(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose:
To propose a novel strategy for the characterization of dose maps (DMs) properties that impact on significance maps from Voxel-based analyses (VBAs) evaluating local dose-response patterns via voxelwise statistical analysis on spatially normalized DMs.
Methods:
DMs of 178 lung cancer patients, treated with IMRT or protons, were normalized on XCAT digital phantom [Segars 2010]. We analyzed:
1. the uniformity of voxelwise mean (µ) and standard deviation (s) of DMs over patients, which determines the homogeneity of VBA statistical power;
2. the probabilistic independent component analysis (PICA) [Beckmann & Smith 2005], blindly inferring the number of statistically-significant independent maps (model order) that generate the DMs;
3. the connectogram [Irimia 2012], linking pairs of substructures by Spearman correlations (Rs) between their mean doses. We analyzed the cardiac substructures included in XCAT and the lung subregions segmented by a radiologist.
Points 2-3 elucidate the spatial resolution of the significance map from VBA for a given effect.
Results:
The contrast over the 80% of the analyzed volume was 0.8 for µ-map and 0.5 for s-map. PICA detected 43 dose clusters homogenously spread across the thorax. Connectograms showed that, while doses to main structures (cardiac chambers and lung lobes) were weakly correlated (Rs²<0.2), Rs² between adjacent lobe segments or chambers and related walls can reach 0.8.
Conclusion:
The homogeneity of the s map and the spread of PICA clusters suggests a uniform power of possible VBAs on the dataset.
PICA order, comparable with the cohort size, hints that a large number of DMs contributes to split the analyzed volume into independent patches that could highlight via VBAs distinct dose-response correlations.
Connectograms showed that the dataset can barely supports a radiobiological differentiation between the tiniest substructures.
The proposed characterization should be ancillary to any dosimetric VBA for a clear insight on the inference limits.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
