Abstract:
|
In order for NMR imaging to detect early pathological changes with precision and reproducibility in diseased muscles, one needs quantitative imaging protocols. Unfortunately, inhomogeneities of static and rotating (B1) transmit and receive fields are an obstacle to muscle structure quantitative analysis. Conventional normalization techniques rely on the assumption that the signal is homogeneous inside each tissue. Such an assumption is not valid when dealing with pathological cases. In this work, we introduced novel techniques that relied on the assumption of subcutaneous fat signal homogeneity. We used two parametric models to approximate the nonuniformity function: sum of cosine functions and sum of Legendre polynomials. These functions have the ability to model any low frequency function. To estimate the model parameters we minimized a cost function composed of two terms: one referred to the uniformity assumption in the subcutaneous fat, the other reduced the amount of variation of the estimated non-uniformity function.The correction was applied to 3D NMR images from thighs of 13 healthy subjects. They were acquired on a 3T scanner using 3pt Dixon sequence (TR=10ms, TE1=2.75ms TE2=3.95ms TE3=5.15 ms, flip angle =3_). The NMR imaging data were obtained using a CP extremity coil (n=6) and a body matrix coil (n=7). First, we estimated the B1 transmit non-uniformity using the AFI technique to correct the water and fat volumes. Next, we estimated the B1 receive non-uniformity, using the fat volume and the subcutaneous fat information. Quality evaluation was done by computing the coefficient of variation (CV). For healthy subjects, we determined the CVs of the subcutaneous fat using fat volumes and inside the muscles using the water volumes. Results showed that for the body matrix coil, the use of Legendre polynomials or the cosine functions yielded close results. The CVs were reduced by a mean value of 60% for the subcutaneous fat, and by 42% for the muscles. Regarding the CP extremity coil, the mean value of CVs reduction in the subcutaneous fat area was 43% with cosine functions and 50% using polynomials. For muscles, CV mean reduction value was 24% with both models.Figure 1 displays a Dixon-based water fraction of an axial slice of the thigh and illustrates the homogeneity improvement obtained using our technique. These preliminary data suggest that such an approach might significantly facilitate truly quantitative muscle imaging protocols.
|