2025-Jan-17, 9AM, Zoom
Yunmei Chen (University of Florida)
Title: Provably Convergent Learned Descent Algorithm for Low-Dose CT Reconstruction
Abstract: We propose a general learning based framework for solving nonsmooth and nonconvex inverse problems with application to low-dose CT (LDCT) reconstruction. We model the regularization function as the combination of a sparsity enhancing and a non-local smoothing regularization. We develop an efficient learned descent-type algorithm (ELDA) to solve the nonsmooth nonconvex minimization problem by leveraging the Nesterov’s smoothing technique and incorporating the residual learning structure. We proved the convergence of the algorithm and generate the network, whose architecture follows the algorithm exactly. Our method is versatile as one can employ various modern network structures into the regularization, and the resulting network inherits the convergence properties, and hence is interpretable. We also show that the proposed network is parameter-efficient and its performance compares favorably to the state-of-the-art methods.
[Inverse Problem Study Group Seminar] 2019-06-28 (16:30 ~), 산업경영학동(E2) Room 3221
Hyeuknam Kwon (Yonsei University, Wonju)
title: Recent advances in electrical bioimpedance
abstract:
Over the past decade, electrical bioimpedance has been undergoing a rebirth as enhanced methodologies and new theories are greatly extending its use in the field of neuromuscular disease (NMD). Simply put, NMDs change the structure and internal composition of skeletal muscle which, in turn, alter the electrical properties muscle. Thus, the capability of measuring the electrical properties of muscle with accuracy has great potential to provide valuable new insights to inform medical assessment and diagnosis of NMDs. One technique well-suited for measuring the electrical properties of muscle is electrical bioimpedance, where an electrical current is applied to the muscle using two electrodes and the resultant voltage is measured using two additional electrodes. However, the accuracy to detect onset of disease, track disease progression and response to therapy using surface electrodes placed on the skin is limited: data are largely influenced by skin and subcutaneous fat (SF) overlying the muscle. Here, we will present a new source separation (SS) technique that, unlike existing blinded SS techniques principal component analysis (PCA) and independent component analysis (ICA), can distinguish muscle from SF with the accuracy being 99.2%.
However, the standard procedure of patient care for diagnosing NMDs consists of inserting needles electrodes into the muscle to measure the electrical activity at rest and during muscle contraction. To take advantage of this, we have designed an enhanced needle device also integrating impedance recording capabilities. Our new needle improves the accuracy measuring the electrical properties by recording these properties and their direction dependence directly in the muscle, the latter also known as anisotropy. Ongoing work in this area promises exciting and valuable new applications in the years to come.
2018-11-23 (15:00 ~16:30), 산업경영학동(E2) Room 2222
Won-Kwang Park (Kookmin University)
title: Direct sampling method in inverse scattering problem: from theory to real-world application
abstract: Direct sampling method (DSM) is a well-known, non-iterative imaging technique in inverse scattering problem. Throughout various researches, DSM has been applied various research area e.g., diffusion tomography, electrical impedance tomography, source detection in stratified ocean waveguide, etc.; however, due to the small number of incident fields or sources, further improvements are still required. In this presentation, we carefully identify mathematical structure of indicator function of DSM to show the feasibilities and limitations, design a method of improvement, and apply in real-world microwave imaging. Simulations results with synthetic and experimental data are shown for supporting identified structure.