Direct 3D Reconstruction Methods for Electrical Impedance Tomography for Stroke Imaging

点击翻译按钮获取中文摘要

Project Summary Every four minutes an American dies from stroke, equating to approximately 1 in every 19 US deaths annually. Strokes are classified as ischemic and hemorrhagic. Ischemic strokes make up 87% of all strokes and are caused by a blockage in a blood vessel (or artery) resulting in a lack of blood to the brain. A hemorrhagic stroke occurs when an artery in the brain leaks or ruptures releasing excess blood in or around the brain. Incorrect classification can have dire consequences as treating a patient suffering from a hemorrhagic stroke (bleed) with anticoagulant drugs (used to dissolve blood clots for ischemic strokes) can prove fatal. Early action is of the utmost importance as each passing minute that brain cells lack the proper blood flow additional cells die. Current classification methods require tests performed at the hospital, e.g., CT or MRI scans of the patient's brain, leading to treatment delays. These delays are particularly lengthy for patients living in rural communities. Electrical Impedance Tomography (EIT) is an emerging medical imaging modality that is inexpensive, has no ionizing radiation, and provides portable high-contrast images using harmless surface current and voltage measurements (e.g., on the head using a flexible hat) to recover the internal point-wise electrical properties (e.g., inside the brain). EIT can recover conductivity, a measure of how easily current flows through a material, as well as permittivity, a measure of the ability of a material to store a charge. A hemorrhagic stroke corresponds to an area of abnormally high conductivity due to the bleed, whereas an ischemic stroke presents as an area of lower conductivity than expected due cellular swelling from energy failure. The proposed project addresses the important problem of early, fast, portable stroke classification with EIT. A critical barrier for the use of EIT for stroke imaging has been the sensitivity of the image reconstruction algorithms to incorrect domain modeling and noise in the data. Due to these challenges, most research has focused on monitoring applications, not helpful for the classification task. By contrast, the D-bar reconstruction method proposed here is the only proven noise and modeling error robust reconstruction method capable of recovering the true conductivity/permittivity using a low-pass filtering in a nonlinear Fourier domain. D-bar methods have been successful in 2D but their development in 3D is stunted. This proposal focuses on the development of fast, robust D-bar based reconstruction methods for the 3D partial boundary problem, critical to working with stroke EIT data. Numerical algorithms will be developed for the full and partial boundary problems in 3D and validated on simulated and experimental data. A priori information, from anatomical atlases, will be embedded into the methods for increased resolution and stability. As the low-pass filtering in D-bar methods leads to blurred reconstructions, post-processing through Convolutional Neural Networks will provide improved image quality. This work will be the first to develop robust computational algorithms for 3D EIT data, opening the door for stroke imaging.