Article Figures & Data
Figures
- Fig 1.
A–D, Scatterplots correlating rCBV metrics with and without preload dosing (PLD), as measured by 2 separate modeling algorithms (IBN, NICE without γ variate fitting). PLD- and non-PLD corrected values are shown in the x- and y-axes, respectively. Overall, IBN measurements demonstrate consistently higher Pearson (r) and Spearman (ρ) correlations for mean rCBV, mode rCBV, fractional tumor burden (FTB), and percentage of voxels > 1.75. The thresholding metrics (FTB, percentage > 1.75) correlate most strongly between PLD- and non-PLD-corrected conditions.
- Fig 2.
Image of a representative case in a 39-year-old patient with recurrent high-grade ganglioglioma shows an enhancing mass (A). Color overlay percentage > 1.75 thresholding maps (B–E) depict orange voxels with high rCBV > 1.75, compared with intermediate yellow voxels (rCBV, 1.0–1.75) and low green voxels (rCBV < 1.0). With NICE, both spatial distribution and percentage of orange voxels show high discrepancy between non-PLD- (70%, B) and PLD-corrected (35%, C) maps. With IBN, the percentage of orange voxels on the non-PLD map (54%, D) approximates that on the PLD-corrected map (51%, E) with high spatial congruence.
- Fig 3.
Receiver operating characteristic curves for fractional tumor burden to predict histopathology (tumor versus posttreatment effect) in patients with recurrent glioblastoma multiforme (n = 25). FTB by IB Neuro (blue) demonstrates a significantly larger area under the curve (AUC) compared with nordicICE (without γ variate fitting, orange) FTB measurements (0.85 versus 0.70, P < .01), suggesting that different modeling algorithms can impact the accuracy in predicting histopathologic diagnosis. Adding γ variate fitting further reduces NICE estimates of FTB (0.67, green).
Tables
- Table 1:
Pearson (r) and Spearman (ρ) correlations between rCBV metrics under PLD-corrected and non-PLD-corrected conditions, as measured by IBN and NICE perfusion software algorithmsa
rCBV Metric Non-PLD vs PLD (IBN) P Value Non-PLD vs PLD (NICE) with gvf P Value Non-PLD vs PLD (NICE) without gvf P Value Mean r = 0.87 <.001 r = 0.11 .54 r = 0.43 .01 ρ = 0.86 <.001 ρ = 0.42 .02 ρ = 0.62 <.001 Mode r = 0.78 <.001 r = 0.44 .01 r = 0.51 .01 ρ = 0.76 <.001 ρ = 0.65 <.001 ρ = 0.65 <.001 % < 1.75 r = 0.93 <.001 r = 0.55 <.001 r = 0.59 <.001 ρ = 0.91 <.001 ρ = 0.61 <.001 ρ = 0.60 <.001 FTB r = 0.96 <.001 r = 0.79 <.001 r = 0.70 <.001 ρ = 0.94 <.001 ρ = 0.72 <.001 ρ = 0.71 <.001 Note:—gvf indicates γ variate fitting.
↵a NICE calculations were performed with and without gvf. IBN modeling shows substantially higher correlation between PLD and non-PLD metrics (compared with NICE), suggesting higher rCBV accuracy in the absence of PLD correction. Statistical significance is P value < .05.
- Table 2:
Correlations between rCBV and fractional MVA under different PLD and modeling conditionsa
Conditions for rCBV Measurement Pearson Correlation (r) P Value Spearman Correlation (ρ) P Value Fractional MVA 1.00 – 1.00 – No PLD (IBN) 0.46 .02 0.33 .12 No PLD (NICE + gvf) 0.51 .01 0.26 .19 No PLD (NICE − gvf) 0.35 .10 0.18 .39 PLD (IBN) 0.64 <.001 0.58 .001 PLD (NICE + gvf) 0.53 <.01 0.28 .15 PLD (NICE − gvf) 0.59 .001 0.40 .04 Note:—+ indicates with; −, without; –, not applicable; gvf, γ variate fitting.
↵a Both PLD correction and IBN software modeling were needed to achieve maximal correlation.
Jump to section
Related Articles
Cited By...
- Arterial Spin-Labeling and DSC Perfusion Metrics Improve Agreement in Neuroradiologists Clinical Interpretations of Posttreatment High-Grade Glioma Surveillance MR Imaging--An Institutional Experience
- Uncertainty Quantification in Radiogenomics: EGFR Amplification in Glioblastoma
- Machine learning assisted DSC-MRI radiomics as a tool for glioma classification by grade and mutation status
- Performance of Standardized Relative CBV for Quantifying Regional Histologic Tumor Burden in Recurrent High-Grade Glioma: Comparison against Normalized Relative CBV Using Image-Localized Stereotactic Biopsies
- Perfusion MRI-Based Fractional Tumor Burden Differentiates between Tumor and Treatment Effect in Recurrent Glioblastomas and Informs Clinical Decision-Making
- Accurate Patient-Specific Machine Learning Models of Glioblastoma Invasion Using Transfer Learning
- Multisite Concordance of DSC-MRI Analysis for Brain Tumors: Results of a National Cancer Institute Quantitative Imaging Network Collaborative Project
- Discrimination between Glioma Grades II and III Using Dynamic Susceptibility Perfusion MRI: A Meta-Analysis