Automated Arterial Input Function Extraction for T1-Weighted DCE-MRI

Abstract

2 , consisting of 3-D gradient echo acquisitions acquired over a period of approximately 3.5 minutes, with a temporal resolution of 5-8.5 s was used. Baseline T1 was calculated using 3 acquisitions with different flip angles. Concentration of contrast agent was estimated from the change in T1 during the dynamic series. An intravenous contrast agent bolus of less than 10 s duration was manually injected. Acquisitions in the lung and prostate utilised similar protocols, but with 4 s and 2.3 s temporal resolution, respectively, and with the bolus administered using a power injector. AIF estimation: 1. Manually identify a slice that contains a suitable artery, and does not suffer in-flow effects (note that the use of 3- D acquisition allows significantly reduced in-flow effects compared with multi-slice techniques when a slice in the centre of the volume is selected); 2. Extract the time course for every voxel in slice and identify those that show a maximum in concentration within 20 s of the arrival time (early max voxels); 3. Identify the early max voxels within the top 5% of peak concentrations; 4. Average the resultant time courses together to create a mean AIF. Results and Conclusion The AIF identification method proved to extract plausible AIFs from a range of regions of the body, including the lung and prostate (Fig. 1). The positions of the voxels that contribute to the AIF may be determined and mapped onto an anatomical image, showing that the mean AIF is extracted from voxels positioned in arteries. The selection of the top 5 % of early max voxels ensures that partial volume effects are minimised. Figure 2 shows the results of the technique in a glioma patient scanned twice, showing qualitatively good reproducibility. Figure 3 shows measures of the repeatability of the method across all 6 glioma patients. The method provides an AIF estimate with a minimum of user input and is applicable to any area of the body with a suitable artery. The method shows good scan-rescan reproducibility, even in the case shown using manual bolus administration. References