Precision and accuracy of K estimated by fitting the extended Kety model parameters to DCE-MRI time course data is unaffected by the choice of optimisation algorithm or estimation of T1 using linearisation

Introduction Microvascular characteristics of tumours can be assessed by fitting a tracer kinetic model to contrast agent concentration time course data derived from dynamic contrast-enhanced (DCE-) MRI time series. The estimated model parameters, such as K, can be used to monitor the efficacy of anti-angiogenic drugs. Many optimisation algorithms are available for model fitting and it is important to ensure that the selected implementation gives precise and accurate results. We present a comparison of optimisation algorithms used to fit the extended Kety model to contrast agent concentration time course data. The parameter values could also be affected by the accuracy of the T1 value (which is required to convert signal intensity to contrast agent concentration). Therefore, we also assess algorithms (including function linearisation) used for estimating T1 from the commonly used variable flip angle (VFA) spoiled gradient echo (SPGR) method. Synthetic data We used a software phantom generator to provide VFA SPGR images and DCE time-series with known ground truth. Pre-contrast SPGR images for T1 estimation had flip angles of 2°, 10° and 30° and a TR of 4 ms. DCE time-series SPGR images were simulated using the extended Kety model and a functional form of a population AIF. K, ve and vp were varied along the 3 axes of the synthetic image volume to give 400 different parameter combinations. 20 evenly-spaced values of each parameter were used with ranges of 0.1 – 0.5 min for K; 0.15 – 0.6 for ve; and 0.025 – 0.1 for vp. Three synthetic data sets with ground truth T1 values of 600 ms, 1000 ms and 1400 ms were generated. For assessing T1 estimation, VFA SPGR images were produced with 20 evenly-spaced T1 values from 100 ms to 2000 ms. 400 samples were generated for each T1 value. Zero mean Gaussian noise was added to the signal intensity data to give a signal-to-noise ratio equivalent to 10 in a 30° flip angle pre-contrast SPGR image.