Showing 1–3 of 3 results for author: Seibel, E J

Real Time Full-Color Imaging in a Meta-Optical Fiber Endoscope

Authors: Johannes E. Froech, Luocheng Huang, Quentin A. A. Tanguy, Shane Colburn, Alan Zhan, Andrea Ravagli, Eric J. Seibel, Karl Boehringer, Arka Majumdar

Abstract: Endoscopes are an important component for the development of minimally invasive surgeries. Their size is one of the most critical aspects, because smaller and less rigid endoscopes enable higher agility, facilitate larger accessibility, and induce less stress on the surrounding tissue. In all existing endoscopes, the size of the optics poses a major limitation in miniaturization of the imaging sys… ▽ More Endoscopes are an important component for the development of minimally invasive surgeries. Their size is one of the most critical aspects, because smaller and less rigid endoscopes enable higher agility, facilitate larger accessibility, and induce less stress on the surrounding tissue. In all existing endoscopes, the size of the optics poses a major limitation in miniaturization of the imaging system. Not only is making small optics difficult, but their performance also degrades with downscaling. Meta-optics have recently emerged as a promising candidate to drastically miniaturize optics while achieving similar functionalities with significantly reduced size. Herein, we report an inverse-designed meta-optic, which combined with a coherent fiber bundle enables a 33% reduction in the rigid tip length over traditional gradient-index (GRIN) lenses. We use the meta-optic fiber endoscope (MOFIE) to demonstrate real-time video capture in full visible color, the spatial resolution of which is primarily limited by the fiber itself. Our work shows the potential of meta-optics for integration and miniaturization of biomedical devices towards minimally invasive surgery. △ Less

Submitted 1 November, 2022; originally announced November 2022.

Synergistic Network Learning and Label Correction for Noise-robust Image Classification

Authors: Chen Gong, Kong Bin, Eric J. Seibel, Xin Wang, Youbing Yin, Qi Song

Abstract: Large training datasets almost always contain examples with inaccurate or incorrect labels. Deep Neural Networks (DNNs) tend to overfit training label noise, resulting in poorer model performance in practice. To address this problem, we propose a robust label correction framework combining the ideas of small loss selection and noise correction, which learns network parameters and reassigns ground… ▽ More Large training datasets almost always contain examples with inaccurate or incorrect labels. Deep Neural Networks (DNNs) tend to overfit training label noise, resulting in poorer model performance in practice. To address this problem, we propose a robust label correction framework combining the ideas of small loss selection and noise correction, which learns network parameters and reassigns ground truth labels iteratively. Taking the expertise of DNNs to learn meaningful patterns before fitting noise, our framework first trains two networks over the current dataset with small loss selection. Based on the classification loss and agreement loss of two networks, we can measure the confidence of training data. More and more confident samples are selected for label correction during the learning process. We demonstrate our method on both synthetic and real-world datasets with different noise types and rates, including CIFAR-10, CIFAR-100 and Clothing1M, where our method outperforms the baseline approaches. △ Less

Submitted 27 February, 2022; originally announced February 2022.

RetinaMatch: Efficient Template Matching of Retina Images for Teleophthalmology

Authors: Chen Gong, N. Benjamin Erichson, John P. Kelly, Laura Trutoiu, Brian T. Schowengerdt, Steven L. Brunton, Eric J. Seibel

Abstract: Retinal template matching and registration is an important challenge in teleophthalmology with low-cost imaging devices. However, the images from such devices generally have a small field of view (FOV) and image quality degradations, making matching difficult. In this work, we develop an efficient and accurate retinal matching technique that combines dimension reduction and mutual information (MI)… ▽ More Retinal template matching and registration is an important challenge in teleophthalmology with low-cost imaging devices. However, the images from such devices generally have a small field of view (FOV) and image quality degradations, making matching difficult. In this work, we develop an efficient and accurate retinal matching technique that combines dimension reduction and mutual information (MI), called RetinaMatch. The dimension reduction initializes the MI optimization as a coarse localization process, which narrows the optimization domain and avoids local optima. The effectiveness of RetinaMatch is demonstrated on the open fundus image database STARE with simulated reduced FOV and anticipated degradations, and on retinal images acquired by adapter-based optics attached to a smartphone. RetinaMatch achieves a success rate over 94\% on human retinal images with the matched target registration errors below 2 pixels on average, excluding the observer variability. It outperforms the standard template matching solutions. In the application of measuring vessel diameter repeatedly, single pixel errors are expected. In addition, our method can be used in the process of image mosaicking with area-based registration, providing a robust approach when the feature based methods fail. To the best of our knowledge, this is the first template matching algorithm for retina images with small template images from unconstrained retinal areas. In the context of the emerging mixed reality market, we envision automated retinal image matching and registration methods as transformative for advanced teleophthalmology and long-term retinal monitoring. △ Less

Submitted 28 November, 2018; originally announced November 2018.