Deep Learning for Fast and Spatially-Constrained Tissue Quantification from Highly-Undersampled Data in Magnetic Resonance Fingerprinting (MRF)

摘要: Magnetic resonance fingerprinting (MRF) is a novel quantitative imaging technique that allows simultaneous measurements of multiple important tissue properties in human body, e.g., T1 and T2 relaxation times. While MRF has demonstrated better scan efficiency as compared to conventional techniques, further acceleration desired, especially for certain subjects such infants young children. However, the framework only uses simple template matching algorithm quantify properties, without considering underlying spatial association among pixels signals. In this work, we aim accelerate acquisition by developing new post-processing method accurate quantification with fewer sampling data. Moreover, improve accuracy quantification, signals from surrounding are used together estimate at central target pixel, which was simply done signal pixel original method. particular, deep learning model, i.e., U-Net, learn mapping evolutions property map. To reduce network size principal component analysis (PCA) dimensionality input Based on vivo brain data, our can achieve both using 25% time points, four times data