Work

End-to-End Mapping of Microwave Data to Synthetic Computed Tomography Sinograms for High-Resolution Stroke Imaging

Year: 2026

Type: article

Abstract: Computed Tomography (CT) is the gold-standard imaging modality for stroke applications, but its hardware is complex, bulky, and costly, and prohibits safe and continuous monitoring due to the use of ionizing radiation. On the other hand, microwave (MW) imaging (MWI) modalities exhibit advantages in terms of portability, cost, safety, and continuous monitoring, but they are ill-posed and suffer from low resolution. In an attempt to overcome these limitations, we report MSSNet, a neural network model developed using synthetic data modeling, which takes MW data (scattered fields from a TMz line source) and maps them to the sinogram projection obtained from CT imaging over anatomical 2-dimensional (2D) head models. MSSNet consists of a mapping block and an enhancement block containing custom spatial-spectral blocks (SSB). The trained model has been tested over arbitrary 2D synthetic slices containing ischemic and hemorrhagic strokes. As reference media for extracting the MW data, both a coupling medium (CM) and free space (FS) have been considered. MSSNet provides high-accuracy mapping from MW data to sinograms with Peak Signal to Noise Ratio (PSNR) above 25 dB across different training scenarios. Comparing the absolute and differential images between MSSNet-mapping and a standard MWI algorithm (Born Iterative Method) shows a significant improvement in the Normalized Cross Correlation (NCC > 0.90) with the ground truth, when CM reference is considered. MSSNet-mapping also provides meaningful images with FS reference, where standard MWI fails completely. This approach may pave the way for low-cost, portable, and clinical-grade resolution MWI. Future studies will refine the model to learn more general physics, as well as study the transition of performance over 3D scenarios with realistic antennas.

Source: IEEE Access

DOI: 10.1109/access.2026.3727144

Authors: M. Shifatul Islam, Golrokh Mirzaei, Md Asiful Islam, Asimina Kiourti

Institutions: The Ohio State University, Bangladesh University of Engineering and Technology

Language: English

Topic: Microwave Imaging and Scattering Analysis

Subfield: Biomedical Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Good health and well-being

Open Access status: gold

Estimated APC paid: $2,075