This work targets those interested in fast quantitative imaging and breast MRI.MRI is increasingly important in breast imaging for lesion detection and characterization. While it provides high sensitivity (approximately 90%), its specificity is relatively low (between 37% and 86%) because many benign lesions exhibit features similar to cancer¹. Previous breast MRI has also demonstrated significant difference in both T₁ and T₂ relaxation times between malignant and benign lesions². However, quantitative measurements of relaxation times are rarely performed in clinical settings because of the long scan time, especially for volumetric coverage. Recently, a new quantitative imaging framework, named MR Fingerprinting (MRF), has been introduced, which can provide rapid and simultaneous quantification of both T₁ and T₂ relaxation times³. The objective of this study is to adopt this MRF framework to develop a quantitative method for 3D relaxometry in breast imaging.

The MRI experiments were performed on a Siemens 3T Verio scanner using a breast coil with 12 receive elements. A FISP-based MRF method, which was originally developed for 2D cardiac imaging, was modified for 3D breast imaging⁴. The MRF data were acquired sequentially through partitions, as shown in Fig. 1. For each partition, the data acquisition was divided into 16 segments, each with a different combination of a fat-saturation module, a inversion-recovery module and T₂-sensitivity module for effective T₁ and T₂ sensitivity. In the following data acquisition window, 48 uniform-density spiral arms were acquired in 48 TRs with variable flip angles ranging from 5° to 12°. A high in-plane reduction factor of 48 was used, so only one spiral arm was acquired for each partition within a 3D volume. The same combination of preparation modules and flip angle pattern was repeated for each partition and a constant time delay of 5sec was applied between partitions for the longitudinal recovery. Other imaging parameters included: FOV=40×40cm; matrix size 256×256; slice thickness 3mm; number of partitions, 16; partial Fourier in the partition direction, 6/8. The overall acquisition time for 16 partitions was about 3.5min.

To retrieve tissue properties (T₁, T₂ and proton density) from the MRF measurement, a dictionary including the signal evolutions from all possible combinations of parameters for a T₁ range of 100 to 3000ms and T₂ range of 10 to 500ms was calculated using Bloch simulations³. The acquired signal in each voxel of the highly undersampled volumes were then matched to an entry in the dictionary using pattern matching, which yielded all underlying tissue parameters used to generate this dictionary entry.

The accuracy of the proposed method was first validated using a phantom and the results were compared to those acquired from the standard method using the single-echo spin-echo sequences. The method was then applied to three normal volunteers (mean age, 21.3 years) and informed consent was obtained from all volunteers.

Fig. 2 shows the results of T₁ and T₂ relaxation times acquired from the central partition in a phantom experiment. A close match to the results from the standard method was observed for a large range of T₁ (200~1600 ms) and T₂ (20~105ms) values. Fig. 3 shows the average T₁ and T₂ values measured from the 10 vials in all 16 partitions. Despite of a small variation in T₁ (2.6%±2.1%) and T₂ values (11.1%±14.1%) at the end partitions, a consistent measurement was observed with a T₁ variation of 0.8%±0.3% and a T₂ variation of 3.5%±2.7% across the central 14 partitions.

Fig. 4 shows representative T₁, T₂ and proton density maps acquired from a normal volunteer. The 8 odd numbered partitions out of a total of 16 partitions are presented and the rest are omitted here for ease of viewing. An average T₁ of 1449±163ms and T₂ of 50±7ms for fibroglandular tissues were obtained from the three normal subjects, which agrees well with the literature values acquired from 3T⁵.

In the current study, a rapid and accurate volumetric relaxometry method was developed for breast imaging using the MRF technique and the method was tested in both phantom studies and normal subjects. A small variation mainly in T₂ relaxation times was observed at the edge partitions, which likely relates to slice profile uniformity. This variation can be corrected in the future by including the slice profile in the Bloch simulation. In this proof-of-concept study, a limited number of partitions were acquired in 3.5min with no data acceleration or undersampling along the partition direction. Data undersampling in this direction will be explored in the future to achieve whole-breast coverage within a similar time window.