Deep Learning Imaging through Fully-Flexible Glass-Air Disordered Fiber Jian Zhao,* ,† Yangyang Sun, † Zheyuan Zhu, Jose Enrique Antonio-Lopez, Rodrigo Amezcua Correa, Shuo Pang, and Axel Schü lzgen CREOL, The College of Optics and Photonics, University of Central Florida, Orlando, Florida 32816, United States * S Supporting Information ABSTRACT: We demonstrate a fully flexible, artifact-free, and lensless fiber-based imaging system. For the first time, this system combines image reconstruction by a trained deep neural network with low-loss image transmission through disordered glass-air Anderson localized optical fiber. We experimentally demonstrate transmission of intensity images through meter-long disordered fiber with and without fiber bending. The system provides the unique property that the training performed within a straight fiber setup can be utilized for high fidelity reconstruction of images that are transported through either straight or bent fiber making retraining for different bending situations unnecessary. In addition, high quality image transport and reconstruction is demonstrated for objects that are several millimeters away from the fiber input facet eliminating the need for additional optical elements at the distal end of the fiber. This novel imaging system shows great potential for practical applications in endoscopy including studies on freely behaving subjects. KEYWORDS: transverse Anderson localization, microstructured optical fiber, fiber imaging, lensless imaging, convolutional neural network F iber optical endoscopes (FOEs) are very important and widely used tools in biomedical research, clinical diagnostics and surgical operations. 1,2 They enable imaging under conditions in which conventional microscopy cannot work well. For example, FOE-based optical imaging can be performed for cells reside within hollow tissue tracts or deep within organs in a minimally invasive way, while those locations are inaccessible for conventional microscopy. 2,3 Furthermore, FOEs can be implanted within freely behaving subjects, such as mice, for long-term imaging research. 4,5 This ability can benefit several areas, such as fundamental biological research and application research in developing in vivo methods for drug testing. Different types of fibers have been proposed for FOEs. 1 However, current solutions suffer from several limitations regarding system complexity and size, image quality and bending sensitivity. For example, single mode fibers can be used as the smallest imaging acquisition unit, but a mechanical scanning head or a spectral encoding device is often installed at the distal end of the fiber to deliver 2D imaging information, which makes the system bulky and complex. 1,6,7 Alternatively, commercially available fiber bundles are able to transport 2D imaging information directly through FOEs. However, pixelation artifacts dictated by the individual cores fundamentally limit the transported image quality. In addition, any crosstalk between fiber cores in the bundle blurs the transmitted images 8 and the cost of materials and fabrication of fiber bundles is rather high. 9 Instead of utilizing thousands of fiber cores, the thousands of different spatial modes in multimode fiber (MMF) can also be used to transmit 2D imaging information. Current MMF based FOEs mainly rely on compensating randomized phases by wavefront shaping after calibrating the transmission matrix (TM) of the fiber. 5,10-14 This method suffers from several limitations. The most critical one is its intolerance to fiber movement. Any tiny movement or bending of the MMF will change the TM resulting in impaired imaging unless recalibration is performed, or very precise knowledge of the bending and its shape is Received: June 20, 2018 Published: September 17, 2018 Letter pubs.acs.org/journal/apchd5 Cite This: ACS Photonics 2018, 5, 3930-3935 © 2018 American Chemical Society 3930 DOI: 10.1021/acsphotonics.8b00832 ACS Photonics 2018, 5, 3930-3935 Downloaded via UNIV OF CENTRAL FLORIDA on October 24, 2018 at 00:37:39 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.