This repository contains trained nnU-Net model weights used in our baseline algorithm for centerline and morphometric feature extraction of the Circle of Willis (CoW). The model was trained on multi-class segmentation masks from the publicly available TopCoW dataset, which includes CTA and MRA scans of 125 stroke patients, and was designed to predict binary centerline skeletons. The ground truth skeletons used for training were obtained from our public centerline dataset. The baseline pipeline combines nnU-Net-based skeletonization with A* graph connection and was developed to support anatomically accurate centerline extraction and robust feature computation. These weights are intended for reproducibility and benchmarking of the method described in our manuscript: Musio et al., “Circle of Willis Centerline Graphs: A Dataset and Baseline Algorithm” (2026) The full pipeline and code are available on GitHub: https://github.com/fmusio/CoW_Centerline_Extraction This pipeline is designed to be used on top of the TopCoW segmentation models, which segment CTA or MRA images into 13 anatomical vessel segments. After segmenting the angiographic images using the TopCoW models, the baseline algorithm can be applied to extract centerlines and compute morphometric features taking the multi-class mask as input. Usage After downloading and unzipping the weights, place the nnUNet_results folder in the following location within the cloned GitHub repository: CoW_Centerline_Extraction/predict_skeleton/ This folder should contain the nnU-Net model weights in the format expected by the nnUNet_predict command.