Abstract Body: Introduction: Intracranial aneurysm (IA), a cerebrovascular disease affecting 3-5% of US population, leads to subarachnoid hemorrage. A non-invasive diagnostic to identify those with an IA could facilitate more widespread screening and better disease management.

Objective: We have designed a PCR-based diagnostic assay for IA, AneuScreen^TM, that is comprised of a previously-published 50-gene panel. This project aimed to further develop the assay’s predictive model in a large cohort and its predictive power in independent cases.

Methods: Before testing, the assay underwent rigorous standardization to test signal strength, dynamic range and linearity, and variation across replicates. Blood from consenting individuals with/without IA (angiogram confirmed) were collected under IRB approval from University at Buffalo, University of Pennsylvania, and University of South Florida. One mL of blood was collected into DxTerity RNA-stabilizing blood tube. After running each sample on the assay and standardizing data, gene expression levels of all 50 genes (as well as gene ratios and patient data) were used for support vector machine (SVM) model development in 85% of the data. Top 50 features selected by statistical filtering and ranking were used in nested loops to maximize training performance. Optimal model was validated in remaining 15% of samples.

Results: We enrolled 497 individuals (276 had IAs). In training, 12 features were most important. The final SVM model performed well in testing set without showing evidence of overfitting, achieving an AUC of 0.74. In independent validation set, the model reached an AUC of 0.80 (Fig. 1A and B).

Conclusion: SVM model developed using expression data from AneuScreen^TM in a large cohort of individuals with and without IA was able to predict IA cases with an AUC of 0.80. This demonstrates the potential of using circulating blood to screen for IA presence. Additional testing is required to finetune model and improve accuracy.