MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Predicting Meningioma Tumor Recurrence Using Radiogenomics Data

T Coroller*, W Bi , N Greenwald , R Zeleznik , E Huynh , C Parmar , R Huang , H Aerts , Dana-Farber Cancer Institute, Brigham & Women's Hospital, Harvard Medical School, Boston, MA

Presentations

(Thursday, 8/2/2018) 7:30 AM - 9:30 AM

Room: Davidson Ballroom B

Purpose: Chromosome alteration burden has been linked to tumor malignancy among meningiomas, where increasing chromosomal disruption is associated with an increased risk of recurrence. In this study, we investigated the ability of radiographic features to distinguish chromosome burden. Capturing such genetic burden non-invasively could help improve our understanding of tumor mechanisms. In this study we investigate the use of radiogenomics data to predict tumor recurrence.

Methods: 250 meningioma patients with pre-operative contrast-enhanced T1 MRI array comparative genomic hybridization (aCGH) were included in this study. Predictors included 102 unfiltered radiomic features and 40 chromosome copy number variation (CNV). The Chromosome alteration score (CAS) was computed by using +/-0.2 thresholds on the CNV values and summing the events across all chromosomes for a given patient. Spearman’s correlation coefficient was used to assess univariate radiogenomics associations and all p-values were corrected for multiple testing hypothesis. Tumor recurrence was predicted using LASSO classifier and cross-validation (k=100) and significance between models using permutation test.

Results: We found that 18.1% of the features were overall significantly correlated with CNV value. This number was up to 49% for CAS. When looking at tumor recurrence, we shown that radiomics alone (median AUC = 0.55) performed poorly compare to genomics alone (median AUC = 0.70, permutation p-value = 0.01). The radiogenomics model (combination of radiomics and genomics information) had higher performance (median AUC = 0.72) than genomics, but did not significantly performed better than it (permutation p-value =0.28)

Conclusion: In this study we have shown a link between meningioma tumor phenotype and genotype. This could potentially enable future research focusing on the impact of biology to tumor characteristics. Combining these information resulted in an improvement in prediction performance in tumor recurrence than radiomics or genomics alone.

Keywords

MRI, Statistical Analysis, Brain

Taxonomy

IM- MRI : Biomarkers

Contact Email