Random Forest model classifies blood vessel tortuosity with 92% F1 score

A team from multiple institutions—Yu Zhong, Jingzhi Guo, Luyao Li, and co-authors—published a paper on arXiv (2607.14195) presenting a fully automated framework for classifying blood vessel morphology, specifically targeting the internal carotid artery (ICA-C1) segment. Traditional subjective visual grading of tortuosity relies heavily on clinician experience, while simple distance-based indices fail to capture 3D spatial deformation. To address this, the framework extracts 13 tortuosity features from 379 clinical vascular centerlines using discrete geometric methods, then applies Information Gain to select the six most discriminative features ($\mathcal{TI}$, $\mathcal{AC}$, $\mathcal{TC}$, $\mathcal{AC}/\mathcal{AT}$, $\mathcal{AT}$, $\mathcal{TT}$). A Random Forest classifier is trained on these features.

The model was evaluated on two tasks: binary classification (non-severe vs. severe tortuosity) achieved a Macro-F1 score of 0.9206, while ternary classification (straight, low-tortuosity, high-tortuosity) reached 0.8626. Elongation and curvature features proved most informative for basic screening, while torsion-related features improved finer-grained grading. Beyond classification, the team derived a Morphological Risk Index (MRI) from Random Forest feature importances, offering a direct numerical reference for vascular morphology. This could enable more objective, consistent stroke-risk assessment in clinical settings, reducing reliance on subjective expert judgment.