THANK YOU FOR SUBSCRIBING
Medical Tech Outlook Weekly Brief
Be first to read the latest tech news, Industry Leader's Insights, and CIO interviews of medium and large enterprises exclusively from Medical Tech Outlook
.webp)
Sophie Morales, Research Engineer and Project Manager, CEA-LetiA lens-free microscope is a simple device which records the diffraction pattern of a thin sample at a short sample-detector distance (typically <1mm) on a CMOS sensor. Focusing optics are not needed because the sample image is generated via computation, using holographic reconstruction techniques. Advantages of this time-lapse imaging technique include low cost, robustness, large field of view (
30mm²) and compact size, which allows direct dynamic monitoring of cell cultures within an incubator.
This novel optical method can provide robust real-time analysis of cell cultures based on cell count, cell tracking, and cell viability without any staining. It gives access to a large number of parameters useful for real-time cell culture monitoring, including cell area, dry mass, thickness, major axis length, and cycle duration. Moreover, all those single-cell metrics can be quantified across thousands of individual cell cycle tracks with high statistical significance, as the technology’s field of view is large enough to image tens of thousands of adherent cells and up to 40.106 floating cells/ml in a single shot.
“The convolution neural network—which is embedded in the holographic reconstruction—is applied to better initialize the gradient descent used in the inverse-problem approach.”
One challenge is the imaging of very thick cells (several microns) because the quality of the holographic reconstruction is reduced by the problem of phase-wrapping artefacts (i.e., phase measurements are invariant to shifts with 
an integer and 
the optical wavelength being used). This occurs when eukaryotic cells are in suspension or at the mitosis stage.
To solve this problem of phase under-determination, CEA-Leti has developed reconstruction algorithms based on an alternation of the inverse-problem method and deep-learning approaches. The convolution neural network—which is embedded in the holographic reconstruction—is applied to better initialize the gradient descent used in the inverse-problem approach. This allows bypassing of possible local minima, where the gradient descent can stagnate. Inverse-problem resolution simultaneously ensures data fidelity.
This dual approach has been successfully used to unwrapped lens-free reconstructions. We validated the method for phase-image recovery of high-density floating cell samples acquired via lens-free microscopy. This is a challenging case with many phase-wrapping artefacts that have not been successfully solved using inverse-problem approaches alone. The reconstruction scheme considerably reduces artefacts and will enable better quantification for further biological analyses. These developments open new perspectives for applications such as drug screening and monitoring of bio-production processes in incubators or bio reactors (including monitoring, counting and characterization of cultivated cells). This technology has been transferred to the French company Iprasense.
Read Also
