Heuristic wavelet approach for low-dose EPR tomographic reconstruction: An applicability analysis with phantom and in vivo imaging

Abstract

Electron paramagnetic resonance imaging (EPRI) is a new functional imaging modality that can provide valuable in vivo physiological information and aids as a complimentary imaging technique to MRI and PET of tissues especially with respect to in vivo pO2, redox status and pharmacology. EPRI deals with the measurement of distribution and in vivo dynamics, using exogenous paramagnetic spin probes injected into the scan subject. The bio-clearance and dosage of these spin probes are issues of concern in EPRI. As a consequence, tomographic reconstruction from ‘noisy’ and ‘sparse’ number of projections is highly desirable in EPRI for the purpose of dose reduction and fast acquisition time respectively. The aim of research is to address the incompleteness of such projection data by developing a software which requires no change in acquisition hardware and/or no a priori knowledge of imaging process. The new software approach integrates soft computing and multiresolution within tomographic reconstruction to “intelligently” extract the details of importance at several levels of resolution. The new multiresolution reconstruction algorithm is based on wavelet transform with computational complexity same as the clinically used, filtered backprojection (FBP) method, except that the filters are now angle dependent. Gain in intelligence is achieved employing multiobjective genetic algorithm (GA) to find values for wavelet denoising threshold with optimum performance in terms of signal to noise ratio (SNR) and resolution of the reconstructed image. Feasibility of the approach for fast and low dose tomographic reconstruction is demonstrated with simulated SheppLogan head phantom. Subsequently, the experimental results with phantom and in vivo EPRI proves that the developed method can reduce the dose level and number of projections by 60–75% in tomographic reconstruction. In particular the quantitative analysis, using RMSE, PSNR and Liu’s error factor, shows that our approach outperforms the widely used, FBP and state-of-art wavelet-based tomographic reconstruction method in achieving image quality with acceptable diagnostic accuracy.