High Accuracy Image Reconstruction Using Microwave Measurements from Bio-Matched Antennas and Deep Learning: A Synthesized X-ray Computed Tomography Approach

Ohio State University, The, Columbus OH

Investigators

Abstract

Several technologies are clinically available to image biological tissues, each with their own merits and limits. Focusing on stroke, the application of interest in this proposal, X-ray computed tomography (CT) and magnetic resonance imaging (MRI) are typically used. Though the spatial resolution is excellent, their hardware is bulky and not suitable for bedside applications. Furthermore, the ability to differentiate between ischemic and hemorrhagic strokes in the ambulance or on-site and for bedside monitoring will have significant potential to improve outcomes and reduce mortality. In this context, microwave tomography is a promising imaging modality, yet it suffers from poor imaging resolution that restricts its clinical use. In this research, an expansion of the fundamental limits of microwave tomography resolution is proposed via an alternative imaging modality that combines the advantages of X-ray CT (high resolution) and microwave tomography (non-ionizing, low-cost, portable). The approach uses non-ionizing microwave measurements and a deep learning neural network to estimate data that would have been collected by an X-ray CT scanner at different angles around the patient. We expect the science developed in this research to be of great use in myriads of healthcare applications (imaging, radiometry, implant telemetry/powering, ablation, etc.) and beyond (e.g., industrial imaging applications). In addition to the intellectual advances, the proposed research is expected to be of significant interest to students and the public. Through interdisciplinary education and diverse recruitment efforts, we intend to expose new audiences to STEM concepts via workshops and family-friendly outings. The proposed research leverages advances in: (a) deep learning to synthesize X-ray CT projection data while relying solely on non-ionizing microwave tomography measurements, and (b) new classes of into-body radiating antennas, namely bio-matched antennas, with unprecedented efficiency of electromagnetic wave propagation towards human body. With the estimated CT projection data in hand, images can be reconstructed using standard CT reconstruction methods, such as filtered back projection. These images are referred to as synthesized CT and an improvement of more than two times over current state-of-the-art peak signal to noise ratio (PSNR) is targeted to provide good image reconstruction. Without loss of generality, focus is on stroke as an example application. The specific goals are: (1) developing a deep learning neural network to learn the complex relationship between microwave tomography measurements and X-ray CT projection data using synthetic/simulation data and line sources in two dimensions, (2) developing a theoretical modeling and experimental framework for bio-matched antennas with unprecedented efficiency of electromagnetic wave transmission towards human body while also being versatile for diverse applications, (3) integrating the deep learning neural network with optimized bio-matched antennas by considering three dimensional scenarios and building a prototype head imager for validation on head phantoms. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.