Historically, bioimaging at the microscopic scale has been performed using lenses, usually the compound microscope and its adaptations, which are typically bulky and expensive. Lucendi's technology is based on digital holography (basic principle is shown on the Figure) where a diffraction pattern resulting from an object (based on, e.g., scattering or fluorescence) is recorded directly on a digital image sensor array without being optically imaged or magnified by a lens. This recorded diffraction pattern is then computationally reconstructed to form an “image” of the object. All of this can be done while locating the object in 3D volume within the sample, recoving its color, shape and other attributes for a follow on analysis.
Digital holography offers several advantages over competing micro-object measurement systems, including:
- Trading optical and mechanical for computational complexity
- Provides large space-bandwidth product (large FOV and high resolution)
- Cost-effective and portable
- Offers depth-resolved 3D imaging
- Supports color imaging
- Supports volumetric imaging (focusing on different heights of sample plane)
Once the digital holograms of the objects were captured and reconstructed they may still include imperfections (i.e. twin-image noise). For some applications this does not pose problems. However, if a refined image of the object is required, then phase recovery needs to be performed. Phase recovery is an interative, computationally intensive process, which naturally lands itself for optimization with Deep Learning. In our system a convolutional neural network is trained with a very large database of reconstructed images as an input and fed images that were phase recovered via a conventional algorithm. After training, Lucendi Deep Learning system is then capable of rapid phase recovery of an inputted reconstructed image.