic Menard, and Keekyoung Kim, Member, IEEE
Abstract—Objective: Compact, cost-effective and high-
performance microscope that enables the real-time imaging of
cells and lab-on-a-chip devices is highly demanded for cell biology
and biomedical engineering. This paper aims to present the design
and application of an inexpensive wireless mini-microscope with
resolution up to 2592 x 1944 pixels and speed up to 90 fps. Methods:
The mini-microscope system was built on a commercial embedded
system (Raspberry Pi). We modified a camera module and
adopted an inverse dual lens system to obtain the clear field of view
and appropriate magnification for tens of micrometer objects.
Results: The system was capable of capturing time-lapse images
and transferring image data wirelessly. The entire system can be
operated wirelessly and cordlessly in a conventional cell culturing
incubator. The developed mini-microscope was used to monitor
the attachment and proliferation of NIH-3T3 and HEK 293 cells
inside an incubator for 50 hours. In addition, the mini-microscope
was used to monitor a droplet generation process in a microfluidic
device. The high-quality images captured by the mini-microscope
enabled us an automated analysis of experimental parameters.
Conclusion: The successful applications prove the great potential
of the developed mini-microscope for monitoring various
biological samples and microfluidic devices. Significance: This
paper presents the design of a high resolution mini-microscope
system that enables the wireless real-time imaging of cells inside
the incubator. This system has been verified to be a useful tool to
obtain high-quality images and videos for the automated
quantitative analysis of biological samples and lab-on-a-chip
devices in the long term.
Index Terms—Cell imaging, Microfluidics, Mini-microscope
I. INTRODUCTION
A microscope is an essential tool for many research areas
including cell biology, material science, and
nanotechnology. In addition to traditional bright-field
microscopes, different types of microscopy techniques have
been developed for specific applications. Phase contrast
microscopy [1], [2] is used for low-contrast object imaging,
polarization microscopy [3] for ordered molecules, and
differential interference contrast microscopy [4] for the
inspection of cultured cells. Recently, miniaturized
microscopes have been developed to check long-term cell
This work was supported by Natural Sciences and Engineering Research
Council of Canada (RGPIN-2014-04010).
Z. Wang, A. Boddeda, B. Parker, R. Samanipour and K. Kim are with School
of Engineering, University of British Columbia, Kelowna, BC V1V 1V7,
Canada (e-mail: keekyoung.kim@ubc.ca).
Copyright (c) 2016 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to pubs-permissions@ieee.org
0018-9294 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1
behavior inside an incubator and monitor lab-on-a-chip devices
[5]. Kim et al. demonstrated that the mini-microscope was able
to monitor cellular processes, including embryonic
development, wound healing, and disease progression over time
[6]. Previously, monitoring such processes required a
microscope equipped with an incubating chamber and anti-
vibration table that involve additional cost and maintenance.
Mini-microscope systems using an image sensor have been
developed and used inside a standard cell culture incubator [6]–
[8]. A camera module containing a complementary metal oxide
semiconductors (CMOS) image sensor was disassembled from
a webcam. The position of the camera lens placed over the
image sensor was adjusted to achieve suitable magnification for
biological samples. Table S1 in the supplementary file provides
a summary of previously reported webcam-based mini-
microscopes. In addition, a number of lens-free systems have
been developed with advantages in cost, weight and portability
since it does not require any lens [9]–[12]. The lens-free
systems have been used in a variety of applications such as
monitoring cell behavior [9], [12] and detecting biomarkers in
microfluidic devices [10], [11]. However, the mini-microscope
systems without lenses can only focus objects very close (e.g.,
~100 – 200 μm) to the image sensor which is less than the
thickness of petri dishes and glass slides (e.g., ~2 mm). Also,
most of the previously reported systems have the resolution of
640 x 480 (0.3 megapixels) [6]–[9], which is much lower than
the resolution of the camera designed for a bright-light
microscope (usually higher than 3 megapixels). The limited
resolution usually results in the poor quality of images and
makes post image processing difficult. In addition, cell phones
have been employed as a portable and rapid cell monitoring
system [13]. The image sensors on cell phones usually have 5
to 8 megapixel resolution, which is much better than the
webcam-based mini-microscopes. However, the cell phone-
based systems are relatively expensive (~ US$500) than other
mini-microscopes, not robust, and has limitations to control
since the operating program are not open to the public.
Moreover, current cell phone-based systems require keeping a
substrate inside a small, closed chamber, which is not suitable
for cell culturing and monitoring.
K. Kim is with Biomedical Engineering Program, University of British
Columbia, Vancouver, BC V6T 1Z4, Canada.
S. Ghosh and F. Menard are with Irving K. Barber School of Arts and
Sciences, University of British Columbia, Kelowna, BC V1V 1V7, Canada.
A. Boddeda is with Department of Mechanical Engineering, Indian Institute
of Technology Guwahati, Guwahati, Assam, 781039, India.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2017.2749040, IEEE
Transactions on Biomedical Engineering
IEEE Transactions on Biomedical Engineering
Fig. 1. Inverse dual lens system. (A) Schematic of the inverse dual lens system.
(B) Working principle of the dual lens system to achieve different
magnification rate. (C) An image of microfluidic channel obtained by
commercial ball lens (D) An image of microfluidic channel obtained by the
inverse dual lens system. Scale bar = 200 μm.
In this study, we present a low-cost, high-resolution, highly
controllable and multi-functional platform for monitoring cell
behavior inside an incubator and experimental situation on a
lab-on-a-chip device. We used an inexpensive embedded circuit
board (Raspberry Pi) and a 5 megapixel CMOS camera module
to build the mini-microscope. This mini-microscope with the
cost around US$100 can be easily placed in a conventional cell
incubator and controlled wirelessly (See Table. S1 for the
specification). We demonstrated that the mini-microscope was
able to observe the long-term behavior of biological cells. Real-
time image processing was conducted to analyze the coverage
rate of cells during proliferation. Also, to demonstrate the high
speed image capturing capability, the mini-microscope was
used to examine a droplet generation process in a continuous
microfluidic system. The various experiments demonstrated
that the mini-microscope could facilitate many studies in long-
term cell monitoring inside the incubator and examining
experimental processes in lab-on-a-chip devices.
II. MATERIALS AND METHODS
A. System Design and Integration
Raspberry Pi Model B+ (Raspberry Pi Foundation, UK) was
adopted as a platform to construct the mini-microscope. The
system was connected to a host computer wirelessly through a
USB Wi-Fi adapter (Cana Kit Corporation, North Vancouver,
BC, Canada). Debian-based Raspbian operating system was
installed in the system to control all hardware. The CPU and
GPU of Raspberry Pi were controlled by Python-based scripts
to realize real-time image capturing and video recording. One
USB port was used to supply 5V via transistor–transistor logic
(TTL) interface to turn on a white LED array for illumination.
An open source software, VLC player (VideoLAN, Paris,
France), was installed in the system for the real-time video
streaming and broadcasting. The video streaming and
Copyright (c) 2016 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to pubs-permissions@ieee.org
0018-9294 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
2
broadcasting allowed other devices in the same network, such
as the host computer and portable devices, to receive the real-
time images or videos that enable users to do the remote
monitoring and data analysis.
A Raspberry Pi camera module (Raspberry Pi Foundation,
UK) was connected to the embedded system. The camera is
capable of recording video clips at 7, 30, 60 and 90 frames-per-
second (FPS) with the resolutions of 2592 x 1944, 1920 x 1080,
1280 x 720 and 640 x 480 pixels, respectively. The detailed
specification can be found in the supplementary material (Table
S2). The camera module was originally designed for normal
distance recording, which is not suitable to provide
magnifications for microscale objects. Therefore, we flipped
the original lens and added another lens to build a two-lens
system for capturing clear and magnified images (Fig. 1A and
1B). Since the camera module does not have the autofocus
function, a 3D printed z-axis stage is needed to find the best
focal plane for different types of cell substrates (e.g. petri dishes,
flasks, or glass slides) with various thickness.
The camera module, z-axis stage, and embedded system were
integrated and assembled into an acrylic box to build the mini-
microscope system. The entire system was operated by a USB
power adaptor with the output voltage of 5V and current of 1A.
This system can also be operated by a mobile supplementary
battery, making the system fully remote and portable. The
schematic diagram of the system configuration is illustrated in
the supplementary material (Fig. S1). The Raspberry Pi system
is controlled by a Python program for recording images/videos,
storing data into a flashdisk, and broadcasting a video stream
into a wireless network. The host computer receives the video
stream from the wireless network and further processes the data
for analysis via a MATLAB-based program (Mathworks,
Natick, MA, USA). Other portable devices, such as a mobile
phone and a tablet computer, can also receive the captured
video stream for remote monitoring.
B. Lens System Design
Although there is a commercial polymer-based ball lens with
certain magnification, only the central part of the image is clear,
while the rest of the area is blurred due to the effect of field
curvature and spherical aberration [14]. Switz et al. developed
a novel inverse dual lens configuration to have both the wide
field of view and appropriate image quality of microscale
sample [14]. We adopted this configuration to design the lens
system of the mini-microscope. The schematic and working
principle of the lens system is given in Fig. 1. Briefly, a lens
from the camera module was flipped to make an inverse lens.
Then, another lens in an ordinary direction was placed over the
inverse lens to build the dual lens system. The CMOS sensor
captured object images through the dual lens system. As
illustrated in Fig. 1B, the object was demagnified through the
inverse lens and then magnified by the ordinary lens.
Theoretically, another advantage of the dual lens method is that
the magnification can be adjustable by changing the distance
(d1) between the CMOS sensor and inverse lens [14]. However,
we fixed the d1 to be 1.4 cm for two reasons: First, the typical
field of view (around 1 mm x 1 mm) is suitable for many
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2017.2749040, IEEE
Transactions on Biomedical Engineering
IEEE Transactions on Biomedical Engineering 3

Fig. 2. Mini-microscope platform. (A) A schematic of the mini-microscope system. (B) and (C) Photographs of the assembled mini-microscope system. Scale bar
= 4 cm. (D) An image of a microscope calibration slide obtained by the mini microscope. Scale bar = 200 μm.

biomedical engineering applications. Second, the adjustment of
d1 can easily introduce dusts on the surface of the CMOS
camera, resulting in the ‘dead zone’ of the CMOS sensor. Fig.
1C and 1D show images of generating droplets in a microfluidic
device using a single lens and dual lens system, respectively.
The image from the dual lens system was much clearer with the
same field of view in a single lens system, offering the better
quality of image without blurred areas resulted from the lens
system.
C. Experimental Setup
First, NIH-3T3 fibroblast and HEK 293 human kidney cells
were cultured in Dulbecco's minimum essential medium
(DMEM) supplemented with 1% v/v penicillin streptomycin
and 1% v/v amphotericin B. All items were purchased from Life
Technologies, Carlsbad, CA, USA. These cells were cultured
on a T25 cell culture flask (VWR scientific, Radnor, PA, USA)
and passaged every three days when the cells were confluent.
Cells were trypsinized for the experiments. The half of the cells
were seeded onto a new flask that was placed on the mini-
microscope for monitoring long-term cell behavior. Time-lapse
images were captured after two hours from seeding the cells to
allow them attaching on the surface of the flask. Before
recording images, we manually adjusted the focus of the mini-
microscope to find the best focal plane of the cells. Then, high-
quality images with the 2952 x 1944 pixel resolution were taken
every minute and stored in the shared folder on the embedded
system. A MATLAB-based program was operated to read the
captured images and complete the real-time image processing
to measure the coverage rate of cells. The image processing
algorithm is based on the threshold Canny edge detection
method [15] and morphological image closure. Detail
information for the image processing is discussed in the
supplementary data.
A flow-focusing microfluidic device was fabricated by
standard photolithography and softlithography processes as
described in [16], [17] (See supplementary data for the detailed
protocol). The fabricated device has a cross junction where a
water phase fluid and mineral oil with 20% v/v surfactant (Span
80, Sigma Aldrich, St. Louis, MO, USA) infused by two syringe
pumps (Kent Scientific Corp., Torrington, CT, USA) were met
together to generate droplets. The water phase fluids were made
of 5% and 8% w/v gelatin methacrylate (GelMA) prepolymer
dissolved in a phosphate buffered saline (PBS). The GelMA
was synthesized by a protocol previously described in [18], [19]
(See supplementary data for the detailed protocol). The
viscosities of the GelMA prepolymer solution and mineral oil
with the surfactant were measured by a viscometer (Cannon
Instrument, State College, PA, USA).
During the droplet generation process, the device was placed
on the mini-microscope and the cross junction of the device was
precisely focused to obtain clear images of the droplet
generation process. Subsequently, the droplet generation
process was recorded by the mini-microscope with the
resolution of 640 x 480 pixels and frame rate of 90 fps. The
recorded video clips were post-processed by MATLAB to
analyze the diameter of droplets, the generation frequency, as
well as the length of thread at the junction. The image
processing is based on Canny edge detection method and the
line traverse method developed by Wang et al.[20]. The details
of the imaging processing method are discussed in the
supplementary data. To compare the effect of the frame rate on
analysis, an inverted microscope (ToupTek Photonics,
Hangzhou, Zhejiang, China) was also used to record the droplet
generation at its maximal frame rate of 24 fps.
D. Statistical Analysis
A one-way analysis of variance (ANOVA) was proceeded to

Copyright (c) 2016 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to pubs-permissions@ieee.org
0018-9294 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2017.2749040, IEEE
Transactions on Biomedical Engineering
IEEE Transactions on Biomedical Engineering 4

Fig. 3. Experiment for monitoring cell attachment and proliferation inside the incubator. (A) Representative images show NIH-3T3 and HEK 293 cells at 2, 14,
26, and 50 hours after seeding. Scale bar = 100 μm. (B) Microscopy images show a cell division process. Scale bar = 100 μm.

statistically analyze the measured data. Results were shown as
an average ± standard deviation. The curve fitting was obtained
through the built-in curve fitting function in Microsoft Excel
(Microsoft, Redmond, WA, USA). Among three curve fittings
(exponential, linear, and logarithmic), the one with the highest
R2 was selected.
III. RESULTS AND DISCUSSION
A. System Characterization
The schematic of the mini-microscope was shown in Fig. 2A.
The lens system and camera module were positioned on the top
of a manual z-axis stage to adjust distance between the object
and the camera (adjustable d2 in Fig. 1A). With this stage, the
mini-microscope can precisely focus cells on various cell
substrates (i.e., petri dishes, flasks, and glass slides) with
different thickness. The CMOS camera was connected to the
Raspberry Pi system for the image acquisition control and
wireless data transfer. The entire mini-microscope was installed
in an acrylic box with 16 cm long and 6 cm tall and wide which
is suitable for keeping it in an incubator for biomedical
applications. The photography of the developed mini-
microscope is given in Fig. 2B and 2C. We briefly examined
the system for the feasibility of cordless power operation. With
a 3000 mAh power bank, the mini-microscope can be operated
to wirelessly record and transport full HD video (1920x1080,
30 fps) for about 5 hours. It may be possible to run the mini-
microscope system cordlessly for a day with a commercial
20000 mAh power bank.
During the operation, the object was placed on the window
of the acrylic box, and a simple array of LED lights was
positioned above the object for providing illumination. We
firstly examined the field of view and clearness of the image
through the mini-microscope using the microscope calibration
slide (the smallest gap = 10 μm) as given in Fig. 2D. The field
of view was approximately 1 mm by 1 mm when d1 was set to
1.4 cm. 10 μm gap can be clearly distinguished in most of the
area. The blurred area near the edge of the field of view may be
attributed from that the CMOS sensor was not perfectly
positioned horizontally on the stage. It is noticed that there were
several artifacts appeared in the field of view resulted from the
dusts introduced when assembling the lens. It is also worth
noting that the light from the LED array was not uniform
throughout the entire field of view, causing shades near the edge
of images (i.e., Fig. 1D and 2D). Introducing the so-called
‘illumination shaping filter’ [14] may further improve the light
uniformity to obtain a better quality of images near the corner.
We also took the images of microscope calibration slide from
other microscopes as shown in the supplementary Fig. S2. The
mini-microscope and the high resolution microscope show
sharp images of 10 μm gaps in the middle portion of the field
of view, while the image from the medium resolution
microscope was not clear. Only the high resolution microscope
could maintain the clear image around both ends of the field
view. In summary, although it is not comparable to the high
resolution microscope, the mini-microscope offers an
acceptable image quality at a low cost. The image quality of the
developed mini-microscope can be further improved by
introducing a better lens system (i.e., ISF) or using the latest
camera module with a higher resolution.
B. Cell Behavior Study
Understanding the attachment and proliferation of
mammalian cells is critical for the researches in stem cells,
tissue engineering, as well as the clinical trials [21]. One of the
most widely used methods to monitor cell attachment and
proliferation is the time-lapse microscopy of live cells [22], [23].
However, it usually requires to use the expensive incubation
accessories of the microscope. It is desired to develop an
alternative method offering the inexpensive but high quality of
time-lapse cell images. Here, we developed the mini-
microscope to place into a commercial incubator, and took the

Copyright (c) 2016 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to pubs-permissions@ieee.org
0018-9294 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2017.2749040, IEEE
Transactions on Biomedical Engineering
IEEE Transactions on Biomedical Engineering
Fig. 4. Image processing data to show cell coverage rates. (A) A plot shows
cell coverage rate over time. Curve fitting results show that the coverage rate
of NIH-3T3 cells (B) is faster than HEK293 cells (C) during culturing.
high quality time-lapse images of NIH-3T3 fibroblasts and
HEK 293 human embryonic kidney cells for monitoring long-
term cell behavior during attachment of proliferation.
Using the developed mini-microscope for time-lapse
imaging has several advantages. First, the mini-microscope is
relatively small (16 cm x 6 cm x 6 cm), making it suitable to be
placed directly into the incubator. No modification of the
incubation system is needed. Second, the mini-microscope does
not require a host computer near the incubator as the mini-
microscope can be controlled wirelessly. The mini-microscope
also provides the live stream video for remote monitoring to
maintain the reliability of the experiment over days. The
capability of wirelessly monitor and control can greatly
improve the observability, reliability, and efficiency of the
experiments.
In addition to the ease of high observability, our mini-
microscope provides cell images with better quality than the
previously reported mini-microscopes. The recent
improvement of the CMOS sensor (maximum resolution: 2592
x 1944) enabled us to obtain a better quality of images
compared to the previously reported mini-microscopes based
on the low resolution sensor (maximum resolution: 640 x 480)
[6]–[8]. The morphology of single NIH-3T3 cell can be clearly
distinguished (See Fig. S3 for the comparison of the images of
NIH-3T3 cells from the mini-microscope in this study and
previous works by Kim et al. [6]). The representative images
captured at 2, 14, 26 and 50 hours after seeding cells on a cell
culture flask were presented in Fig. 3A. The shapes and
boundaries of single cells can be clearly seen from the captured
images. After 2 hours, most of the cells were remained in a
round shape. However, cells (especially NIH-3T3 cells)
attached and elongated after 14 hours. The confluency of cells
was significantly increased when comparing the coverage rate
of cells after 14 hours and after 50 hours. We attached a video
clip (supplementary video 1) to show the division process of
NIH-3T3 during proliferation in-between 30 and 50 hours. Fig.
3B shows four snapshots of supplementary video 1. It is noticed
that in the highlighted area, the attached cells became round for
Copyright (c) 2016 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to pubs-permissions@ieee.org
0018-9294 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
5
division, which matches the reported results from [24].
We conducted the coverage rate analysis using the image
processing method. The processed images are also shown in Fig.
3A. It is noticeable that the processed images could not identify
the border between the single cells as the cells were connected
to each other when the confluency reached high. Therefore, the
edge detection-based image processing algorithm presented
here is not feasible to find the morphology of single cells and
count the quantity of cells. However, the bright area of the
processed images can still be used to approximately reflect the
areas covered by cells in the field of view. The relationship
between the coverage rate and culturing time was plotted in Fig.
4A. Supplementary video 2 is a real-time video clip showing
the image analysis process of HEK 293 cells. We found the
significant difference between NIH-3T3 and HEK 293 cells in
their behavior of attachment and proliferation. For NIH-3T3
cells, most of the cells attached and elongated rapidly and the
coverage rate was dramatically increased after 7 hours.
However, HEK 293 cells formed a small group of cells before
attaching on the flask (see supplementary video 2), which
resulted in the low coverage rate within 17 hours. After small
groups of cells attached, HEK 293 cells started to proliferate.
There was a significant difference between the attachment time
for NIH-3T3 fibroblasts (approximately 7 – 8 hours maximally)
and HEK 293 cells (at least higher than 16 hours), as shown in
Fig. 4A. The coverage rate of cells was obtained by the curve
fitting (Fig. 4B and 4C). The slope of the coverage rate plot of
NIH-3T3 cells was significantly greater than that of HEK 293
cells, indicating the different cell behaviors. The successful
observation of the long-term cell culturing process inside the
incubator demonstrates the feasibility of the developed mini-
microscope for the time-lapse imaging and analysis. In short,
the mini-microscope has provided an inexpensive but high
resolution method to monitor and analyze cell development in
vitro. Such a capability would be very useful for the study of
the long-term cell-cell and cell-extracellular interactions and
especially for the factors that regulate stem cell fate [25].
C. Microfluidic Experiment Investigation
Droplets are a useful tool for tissue engineering [17], [26],
drug delivery [27], and organ-on-a-chip systems [28]. The most
widely used devices for generating droplets are the flow-
focusing microfluidic channel devices. In a flow-focusing
device, two immiscible fluid phases (dispersed and continuous
phase) are injected into the cross-junction channel through
syringe pumps with different flow rates (Fig. 5A). Commonly,
the continuous phase is the mixture of oil and surfactant, which
is very viscous and has high surface tension. The dispersed
phase is usually an aqueous solution, such as a hydrogel
prepolymer solution. Droplets are generated when two phases
meet at the cross junction, where the shear force induced by the
continuous phase breaks the dispersed phase to form droplets.
There are many studies on the physics of droplet generation
[29]–[31]. However, it is still hard to use the physics model to
predict the droplet generation process of biomaterials since
those models are based on water as the dispersed phase [32]. In
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2017.2749040, IEEE
Transactions on Biomedical Engineering
IEEE Transactions on Biomedical Engineering 6

Fig. 5. Experiment for monitoring a droplet generation process in a microfluidic device. (A) Flow-focusing droplet generation at the cross junction of the
microfluidic device. (B) Experimental setup with two syringe pumps. Scale bar = 2 cm. (C) Images show periodic changes of the thread during droplet formation.
Scale bar = 200 μm.

addition, previous works using hydrogels as the dispersed phase
briefly reported the relationship between the diameter of
droplets and flow rates [17], [33]. Here, we utilized the mini-
microscope to investigate fundamental parameters for the
droplet generation process of hydrogel prepolymer. The
experimental setup is presented in Fig. 5B. The parameters
include the diameter of droplets, droplet generation frequency,
and the length of thread. The term ‘thread’ was firstly
introduced by Tan et al. to explain the flow-focusing droplet
generation mechanism [34]. It indicates the portion of aqueous
phase which is passed the junction before forming droplets. The
end point of the thread is the point that the highest shear stress
occurred [34]. As can be seen in Fig. 5C, the length of thread
was changed periodically during droplet generation.
The investigation of microfluidic experiment aims to show
the advantages of the developed mini-microscope in terms of
the high frame rate per second. In general, the requirements of
the mini-microscope system for monitoring the droplet
generation process is different from the cell behavior
monitoring. First, the contrast difference between the droplets
and microfluidic channel is significant and the size of droplets
is three times greater than cells. Therefore, the morphology of
the droplets can be easily detected even with the low resolution
of captured video clips. However, since the droplet generation
frequency is very high, the camera with low frame rate per
second may not be able to capture enough number of images to
show the detail process. For example, with the conventional
microscope with the maximal fps of 24 (Fig. 6A), the end of the
thread near the junction area was not clear due to the relatively
low fps. On the contrary, when we used the maximal fps of the
mini-microscope (fps = 90) to record the droplet generation
process, the end of the thread was much clearer (Fig. 6A). In
supplementary video 3, the difference between 24 fps and 90
fps is highly significant. In addition, the developed mini-
microscope with 90 fps offers more frames in the generation
process of each droplet (~ 6 frames for a droplet) compared to
the previously reported mini-microscopes with the maximum
fps of 30 [6], [7]. The mini-microscope with the high fps can
capture the droplet generation process more smoothly and
clearly, providing high quality data for image processing.
Therefore, we employed the highest fps for the microfluidic
droplet generation experiments, although the high fps resulted
in relatively low resolution images.
The processed image is given in Fig. 6A. The white point is
the detected end of threads. The detection of the droplet
generation frequency and the diameter of droplets are presented
in supplementary video 4 and 5, respectively. Thanks to the
capability of the mini-microscope to record videos at high fps,
the generation of each droplet and the corresponding change of
‘thread’ can be accurately tracked and analyzed. As illustrated
in the video 5, the length of threads exponentially increased
over the time. After it reached a peak, a new droplet was formed
and the length went back to a minimum value close to zero.
Since the end point of the thread is the point that the highest
shear stress occurred [34], the dynamic tracking of the thread
length enabled by the high fps of the mini-microscope can help
researchers better understand the droplet generation process of
various materials and study the unusual phenomenon of droplet
generation induced by encapsulating foreign particles (e.g.,
cell-induced instability [35]).
In addition to the dynamic tracking of ‘thread’, the mini-
microscope was also capable of providing quantitative data (i.e.,
average diameter of droplets) for statistical studies. After
testing various conditions listed in the method section, we
summarized the data in Fig. 6B – 6E. The kinematic viscosity
from different fluids was measured by the viscometer. As

Copyright (c) 2016 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to pubs-permissions@ieee.org
0018-9294 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2017.2749040, IEEE
Transactions on Biomedical Engineering
IEEE Transactions on Biomedical Engineering
Fig. 6. Analysis of droplet generation process. (A) Representative images was
captured by the commercial camera (24 fps) and mini microscope (90 fps). The
image processing was based on the image captured by the mini microscope and
a white point indicate the detected end of the thread. Scale bar = 200 μm. (B)
The kinematic viscosity of PBS, 5% GelMA, 8% GelMA, and oil with 20%
surfactant was measured by the viscometer (n = 3, *p < 0.05). (C) Diameters,
(D) Length of thread, and (E) generation frequency of the droplets from PBS,
5% GelMA, and 8% GelMA with two different flow rates of continuous phase
were calculated by the image processing (*p < 0.05, n = 5). (F) Curve fitting
results show the logarithmic relationship between the kinematic viscosity ratios
of continuous phase and dispersed phase and the diameter of droplets (n = 5).
shown in Fig. 6B, with the increment of GelMA concentration,
the dispersed phase become more viscous. The continuous
phase (oil with 20% surfactant), however, is much more viscous
than the dispersed phase. At the same flow rate, viscous
dispersed phase resulted in bigger droplets, the longer length of
thread, and lower frequency of droplet generation. On the
contrary, the increment of continuous phase flow rate
contributed to smaller droplets, the shorter length of thread, and
higher frequency of droplet generation. These trends are well
matched with previous experimental results [16]. The analyzed
data can be further used to find parameters for modeling droplet
generation mechanisms. For example, we found that the
diameter of droplets seems to be a logarithmic function of the
ratios of kinematic viscosities of two flow phases (Fig. 6F). The
R2 values of the fitted curves are around 0.95. The successful
capturing and analysis of the droplet generation process proves
the feasibility of using the developed mini-microscope for
monitoring lab-on-a-chip devices. In the future work, the mini-
microscope can be used not only as a passive observation
Copyright (c) 2016 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to pubs-permissions@ieee.org
0018-9294 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
7
device, but also a main part of active control systems for
microfluidic devices inside the incubator. For example, the
mini-microscope can be integrated with syringe pumps that
regulate flow rates of the droplet generation process. A real-
time feedback control and analysis system based on the videos
obtained from the mini-microscope can actively control the
droplet generation process to produce large amounts of uniform
droplets inside the incubator for the long term. Such a highly
automated system may greatly benefit the labor-intensive
biofabrication process that usually requires large amounts of
uniform cell-laden droplets [36].
IV. CONCLUSION
In this paper, we present an inexpensive, high resolution
mini-microscope system suitable for in situ monitoring of
biological samples and microfluidic experiments. The system
was built on the Raspberry Pi embedded system. To make the
normal camera module clearly visualize microscale images, we
employed the inverse dual lens system that provides the clear
field of view and suitable magnification rate. The mini-
microscope was able to obtain and transfer the image data
wirelessly, offering the capability of operating it in a
conventional cell culturing incubator. To demonstrate the
feasibility of the system, the developed mini-microscope was
used to monitor cell attachment and proliferation processes, as
well as droplet generation processes inside microfluidic devices.
Benefited from high-quality images, the post image processing
was able to automatically investigate many useful information,
such as the coverage rate of cells and the generation frequency,
diameter, and thread length of droplets. Successful applications
demonstrate that the mini-microscope system has a great
potential in many biological and biomedical applications.
REFERENCES
[1] C. R. Burch and J. P. P. Stock, “Phase-contrast microscopy,” J. Sci.
Instrum., vol. 19, no. 5, pp. 71–75, 1942.
[2] C. J. Mann et al., “High-resolution quantitative phase-contrast
microscopy by digital holography,” Opt. Express, vol. 13, no. 22, pp.
8693–8698, 2005.
[3] R. Oldenbourg, “New views on polarization microscopy,” Eur. Cell.
Mater., vol. 1, no. 2, p. 13, 2001.
[4] C. J. Cogswell and C. J. R. Sheppard, “Confocal differential interference
contrast (DIC) microscopy: including a theoretical analysis of
conventional and confocal DIC imaging,” J. Microsc., vol. 165, no. 1, pp.
81–101, 1992.
[5] E. Schwan et al., “A modular and affordable time-lapse imaging and
incubation system based on 3D- printed parts, a smartphone, and off-the-
shelf electronics,” PLoS One, vol. 11, no. 12, pp. 1–15, Dec. 2016.
[6] S. B. Kim et al., “A mini-microscope for in situ monitoring of cells.,” Lab
Chip, vol. 12, no. 20, pp. 3976–82, Oct. 2012.
[7] K. Koo et al., “An optical multi-sensing system for detection of
cardiovascular toxicity.,” Biotechnol. Lett., vol. 36, no. 5, pp. 1089–1094,
2014.
[8] Y. S. Zhang et al., “A cost-effective fluorescence mini-microscope for
biomedical applications,” Lab Chip, vol. 15, no. 18, pp. 3661–3669, 2015.
[9] S. B. Kim et al., “A cell-based biosensor for real-time detection of
cardiotoxicity using lensfree imaging.,” Lab Chip, vol. 11, no. 10, pp.
1801–7, May 2011.
[10] S. Seo et al., “High-throughput lens-free blood analysis on a chip,” Anal
Chem, vol. 82, no. 11, pp. 4621–4627, 2010.
[11] J. Balsam et al., “Lensless CCD-based fluorometer using a
micromachined optical Söller collimator.,” Lab Chip, vol. 11, no. 5, pp.
941–9, Mar. 2011.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2017.2749040, IEEE
Transactions on Biomedical Engineering
IEEE Transactions on Biomedical Engineering
[12] X. Zhang et al., “Lensless imaging for simultaneous microfluidic sperm
monitoring and sorting.,” Lab Chip, vol. 11, no. 15, pp. 2535–40, Aug.
2011.
[13] D. Tseng et al., “Lensfree microscopy on a cellphone.,” Lab Chip, vol. 10,
no. 14, pp. 1787–92, Jul. 2010.
[14] N. A. Switz et al., “Low-cost mobile phone microscopy with a reversed
mobile phone camera lens,” PLoS One, vol. 9, no. 5, p. e95330, May 2014.
[15] J. Canny, “A computational approach to edge detection,” IEEE Trans.
Pattern Anal. Mach. Intell., vol. PAMI-8, no. 6, pp. 679–698, 1986.
[16] R. Samanipour et al., “Computational and experimental study of
microfluidic flow-focusing generation of hydrogel droplets,” J. Appl.
Polym. Sci., 2016.
[17] C. Cha et al., “Microfluidics-assisted fabrication of gelatin-silica core −
shell microgels for injectable tissue constructs,” Biomacromolecules, vol.
15, pp. 283–290, 2014.
[18] Z. Wang et al., “A simple and high-resolution stereolithography-based 3D
bioprinting system using visible light crosslinkable bioinks,”
Biofabrication, vol. 7, no. 4, p. 45009, 2015.
[19] Z. Wang et al., “An ultrafast hydrogel photocrosslinking method for direct
laser bioprinting,” RSC Adv., vol. 6, no. 25, pp. 21099–21104, 2016.
[20] Z. Wang et al., “Development of anatomically realistic numerical breast
phantoms based on T1- and T2-weighted MRIs for microwave breast
cancer detection,” IEEE Antenna Wirel. Propag. Lett., vol. 13, pp. 1757–
1760, 2014.
[21] R. Dai et al., “Adipose-derived stem cells for tissue engineering and
regenerative medicine applications,” Stem Cells Int., 2016.
[22] P. M. Kulesa and S. E. Fraser, “Neural crest cell dynamics revealed by
time-lapse video microscopy of whole chick explant cultures,” Dev. Biol.,
vol. 204, pp. 327–344, 1998.
[23] A. L. Siegel et al., “Muscle satellite cell proliferation and association: new
insights from myofiber time-lapse imaging,” Skelet. Muscle, vol. 1, no. 1,
pp. 327–344, 2011.
[24] A. Lesman et al., “Contractile forces regulate cell division in three-
dimensional environments,” J. Cell Biol., vol. 205, no. 2, pp. 155–162,
2014.
Copyright (c) 2016 IEEE. Personal use of this material is permitted. However, permission to use this material for any other
purposes must be obtained from the IEEE by sending an email to pubs-permissions@ieee.org
0018-9294 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
8
[25] N. Bertani, “Neurogenic potential of human mesenchymal stem cells
revisited: analysis by immunostaining, time-lapse video and microarray,”
J. Cell Sci., vol. 118, no. 17, pp. 3925–3936, 2005.
[26] Y. T. Matsunaga et al., “Molding cell beads for rapid construction of
macroscopic 3D tissue architecture,” Adv. Mater., vol. 23, no. 12, pp. 90–
94, 2011.
[27] H. Shibata et al., “Injectable hydrogel microbeads for fluorescence-based
in vivo continuous glucose monitoring.,” Proc. Natl. Acad. Sci. U. S. A.,
vol. 107, no. 42, pp. 17894–17898, 2010.
[28] Z. Wang et al., “Organ-on-a-chip platforms for drug delivery and cell
characterization : a review,” Sensors Mater., vol. 27, no. 6, pp. 487–506,
2015.
[29] W. Lee et al., “Role of geometry and fluid properties in droplet and thread
formation processes in planar flow focusing,” Phys. Fluids, vol. 21, no. 3,
pp. 1–15, 2009.
[30] C. N. Baroud et al., “Dynamics of microfluidic droplets.,” Lab Chip, vol.
10, no. 16, pp. 2032–2045, 2010.
[31] P. He et al., “Flow of two immiscible liquids with low viscosity in Y
shaped microfluidic systems: effect of geometry,” Microfluid.
Nanofluidics, vol. 9, no. 2–3, pp. 293–301, 2009.
[32] Z. Wang et al., “An automated system for high-throughput generation and
optimization of microdroplets,” Biomicrofluidics, vol. 10, no. 5, p.
013605BMF, 2016.
[33] T.-D. Dang et al., “Preparation of monodisperse PEG hydrogel
microparticles using a microfluidic flow-focusing device,” J. Ind. Eng.
Chem., vol. 18, no. 4, pp. 1308–1313, Jul. 2012.
[34] Y.-C. Tan et al., “Monodispersed microfluidic droplet generation by shear
focusing microfluidic device,” Sensors Actuators B Chem., vol. 114, no.
1, pp. 350–356, 2006.
[35] J. Jung and J. Oh, “Cell-induced flow-focusing instability in gelatin
methacrylate microdroplet generation,” Biomicrofluidics, vol. 8, no. 3,
2014.
[36] Y. Morimoto et al., “Point-, line-, and plane-shaped cellular constructs for
3D tissue assembly,” Adv. Drug Deliv. Rev., vol. 95, pp. 29–39, 2015.