Project Code: PN-III-P1-1.1-TE-2019-1339
Contract Number: TE200/2021
Project Title: Augmenting Micro- and Nanoscale Optical Imaging Techniques with Generative Adversarial Networks
Duration: 24 months (25/01/2021-31/12/2022)
Total project value: 431.900 lei (86.000)
Abstract:
At present we are witnessing unprecedented possibilities for studying the properties of biological species and advanced materials at resolutions ranging from micro- to nanoscale. Despite that, many questions in fundamental fields of science remain still unanswered due to the absence of imaging modalities with the required characteristics and performance. In OPTIGAN we will address this problem by combining latest hour optical characterization techniques with a complementary technology, namely an emerging Deep Learning method known as Generative Adversarial Networks (GAN). OPTIGAN will develop novel GAN-based methods capable to significantly augment the potential of micro- and nanoscale optical imaging techniques operating in the far-field and near-field regimes. The developed methods for GAN-based data simulation, cross-modality and information forecast will make possible faster and easier sample characterization, and will enable novel optical investigation frameworks based on real and virtual imaging modalities, ranging from traditional to cutting-edge technologies.
Objectives:
OPTIGAN’s main objectives consist in:
– Novel GAN-based methods to simulate data collected at best system performance. These methods will enable simulation of data collected with optimal/best imaging conditions based on data collected with imaging conditions that are more appropriate for a specific experiment (e.g. simulation of images collected under high-beam power, based on images collected under low-beam power, in the case of light-sensitive samples)
– Novel GAN-based methods for cross-modality imaging. These methods will enable cross-modal simulation of images corresponding to techniques of low-availability (e.g. emerging ones), using as input images collected from wider spread techniques(e.g. simulation of Re-Scan Second Harmonic Generation Microscopy images from conventional Second Harmonic Generation images, or simulation of scattering-type Scanning Near Field Optical Microscopy data from Atomic Force Microscopy data)
– Novel GAN-based methods for information forecast. Such methods will provide a data/information forecast for sample regions that cannot be imaged with a specific technique due to physical constraints or contrast mechanism limitations (e.g. forecast of s-SNOM images for deeply buried sample regions).
– Design and implementation of novel correlative imaging applications based on GAN-powered cross-modality leading to better understanding of nanoscale data sets collected with emerging nanoscopy techniques
Expected Results:
-An extensive set of GAN-based methods that significantly augment high-and ultra-high optical imaging modalities (available at CMMIP-UPB, and not only)
-Novel characterization approaches based on GAN-based data simulation, cross-modality and information forecast
-Better understanding of biological species and advanced materials by combining experimental and GAN-based virtual data
-Competitive team of young researchers conducting high quality research on Deep Learning augmented high-resolution optical imaging
-Consolidated research collaborations with partner institutions and new ones with research teams highly experienced in the field of microscopy, nanoscopy, bioimage analysis, cellular and molecular biology, biochemistry and medicine, advanced materials
-Increased visibility of the host institution and of the research team involved in this project via publications in high impact factor journals, or invited talks at prestigious conferences
-Increased capacity of the research team and of the host institution to successfully apply for European and international financing instruments, and for intellectual property rights
Funding Agency :
Executive Unit for Higher Education, Research, Development and Innovation Funding (UEFISCDI)