Use case

Automated Microscopy for High Throughput Blood-Cell Analysis via CNN

What is high-throughput microscopy all about?

Which classes of blood cells are laboratory technicians seeing in a sample under a microscope? How many cells are there of each, and what are their sizes and shapes? Answers to questions like these help doctors diagnose diseases such as malaria, tuberculosis or hematological-oncological disorders. However, a look through a microscope is error-prone as well as time-consuming and therefore costly. A computer vision system based on a CNN (Convolutional Neural Network) automates the process.

What is the challenge in manual blood cell analysis?

Lab analyses of blood smears should be reliable, fast and yet inexpensive. All of this is made possible by automating the analysis process with the help of a CNN-based computer vision system. But the requirements for high resolution and high speed result in large data amounts that must be transmitted and processed quickly.

The solution for high-throughput microscopy: blood cell analysis via CNN

In our demo, we use a CNN to identify Plasmodium genera in blood smears, which can cause the infectious disease malaria. The CNN sorts the Plasmodia into one of seven predefined classes and counts them. The result makes it possible to reliably determine the form of malaria. To solve this or a similar problem, we will select the right hardware and software components for you and will configure them into a coherent computer vision system.

The system hardware in this demo includes a dual-channel boost color camera with 20 MP resolution, 1.1” sensor format and CXP-12 interface. The camera enables you to capture very rapidly cycled images with a high resolution – the basis for high scanning speed or a high sample throughput. The (pre-)processing and analysis of the high data amount requires additional suitable components. In this case, the programmable imaWorx CXP-12 Quad frame grabber handles not only the image processing and analysis but also supports features such as the autofocus with the help of FPGA-based real-time data pre-analysis. A C-mount lens and two CXP-12 data cables complete the system hardware.

Basler’s VisualApplets software is used to program the frame grabber FPGA. The configuration and programming of the data pre-analysis, as well as the implementation of the CNN on the frame grabber FPGA are performed on the software’s graphic user interface. The CNN is trained beforehand on the host side. However, the frame grabber FPGA provides sufficient capacity for the implementation of the CNN and enables high-performance inference. Depending on the customer requirements, the CNN is either trained once and implemented on the frame grabber’s FPGA or the customer is given the option to adapt the CNN later.

The advantages of a vision solution for automated microscopy

  • Excellent image quality of the boost camera thanks to Sony Pregius S sensors (e.g. IMX531)

  • Optimal interaction of the hardware, which quickly generates large data amounts and processes them with high-performance (20 MP CXP-12 camera and frame grabber)

  • Data pre-analysis and CNN-based evaluation are performed in real time with over 900 MB/s on the frame grabber, reducing the system requirements and costs on the host side

  • High reliability of the classification and very fast availability of results

Products for this solution

Looking to implement a comparable solution? These products will help you.

Put the setup together yourself in the configurator.

To the configurator

How can we support you?

We will be happy to advise you on product selection and find the right solution for your application.