SU‐FF‐T‐388: Comparison of R2 Estimation Algorithms for Polymer Gel Dosimetry

Purpose: To evaluate R2‐estimation algorithms applied for MRI‐based polymer gel dosimetry. Method and Materials: We evaluated weighted least‐squared (WLS), least‐squared (LS), and maximum‐likelihood estimation (MLE) methods. For MLE, we tested both Gaussian and Rician probability distributions (MLE_G and MLE_R). To increase the accuracy of the R2‐estimation, we proposed a variable echo number technique (VAREC), in which the number of echo signals used for the estimation was optimized depending on the expected R2‐value. The VAREC method was used with WLS and MLE_G. These algorithms were used to estimate R2 values of BANG polymer gel, which was irradiated for 12‐different dose levels ranging from 0 to 50 Gy. The R2 measurements were done by using a 32‐echo CPMG pulse sequence as implemented in Siemens Trio 3T MRI scanners. Estimated R2 values, 〈R2〉, were plotted as a function of dose for the four methods, which were compared with a reference. Results: With WLS and MLE_G, 〈R2〉 increased up to an absorbed dose of about 6 to 10 Gy; however, above these doses, 〈R2〉 decreased with increasing dose. This was a manifestation of an algorithmic error of those methods. When the VAREC technique was used with WLS and MLE_G, the erroneous behavior of the 〈R2〉 and dose relationship disappeared. Bothe LS and MLE_R did not show such an error; but, 〈R2〉 values for higher doses were overestimated. The computing times of LS and MLE_R were 10 and 60 times longer than MLE_G, respectively. Conclusion: WLS and MLE_G lead to equally reliable results in a reasonable computing time. Since MLE is known to be more accurate for noisy data than WLS, we strongly recommend the MLE algorithm with the VAREC technique for R2‐estimation to the MRI‐based polymer gel community.