Meeting Banner
Abstract #2251

Statistical Wavelet Structure Based MRI Compressed Sensing Reconstruction Using a Hidden Markov Tree Model

Enhao Gong1, Xiao Wang1, Kui Ying2, Shi Wang2

1Biomedical Engineering, Tsinghua University, Beijing, China; 2Engineering Physics, Tsinghua University, Beijing, China

Compressed sensing (CS) is an emerging acceleration technique and recently applied for MRI. Conventional CS reconstruction techniques are based on simplistic sparsity of signals and use uniform L1-norm penalty regardless of whether the coefficients contribute to significant information for pathological diagnosis. This results in reconstruction errors, like blurring details. We proposed a new algorithm that uses Hidden Markov Tree model to extract structural information in wavelet domain. Sparsity is regulated by exploiting statistical structural matrices, such that important coefficients are enhanced and artifacts are further reduced. Phantom simulation and in-vivo experiments show the validity and advantages of our algorithm.

Abstract PDF Poster PDF

Keywords

abdominal acceleration achieved acquisition applied approximate artifacts assumption becomes better beyond binary biomedical blurring brain captured china coefficient coefficients complexity compressed computational computed connected considering constrained contribute decaying defined density dependence details diagnosis differentiate distribution domain edges efficient employed encoding engineering enhanced enjoy errors estimation exhibit experimental exploiting exponentially extract faster fewer fine finer fold fully functions gains gong healthy hidden important improve in vivo isolated iterations iterative just like linear magnitude magnitudes matrices matrix measured medical mixture model modeled modeling models negligible noise norm noted obey operator optimized partial pathological penalize penalty people performance persistence persists phantom physics plot preserve preserved probabilities probability progress properties property propose proposed quad quantitative recently reconstructed reconstruction recovery reduced reduction regardless regulate regulating remove republic sampled sampling scale scales selectively sensing several sharing significance simplistic simulation since sparse sparsity statistical stronger structural structure studies subject supports thresh threshing transition tree tuning uniform views wavelet whether zoom