Edgise was approached by a manufacturer of innovative laboratory equipment that specializes in live-cell imaging. As the manufacturer supports applications across various areas of research, including stem cells, oncology, cell therapy, and more advanced applications such as chemotaxis and cell migration, they have to process a lot of samples. This means the laboratory technicians often have to count enormous amounts of cells, a task which takes up a lot of time. As a result, they were in need of a smart solution that could reduce their work time drastically.
Even though the client already worked with a cloud-based solution, they felt the need for a local smart counting solution. In other words, they wanted to make their microscopes more intelligent by moving the AI algorithms that were processing in the cloud to the actual microscopes. By doing this, the client wanted to significantly decrease the time needed for laboratory technicians to count the cells placed underneath microscopes.
As this is a fairly specific use case, there was no ready-to-use solution available on the market. Luckily, we are always on the lookout to provide tailored embedded hardware solutions with Edge AI technology, as this kind of technology is constantly evolving and more possibilities arise. We immediately went to work and started thinking about how Edge AI could provide a fitting solution for this case.
By designing a proof of concept, we quickly discovered that we could in fact deliver a viable solution by using Edge AI. The proof of concept consisted of a device that could process cell images in real time, without the need for cloud processing. Let’s dive into the technicalities, shall we?
We started out by connecting our clients’ microscopes to an Edge device that could process the microscopes’ images in real time. Brace yourself, because what follows, is some real Edge AI magic. The device had to be able to process the images, so we had to optimize the clients’ existing cell detection AI model to run on the device. Since the AI model was quite complex, we had to map everything perfectly to the resources available on NVIDIA Edge AI hardware. This meant that we had to optimize and implement both the AI model and the pre and post processing functions in a smart way. Moreover, using our core expertise, we designed the necessary embedded firmware.
As a result, the proof of concept achieved a substantial speedup in counting biological cells. This means a lot of valuable time is freed up for the laboratory technicians. As a result, they can analyze many more cell samples in the same time frame. Moreover, this concept reduced cloud data processing costs for the manufacturer, as this is a local solution that has no need to process in the cloud. In cases where low power, intelligence, performance, speed and privacy are paramount, we believe Edge AI technology can definitely offer what’s needed.