應用盲壓縮感知加速多激發角度T1圖譜的可行性研究

Abstract

多激發角度T1圖譜是一種基於非平衡梯度回訊序列的T1定量方法，此方法利用多組來自不同激發角度的影像以準確估計T1，因此常需要在掃描時間及不同激發角度的數量上作權衡。本論文探討了利用盲壓縮感知以加速多激發角度T1圖譜的可行性，這種技術可以在不預先假設一組固定的基底函數之下，使用稀疏限制重建影像。我們使用真實大腦影像，運用後處理的方式比較不同取樣方式的結果並討論演算法參數的選擇。我們在初步的結果中展示了5倍加速後的定量T1圖譜，希望能提供此更快速的多激發角度T1圖譜實際應用時的參考。

Variable flip angle T1 mapping is a T1 quantification technique based on spoiled gradient echo (SPGR) sequence. This method requires multiple images at different flip angles for accurate T1 estimations, so a trade-off is often made between acquisition time and the number of flip angles. This thesis aims to investigate the feasibility of faster variable flip angle T1 mapping by incorporating blind compressed sensing. This technique used sparsity constraint to reconstruct images without the assumption of a fixed dictionary. We retrospectively subsample in vivo brain images and compare the results of quantitative T1 maps with a 5-fold acceleration and provide a reference for practical implementation of a faster variable flip angle T1 mapping.