High-Speed Automated Blood Testing Solutions with Embedded AI

Source: Xiaojudeng Network

PerkinElmer was founded in 1937 and is headquartered in Waltham, Massachusetts, USA. Five years ago, the company acquired Chemagen, whose core is a multi-wavelength color vision system capable of capturing images of centrifuged blood samples.

Cognex engineers collaborated with PerkinElmer’s instrument experts to demonstrate how deep learning technology, which falls under artificial intelligence (AI), combined with compact embedded industrial cameras, can overcome one of the challenges at the end of the automated blood separation testing process (a key component of RNA and DNA testing and disease diagnosis).

A major challenge faced by the PerkinElmer team was how to quickly capture images of centrifuged blood sample tubes and reliably convert these images into measurable height information. However, traditional deterministic machine vision software methods could not reliably solve this issue in a repeatable manner. Therefore, the PerkinElmer team sought a solution from Cognex to tackle the challenge of detecting the white membrane layer in blood separation samples at the automated blood separation workstation.

When the PerkinElmer automated blood separation workstation operates, the samples, caps, labels, and fluid components may vary in size, shape, color, and position. For this type of complex application, combining traditional machine vision with deep learning technology is more effective in measuring the internal features of tubes and samples.

Cognex’s VisionPro Deep Learning software platform combines commercial-grade deterministic machine vision algorithms with deep learning software tools that can run on embedded or traditional PCs (depending on the type of embedded application). The deep learning software analyzes images marked as “qualified” or “unqualified” by quality experts. By analyzing dozens, hundreds, or even thousands of sample images, the deep learning software can “learn” what is qualified and what is unqualified, much like a human child, rather than based on rules set by programmers.

After operators load a row of sample tubes into the PerkinElmer automated blood separation workstation, the entire analysis process begins as the single row of tubes moves between two Cognex Advantage vision cameras. Each Advantage vision camera contains an onboard AE3 vision engine module, allowing this ultra-compact vision system to run Cognex’s In-Sight embedded image processing algorithms while connecting to a PC embedded with deep learning software for advanced image processing analysis.

One Advantage 100 camera uses edge detection tools and Cognex’s proprietary IDMax algorithm to read the identification codes of each tube, while the other Advantage 102 color camera captures two images of each tube under two different color light sources. The blood separation workstation ensures the pipette is positioned correctly for aspirating the white membrane layer based on its depth, for final analysis and diagnosis. This operation is easy and saves a lot of time.

Unlike other deep learning solutions, Cognex uses the following four basic tools: localization, analysis, classification, and reading, which makes it easier for users to debug when developing solutions. With VisionPro Deep Learning software, designers can break complex problems into smaller tasks that can be optimized individually, making them easier to understand and maintain. Thus, Cognex’s integrated software environment means that customers like PerkinElmer can build deep learning solutions using hundreds of images instead of thousands or tens of thousands of images, significantly shortening the time to market for OEM customers.

PerkinElmer’s automated blood separation workstation was launched in January this year and received unanimous praise from the industry. “Earlier this year, our team launched a blood separation workstation for research use only,” said Atwood from PerkinElmer. “The project was born out of the demand for intelligent liquid handling of layered blood, which is the front-end liquid handling platform for our industry-leading chemagen™ nucleic acid extraction technology. Cognex’s life sciences OEM team provided us with strong support. From project initiation to successful implementation completion, we have been working with Cognex, and their expertise in image-based deep learning and machine vision technology has been invaluable to us.”