Apr 20 2020
DRVISION Technologies LLC, a pioneer in computer vision and the world's leading AI microscopy software company has been awarded up to 2.16 million USD from the National Institute of General Medical Sciences (NIGMS) to develop the AI Restoring and Staining platform (AIRS).
The R&D funded by this Fast-Track collaborative research grant (# 1U44GM136091-01) will further boost the machine learning enabled solutions currently available in Aivia and Aivia Cloud. The project will be conducted in collaboration with Dr. Hari Shroff (National Institute of Biomedical Imaging and Bioengineering, (NIBIB)), his wider team and a consortium of seven world-class imaging centers across the USA.
Over the last two years, DRVISION researchers in collaboration with Dr. Hari Shroff and his team have successfully created and tested a wide range of deep learning (DL) models for restoration and/or improvement of image resolution and quality. These early studies illustrated how this technology can be used to run experiments that were previously impossible and have uncovered several potential limitations including usability, robustness, data friendliness, trustworthiness, effectiveness, versatility and scalable computing. Addressing these will enable the wider microscopy community to adopt and routinely use AI Microscopy approaches with greater confidence.
"We hope that this work becomes a valuable resource for the field, and spurs further technical developments and applications in image restoration." mentioned Dr. Hari Shroff, a NIH Senior Investigator, Chief of the Laboratory of High Resolution Optical Imaging, Managing Director of the trans-NIH Advanced Imaging and Microscopy Resource and the key academic collaborator on the AIRS project.
This project aims to systematically address the limitations identified in our pilot studies as well as by the research community. Addressing these will accelerate the adoption of DL to greatly benefit the microscopy and the wider AI communities. Namely this project aims to provide (1) a comprehensive suite of validated DL models for microscopy applications; (2) a plug-and-play solution for common imaging routines; (3) semi-automatic update training to tailor pre-trained DL models to match new imaging experiments; (4) user friendly support for new DL model training; (5) confidence scores to assess the output results created by DL models; (6) DL models that avoid image artifacts and allow continuous learning and evolution; (7) streamlined access to the required computing and data storage resources.
To accelerate the adoption of DL for broad microscopy communities, we have thoroughly studied and defined the optimal DL pipeline and computing workflow to benefit most labs in advanced microscopy experiments. This project is enabled by DRVISION's innovative technologies (patents pending) that can handle paired and unpaired Ground Truth, can match models to data and can reject image artifact. The technology also allows continued model improvement through active learning. We look forward to starting this exciting 3-year project to push the frontiers of what is possible in microscopy."
Dr. James Lee, DRVISION's Founder, President and CEO and Principal investigator of the project
"Our team has a single objective - help advance scientific discovery. Deep Learning - primarily convolutional neural networks - can be used to enable microscopy experiments that would previously be impossible, and thus herald in a wave of new discoveries. In this ambitious project we will focus on the hard task of quantifying and addressing (mitigating) the limitations that are intrinsic to the use of AI technology for microscopy applications. Aivia users will be the first to experience the results of this exciting R&D effort" said Dr. Luciano Lucas, DRVISION's Executive Vice President and AIRS' Project Manager.
DRVISION will work closely with Dr. Hari Shroff and his wider team at the NIBIB in addition to 7 other collaborating sites from across the USA (Columbia University, Rockefeller University, Princeton University, Texas A & M, The Marine Biology Laboratory - University of Chicago, University of Michigan and Wake Forest University). The collaborating sites will provide test data sets, detailed user feedback and validate AIRS' platform using a variety of imaging modalities, samples and experimental set ups for applications including SNR restoration, super resolution restoration, spatial deconvolution, spectral unmixing, 2D to 3D images and organelle virtual staining.