Home > Work > A sophisticated and low-cost diagnostic technology demonstrator

A sophisticated and low-cost diagnostic technology demonstrator


The project team needed to develop a demonstrator rig that could have multiple diagnostics applications, including immunodiagnostics and molecular diagnostics. The aim was to find a way to quickly develop a low-cost, yet sophisticated rig that could demonstrate a diagnostic technology.


Our scientists, mechanical engineers and software teams worked closely to build a rig able to image fluorescent particles flowing through microfluidic channels so that they could be tracked and counted using a machine learning algorithm. Our team built the demonstrator using commercially available components and open software libraries to save time and cost.


This project illustrated that a small rig can be quickly developed with the use of off-the-shelf components, available software libraries and tailored engineering and software development. The rig can demonstrate the feasibility of a diagnostic technology and reduce development time and ultimately, cost.

Do you need help developing your diagnostics device?

Let’s chat

Fluorescence for diagnostics applications

Fluorescent particles, or fluorophores, are used widely in many diagnostic processes, including immuno and molecular diagnostics. Nucleic acids and antibodies can be labeled with fluorescent particles to enable their detection. Fluorophores are highly sensitive to light and when illuminated, they transition to an excited state and emit light of different wavelengths, which allows them to be detected.

There are many fluorophores that are commonly used in diagnostics, each one requiring excitation at a specific wavelength and emitting light at a different wavelength. When exciting multiple fluorophores simultaneously in a multiplex assay, the light source should be capable of exciting each fluorophore sufficiently so that the particles can all be easily tracked and counted by an optical system.

Our diagnostics demonstrator rig

The demonstrator rig is made up of an optical system under which a glass cartridge containing the fluidic sample is placed. The fluorophores in the sample move within microfluidic channels in the cartridge. The movement is created by a peristaltic pump controlled by a laptop. A UV light is placed on the rig to illuminate fluorescent particles in the sample. A camera captures video clips, which allow the particles to later be tracked and counted by a machine learning algorithm.

Moving fluorescent particles in a fluidic sample

The rig needed to have precise fluidic control to ensure that the fluorescent particles moved smoothly through the microfluidic channels.

As the aim was to keep the demonstrator low-cost, we identified an off-the-shelf peristaltic pump to manage the fluid movement. The pump was coupled with a custom-designed damper for extra control at the extremely low flow rates required. The team reduced the pulsatile flow to near zero, preventing particle clumping in the channels, allowing for effective tracking and counting.

Developing an optical system using an off-the-shelf
Raspberry Pi camera

To keep the cost of the optical system down, we ensured that the selected fluorescent particles were excitable by the same wavelength, which meant that only one LED source was necessary, and the number of light filters was kept to a minimum.

We knew that the types of fluorophores we selected would dictate the type of optical system that needed to be used. We chose an off-the-shelf Raspberry Pi camera with built-in color filters to image and process the particles.

We designed the optical system to achieve the best resolution possible for our sample field of view. The image quality obtained with the final prototype allowed for accurate tracking and counting of the particles by the machine learning algorithm.

Machine learning to identify, track, and count fluorescent particles in real-time

With smooth fluidic control and camera recordings, we were able to identify, track and count the fluorescent particles in real-time using a machine learning object detection algorithm.

To keep the cost down, we used open software libraries to both find an object detection algorithm model and train it to recognize particles of different colors, size and focus from the camera recordings. We then converted this model to enable it to track and count particles moving through microfluidic channels on an off-the-shelf Raspberry Pi computer. Using only the fluid flow rate and particle count from the different fluorophores, we were able to determine particle concentration.

The software that our team created is a cost-efficient way to quantify analytes of interest in many applications. For example, this method could be used in processes monitoring, to develop multiplexed assays, or develop nucleotide sequencing technologies. It may also facilitate the transfer of laboratory-developed assays into medical devices by providing a means of assessing mixing or flow control.

Rapid and pragmatic diagnostics demonstrator development

Our scientists and software teams worked closely to find a pragmatic approach to build a demonstrator rig for diagnostic technologies. By using off-the-shelf components and adapting already existing machine learning algorithms, we were able to show that a demonstrator rig can be built quickly and at low-cost to prove a technology early on and significantly reduce development time.

Wonder how we could help?

Let’s start working together on your next challenge.