Program Information
A Constrained Linear Reference Region Model For DCE-MRI
Z Ahmed1*, I El Naqa1 , I R. Levesque1,2 , (1) McGill University, Montreal, QC (2) Research Institute of the McGill University Health Centre, Montreal, QC
Presentations
TH-CD-207-3 (Thursday, July 16, 2015) 10:00 AM - 12:00 PM Room: 207
Purpose: Dynamic contrast enhanced (DCE)-MRI characterizes perfusion in tissues using time-series data and a quantitative model. This study presents a modification to the linear reference region model (LRRM) that uses two fitting parameters instead of three, resulting in improved accuracy and precision of estimated parameters.
Methods: The number of fitting parameters in the LRRM is reduced by noting that the 1st and 2nd parameters are related by the kep of the reference region. This relation is accounted by the proposed constrained LRRM (CLRRM). The performance of CLRRM was compared against LRRM through simulations of concentration-time curves produced using the Tofts model, with KTrans=0.1, v_e=0.1 for reference tissue, and KTrans=0.25, v_e=0.4 for tumour. SNR in reference curve is generally high, therefore Gaussian noise was added only to the tumour curves for an SNR range of 5 to 50 in 10 steps with temporal resolutions of 1s, 10s, 30s and 60s. 10,000 realizations were simulated for each SNR and temporal resolution combination. The in-vivo performance of both methods was evaluated by applying them to data acquired in sarcoma.
Results: The accuracy was evaluated through the mean percent error, and precision was judged by the standard deviation of the percent error. Both CLRRM and LRRM had similar accuracy at high SNR, but CLRRM was more accurate at lower SNR, especially for kep. The mean percent error for kep was -36% for LRRM and -6% for CLRRM at SNR 5 and 30s temporal resolution. The standard deviation for CLRRM was roughly 5 times lower. The in-vivo results showed that CLRRM produced kep maps with less noise and higher values in the solid tumour region.
Conclusion: A modified version of the linear reference region model improves quantitative DCE-MRI parameter estimates at low SNR and low temporal resolution.
Funding Support, Disclosures, and Conflict of Interest: Funding from the RI-MUHC (Montreal General Hospital Foundation), NSERC CREATE MPRTN (Grant no. 432290), and McGill University Faculty of Medicine Fellowships.
Contact Email: