Development and validation of a model for predicting the risk of brain arteriovenous malformation rupture based on three-dimensional morphological features.

Abstract

Brain arteriovenous malformation (bAVM) is an important reason for intracranial hemorrhage. This study aimed at developing and validating a model for predicting bAVMs rupture by using three-dimensional (3D) morphological features extracted from Computed Tomography (CT) angiography. The prediction model was developed in a cohort consisting of 412 patients with bAVM between January 2010 and December 2020. All cases were partitioned into training and testing sets in the ratio of 7:3. Features were extracted from the 3D model built on CT angiography. Logistic regression was used to develop the model, with features selected using L1 Regularization, presented with a nomogram, and assessed with calibration curve, receiver operating characteristic (ROC) curve and decision curve analyze (DCA). Significant variations in associated aneurysm, deep located, number of draining veins, type of venous drainage, deep drainage, drainage vein entrance diameter (Dv), type of feeding arteries, middle cerebral artery feeding, volume, Feret diameter, surface area, roundness, elongation, mean density (HU), and median density (HU) were found by univariate analysis (p < 0.05). The prediction model consisted of associated aneurysm, deep located, number of draining veins, deep drainage, Dv, volume, Feret diameter, surface area, mean density, and median density. The model showed good discrimination, with a C-index of 0.873 (95% CI, 0.791-0.931) in the training set and 0.754 (95% CI, 0.710-0.795) in the testing set. This study presented 3D morphological features could be conveniently used to predict hemorrhage from unruptured bAVMs.