Image-guided Tissue Validation of Combined Preload Dosing and Mathematical Modeling Correction of Perfusion MRI Measures

1.0.454 (Imaging Biometrics, LLC, Wisconsin). This enabled rCBV calculation either without or with modeling based on a previously reported algorithm. 5 Based on different acquisition and post-processing variables, we created three distinct experimental methods to calculate rCBV: A) Modeling without PLD (to assess modeling correction of T1 leakage); B) PLD without modeling (to assess only the effects of PLD without T2/T2* residual correction); and C) Combined Modeling with PLD (to assess modeling correction of T2/T2* residual effects). We created Receive Operator Characteristic (ROC) curves for each experimental group to determine rCBV accuracy to distinguish tumor from PTRE. We calculated Areas under the curve (AUCs) for each group's ROC and statistically compared them using the Delong Clarke-Pearson method (p < 0.05). Sensitivity, specificity, and 95% confidence intervals (CIs) for distinguishing PTRE and tumor were generated from each ROC curve at a number of rCBV cutoff points to determine the optimal threshold value that maximized accuracy (defined as the average of sensitivity and specificity). We also calculated Pearson correlations between rCBV and tissue microvessel number for each group (p<0.05). 6 We calculated total microvessel number on CD-34 stained slides and normalized to the total slide specimen area (μm 2 ), using Axiovision Automeasure 3.4 software module (Zeiss, Germany). A biostatistician performed all analyses. A neuropathologist diagnosed specimens as tumor or PTRE. 1 Results: We included 36 tissue specimens (from 11 subjects) and categorized each specimen as tumor (n=21) or PTRE (n=15). Microvascular analysis was available in 16 of these samples, which included both tumor (n=7) and PTRE (n=9) categories. We summarize AUC and Pearson correlations in the table below. Combined preload dose (PLD) and modeling (group C) provided the highest AUC (0.97), which was significantly higher than AUC in the absence of PLD (group A, p=0.01) or modeling (group B, p=0.04) (figure). Using combined PLD and modeling, the rCBV threshold of 1.07 maximized diagnosis of tumor and PTRE with 95.2% accuracy (95%CI = 73.9%-99.4%), 90.5% sensitivity, and 100% specificity. Combined PLD and modeling rCBV significantly correlated with microvessel number (r=0.524), whereas the other conditions did not. Conclusion: Combined PLD and modeling correction maximizes rCBV correlation with tissue analysis, compared with either condition alone. Modeling provides similar T2/T2*W correction as previously reported BLS 2 , but in a more automated and efficient manner, suggesting the potential utility of this combined method in clinical practice and multi-institutional trials.