The purpose of this study is to achieve better reconstruction of Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCEMRI) with higher acceleration and to obtain better Pharmacokinetic (PK) modelling results at these higher acceleration factors retaining the quality of image.

In the proposed work, the algorithm is tested on in-vivo data instead on phantom data as in ¹. Also, the acceleration factors are increased to 8X/6X to increase the acceleration by a factor of 2 in both Temporal Resolution (TR) and Spatial Resolution (SR) regions (shown in figure 1) taking different undersampling masks in both regions, thus resulting in different accelerations and undersampling in both these regions. The method includes view-sharing that is done by mapping the samples from SR region onto TR region at corresponding frames in both regions. The mapping of samples is done such that there will not be any occurrence of scaling errors. For eg: the samples from first frame in SR region are mapped onto the last frame of TR region. Mapping is performed by taking the high frequency samples of frame in SR region onto corresponding frame in TR region to improve the high frequency information in reconstructed frames. Reconstruction is performed or 8X/6X and 6X/4X accelerations respectively. The reconstructed images are fed into Tofts model ² to obtain the parametric maps.

Data: All experiments were performed using breast DCE data sets downloaded from Quantitative Imaging Network (QIN) ³ with parameters TR=6.2 ms, TE=2.9ms, temporal resolution 18~20 seconds. Number of time points varied from 28~32 frames in each data sets. The Contrast Agent (CA) used was Gd (HP-DO3A) [ProHance] IV with a dosage of 0.1mmol/Kg at 2 ml/s. The reconstruction of this in-vivo data was performed by taking combination of different acceleration factors in TR and SR regions respectively.

The proposed method shows better reconstruction results as in figure 2. From figure 2, it can be inferred that the image obtained by including view sharing (b) is better in terms of blurring than that of the one with only CS (a). Thus, the acquisition is accelerated and the PK maps obtained for the tumor region shown in figure 3 at these acceleration factors using previous method (ASTRA) ¹ and current method (RASTRA) are as shown in figure 4. From figure 4 (Ktrans values), we can observe that the reconstructed images obtained using RASTRA are much closer to 1X maps compared to those of ASTRA ¹.Sampling method used in the proposed method RASTRA is based on mapping which does not lead to temporal blurring on reconstruction as the injection of CA affects change only in the low frequency data and not affecting the high frequency information. RASTRA was able to reconstruct the images at higher acceleration factors with high fidelity. On including view sharing, we could obtain better edge information in reconstructed frames thus improving the image quality and subsequently in the PK maps. The NRMSE value obtained using RASTRA is less than that obtained using ASTRA as shown in figure 5 for all acceleration factors.