Sepsis — a severe, life-threatening reaction to infection within the body — requires a rapid, accurate diagnosis.

Avoiding lethal outcomes from sepsis — a severe, life-threatening reaction to infection within the body — requires a rapid, accurate diagnosis. Historically, it has been a challenge for healthcare providers to beat the clock and intervene with life-saving care. This has contributed to the disease’s lethality, making sepsis the leading cause of hospital-related deaths in the United States.

The VEO 710 features a custom CMOS sensor with 7 gigapixels per second (Gpx/s) of image throughput. (Credit: Vision Research)

To assist medical professionals in preventing sepsis-related deaths, Cytovale®, a life sciences technology company, developed the Intellisep® test, a commercially available medical device that detects sepsis early. By providing test results in under 10 minutes, this diagnostic tool provides clinicians with the probability that a patient may have sepsis. This supports their ability to make critical, time-sensitive decisions based on the physical and mechanical characteristics of the patient’s blood cells.

Key to Intellisep’s development was a Phantom high-speed imaging system, which Cytovale researchers used in combination with advanced microfluidic techniques and machine learning algorithms. The device’s design also marks the first time an embedded Phantom camera was used within a medical device with U.S. Food and Drug Administration (FDA) 510(k) clearance.

Designing the Microfluidic Experiments

Coupling the VEO 710 with a standard microscope enabled researchers to record the cell deformation process in the cross-slot microfluidic channel. (Credit: Cytovale)

Developing Intellisep required a careful study of the physical and mechanical properties of large quantities of cells. This information allowed the Cytovale® researchers to accurately identify and categorize a range of cell types in various stages of development and to also make statistical predictions of how the cells will develop. Traditional methods to observe and measure large collections of cells include atomic force microscopy and micropipette aspiration — methods that require significant time and labor.

Addressing the challenges of phenotyping large cell quantities, Cytovale created a cross-slot microfluidic channel as a stage to observe the cells’ characteristics. During the experiments, hundreds of thousands of cells suspended in a solution flowed through the channel at 3 m per second (m/s) to an observation zone measuring 45 × 120 μm. While under observation, the cells were subjected to hydrodynamic stresses to induce deformations.

To capture the cell deformation process, the Cytovale researchers used a high-speed Phantom VEO 710 camera, which they coupled with a standard inverted microscope to monitor the cells as they passed through the observation zone. Recording at 500,000 frames per second (fps) at 102 × 48 resolution, the VEO 710 camera captured roughly 2,000 cells per second. Each run involved 50,000 single cell events, providing the Cytovale team with both a spatially and temporally rich dataset from which they could extract valuable cell metrics like size, morphology and strain rate. See the sidebar, “A Critical Technology for Success in Cell Cytometry.”

A Critical Technology for Success in Cell Cytometry

By providing test results in under 10 minutes, the Intellisep diagnostic tool indicates to clinicians the probability of sepsis within a patient. (Credit: Cytovale)

To achieve high-quality cytometry images, cameras require larger pixels — typically between 10 and 30 μm in edge — as well as fast frame rates up to 10,000 fps, ultra-short exposure times and excellent light sensitivity. The Phantom VEO 710 meets these requirements, featuring a 20-μm pixel size and frame rates up to 680,000 fps at reduced resolutions. The camera’s large pixel size, coupled with the 35-millimeter, 1-megapixel CMOS sensor, achieves high light sensitivity even at fast recording speeds.

Other notable features of the VEO 710 include:

  • Frame rates of 7,400 fps at full resolution (1280 × 800) and up to 1,000,000 fps at reduced resolutions and with the export-controlled FAST option.

  • A sensor format that is compatible with various microscopic lenses.

  • Compact, lab-friendly housing.

  • 10 Gb Ethernet download option, enabling the Cytovale team to capture and offload massive amounts of data for fast analysis.

  • Minimum exposure time (standard) of 1 μsec, as well as 300 ns with the FAST option.

  • These ultra-short exposure times enabled the team to capture the fast-moving cells within the microchannel without motion blur.

Thanks to this combination of features, the VEO 710 can provide valuable insights into many biological and biomedical events that are otherwise too fast to be seen.

The Intellisep® device. (Credit: Cytovale)

Critical to the success in this application, the VEO 710 features a custom CMOS sensor with 7 gigapixels per second (Gpx/s) of image throughput, supporting the fast offload of data and contributing to Intellisep’s fast, 10-minute sample turnaround time.

Machine Learning Algorithms for Analysis

The Intellisep device works by analyzing the physical characteristics of cells as they undergo deformation to determine the likelihood of sepsis. During its design, the Cytovale team had to determine which properties were better indicators of cell type and future sepsis development. To this end, the team trained support vector machines (SVM), or supervised learning models with algorithms that analyze data for classification and regression analysis. Cytovale utilized SVMs to classify cell types and then weigh the importance of cell parameters.

The researchers used SVMs and control samples to establish a baseline for cell identification and then trained new SVMs with the full data set. They tested the results against an unlabeled mixture of cells to determine the model’s identification accuracy. Next, the team trained the SVMs to rank how important each parameter was, introducing one parameter to the model at a time and then checking the identification results.

The team repeated this process with a new SVM for multiple iterations. After each SVM run, the researchers introduced the most important parameters from the previous iteration earlier into the procedure, refining the accuracy of the model and allowing it to correctly identify greater quantities of cells each time. Enabled by the large quantity of high-fidelity data captured by the VEO 710 camera, this machine learning process was critical to Intellisep’s design, allowing the Cytovale researchers to refine the predictive power of their technology and deliver a fast and accurate diagnostic result.

Looking Ahead to the Future

Not only did the Phantom VEO 710 support the foundational research for Cytovale’s new test, it also facilitated Intellisep’s production. The camera enabled Cytovale to scale its design to a commercially available tool, making it the first time an embedded Phantom high-speed camera was used within a medical device with FDA 501(k) clearance. Intellisep is only just beginning, with Cytovale planning to use Phantom cameras for future research and product development.

This article was provided by Vision Research, Wayne, NJ. For more information, visit here .