Skip to main content
. Author manuscript; available in PMC: 2020 Oct 1.
Published in final edited form as: Med Phys. 2019 Aug 31;46(10):4610–4621. doi: 10.1002/mp.13756

Figure 4.

Figure 4

In vivo apparent T1, T2 and B0 estimation using the proposed multi-task deep neural network with T1 and T2 range limitations, single-task deep neural network with T1 and T2 range limitations, and dictionary-element-like quantization on T1, T2, and B0, as well as the naive single-task approach with different levels of T1 and T2 limitations. Data were acquired with 3D IR-bSSFP, 30 inversion times, and 2 different frequency offsets at 0 and 1/(2×TR). The single-task and dictionary-element like quantization were vulnerable to artifacts in T1, T2, and B0 maps. On the other hand, the deep neural network outputs with T1 and T2 range limitations appeared more accurate relative to brain anatomy and provided more continuous T1, T2 and B0 values.