Collateral automation for triage in stroke treatment correlated with multiphase computed tomography scans in acute stroke

Objective: The aim of our study was to evaluate the interobserver agreement level of the collateral score obtained using multiphasic computed tomography angiography (mCTA) in patients who underwent mechanical thrombectomy due to MCA M1 occlusion, and to investigate the prediction of this scoring and functional outcome using radiomics-based semi-automated analysis results. Materials and Methods: A total of 96 patients who underwent mechanical thrombectomy for MCA M1 occlusion between 2021 and 2024 and had mCTA examinations were included in the study. Demographic and clinical data of the patients were obtained from the hospital information management systems. The mCTA images of the patients were retrospectively reviewed, and collateral scoring was performed independently by two radiologists. The mCTA images were segmented using the "3D Slicer" software, and radiomics data were extracted. Using the obtained data, three different artificial neural network (ANN) models were developed to predict collateral scores and functional outcomes. Results: All three ANN models demonstrated high accuracy; however, Model 3, which underwent cross-validation, showed the highest external validity despite a relatively lower AUC value. Correlation analysis revealed no significant strong relationship between clinical variables and radiomics parameters. Most radiomics features were found to be independent of each other. In the correlation analysis between the Menon score and the examined radiomics parameters, a weak-to-moderate negative correlation was observed between MeanAbsoluteDeviation and LargeAreaHighGrayLevelEmphasis (r = –0.28). No notable correlations were found among the other variables. Interobserver agreement in collateral scoring was found to be low. Conclusion: Our study demonstrates that ANN models can be utilized to predict collateral scoring and functional outcomes. Although such models are not expected to replace final decision-making mechanisms in the short term, they are thought to contribute to standardized evaluation by reducing subjectivity in collateral scoring. Moreover, they offer the potential for rapid and precise assessment in time-critical situations. KEYWORDS: Multiphasic CTA, Collateral Scoring, Radiomics, Artificial Neural Network