Automated Distinction of Intimal and Medial Intracranial Arterial Calcification from CT Head

Benjamin Jin; Maria del C. Valdés Hernández; Richard Bortsov; Joanna M. Wardlaw; Daniel Bos; Grant Mair

eess.IVSep 11, 2026

Automated Distinction of Intimal and Medial Intracranial Arterial Calcification from CT Head

Authors: Benjamin Jin, Maria del C. Valdés Hernández, Richard Bortsov, Joanna M. Wardlaw, Daniel Bos, Grant Mair

Organizations: Institute for Neuroscience and Cardiovascular Research, University of Edinburgh, United Kingdom · Department of Epidemiology, Erasmus MC, The Netherlands · UK Dementia Research Institute, University of Edinburgh, United Kingdom · Department of Radiology and Nuclear Medicine, Erasmus MC, The Netherlands · Department of Clinical Neurosciences, NHS Lothian, United Kingdom

Abstract

Intracranial arterial calcifications (IACs) are a common finding on clinical non-contrast enhanced head CT scans and are associated with neurovascular disease. Calcifications can occur in the intimal or medial layer of the arterial wall, subtypes that differ in aetiology and may have distinct clinical relevance. These subtypes can be visually distinguished by radiologists based on the shape of the calcifications. We investigate three automated approaches for subtype classification of IAC from head CT-derived segmentation masks: (1) an automated adaptation of the established radiological visual score, (2) a sphericity-based method, and (3) a method based on shape embeddings extracted by a medical shape foundation model. All approaches use the same lightweight classification pipeline on top of the features they compute and are evaluated using 5-fold cross-validation. The three methods achieved comparable performance, with the embedding-based approach yielding the best overall results with a weighted F1 (mean $\pm$ SD) of up to 71.5 $\pm$ 3.7 for a single artery and 59.8 $\pm$ 1.7 for the joint artery classification. Performance was largely preserved when using automated instead of manual IAC segmentation masks, and we found the difference in weighted F1 not significant. Our results show that fully automated IAC subtype quantification from head CT is feasible and remains robust to the use of manual and automated IAC segmentation masks. Code at https://github.com/bjin96/iac-subtyping.

Figures & tables

Figure 1: Intracranial arterial calcification subtypes with examples of the calcification from the cavernous part of the internal carotid artery as it appears on CT.

Figure 2: Three methods generating features for distinguishing IAC subtypes. Spheres for sphericity calculation overlayed in red over the IAC objects. $N$ is the number of examples, $P$ is the respective number of points derived from the segmentation masks.

		Macro Avg. $\pm$ SD			Weighted Avg. $\pm$ SD
Method	Side	F1	Prec.	Recall	F1	Prec.	Recall	$\mathrm{\textbf{F1}}^{\mathrm{\textbf{A}}}$
Autom. Visual	L	66.5 $\pm$ 2.7	67.7 $\pm$ 2.3	68.8 $\pm$ 2.5	67.3 $\pm$ 2.7	71.1 $\pm$ 2.2	66.8 $\pm$ 2.7	65.3 $\pm$ 1.9
	R	68.4 $\pm$ 2.1	69.5 $\pm$ 2.1	70.3 $\pm$ 2.2	68.9 $\pm$ 2.1	72.0 $\pm$ 2.2	68.6 $\pm$ 2.1	66.2 $\pm$ 1.9
	T	49.9 $\pm$ 1.8	51.3 $\pm$ 1.6	51.1 $\pm$ 2.1	58.4 $\pm$ 1.8	62.7 $\pm$ 1.5	56.7 $\pm$ 2.0	57.5 $\pm$ 2.5
Spher.	L	67.3 $\pm$ 1.2	67.5 $\pm$ 1.1	68.5 $\pm$ 1.2	68.5 $\pm$ 1.3	70.4 $\pm$ 1.1	68.0 $\pm$ 1.3	69.0 $\pm$ 5.1
	R	67.5 $\pm$ 3.0	67.5 $\pm$ 3.0	68.1 $\pm$ 3.2	68.6 $\pm$ 3.0	69.5 $\pm$ 3.0	68.3 $\pm$ 3.0	69.1 $\pm$ 2.6

Table 1: IAC subtype classification cross-validation metrics for the three proposed methods, mainly from manual segmentations. Macro Avg. is the unweighted mean of per-class metrics, and Weighted Avg. is weighted by per-class sample count. We provide separate results for the left (L) and the right (R) artery, and joint total (T) results. Weighted F1 using automatically generated segmentations in column

\mathrm{\textbf{F1}}^{\mathrm{\textbf{A}}}

		Left		Right		Total
Method Pair		$\kappa$	Agree(%)	$\kappa$	Agree(%)	$\kappa$	Agree(%)
Autom. Visual	Sphericity	0.58	79.1%	0.56	77.8%	0.46	64.3%
Autom. Visual	Deep Learning	0.38	68.8%	0.42	70.7%	0.32	56.0%
Sphericity	Deep Learning	0.39	69.6%	0.43	72.1%	0.32	55.8%

Table 2: Cohen’s

\kappa

and raw agreement percentages for the pairwise agreement between the proposed Automated Visual, Sphericity, and Deep Learning methods.

Figure 3: Correlation between Automated Visual features and mean sphericity.

Figure 4: Confusion matrices for artery-wise classification of IAC subtype using our Automated Visual, Sphericity, and Deep Learning methods.

Figure 5: Deciles summarising the scan-level mean sphericity for each predominant subtype. The additional None category for automated segmentation masks includes false-positive segmentations and calcifications missed by human annotators.

Explore similar work

ArteryX: A Reliable End-to-End Toolbox for Standardized Intracranial Artery Feature Extraction from 3D TOF-MRA
Jul 10, 2025Abrar Faiyaz, Nhat Hoang, Giovanni Schifitto +1Medical Image Analysis Magnetic Resonance Imaging
Department of Neurology, University of Rochester, Rochester, NY 14642, USA · Department of Physics, University of Rochester, Rochester, NY 14627, USA · Department of Imaging Sciences, University of Rochester, Rochester, NY 14642, USA +2
Shape Over Intensity: Directional Topological Encoding for False Positive Reduction in Intracranial Aneurysm Detection
Jul 6, 2026Akshay Gokhale, Mansi DhamneFDR Control Medical Image Anomaly Detection
Computer Science and Engineering Sardar Patel Institute of Technology Mumbai, India · Computer Engineering Sardar Patel Institute of Technology Mumbai, India
DINOv3 with Test-Time Calibration for Automated Carotid Intima-Media Thickness Measurement on CUBS v1
Mar 11, 2026Zhenpeng Zhang, Jinwei Lu, Yurui Dong +1Medical Imaging Ultrasound Image Segmentation
School of Basic Medical Sciences, Fudan University, Shanghai 200032, China · Fudan University, Shanghai, China · Institute of Medical Genetics and Genomics, Fudan University, Shanghai 200032, China

Jul 10, 2025J/K move · Enter open · S save

ArteryX: A Reliable End-to-End Toolbox for Standardized Intracranial Artery Feature Extraction from 3D TOF-MRA

Abrar Faiyaz, Nhat Hoang, Giovanni Schifitto, Md Nasir Uddin

Department of Neurology, University of Rochester, Rochester, NY 14642, USA · Department of Physics, University of Rochester, Rochester, NY 14627, USA · Department of Imaging Sciences, University of Rochester, Rochester, NY 14642, USA+2

Medical Image Analysis Magnetic Resonance Imaging

Open paper AskPDF arXiv

Cerebrovascular research heavily relies on quantitative analysis of intracranial arteries from time-of-flight magnetic resonance angiography, yet existing processing pipelines remain limited by inconsistent artery labeling and a high manual correction burden. We present ArteryX, a toolbox for extracting features that standardizes artery classification across proximal and distal vascular territories. It integrates segmentation handling, isotropic processing, vessel-fused graph construction, and constrained landmark-based classification within a unified artery-specific feature reporting and reproducible workflow. The toolbox extracts morphological, topological, and complexity features including total length, mean radius, volume, surface area, branch count, tortuosity, and fractal dimensionality for standardized artery-segments. Test-and-validation were performed using three complementary datasets: (1)TopBrain-Challenge benchmarking with annotated arteries, (2)synthetic known-reference validation, and (3)exploratory in-vivo cohort of cerebral small vessel disease. In TopBrain analyses, ArteryX with supervised nnUnet segmentation showed minimal bias, while iCafe showed the highest bias and a large limit-of-agreement. ArteryX consistently demonstrated robust downstream quantification performance across segmentation sources (unsupervised/supervised). Agreement analyses showed minimal bias for radius and good sensitivity of extent-dependent metrics throughout the noisier segmentations compared to the state-of-the-art iCafe-toolbox. Furthermore, a stage-wise human-in-the-loop protocol showed lower intervention time than iCafe. In an in-vivo-cohort (48CSVD+, 20CSVD-), ArteryX-derived distal and territory-level features showed group-level differences, not evident with iCafe. To facilitate adoption-and-reproducibility, ArteryX is designed with versioned builds, tutorials, and documentation.

Abstract

Explore similar work

ArteryX: A Reliable End-to-End Toolbox for Standardized Intracranial Artery Feature Extraction from 3D TOF-MRA

Shape Over Intensity: Directional Topological Encoding for False Positive Reduction in Intracranial Aneurysm Detection

DINOv3 with Test-Time Calibration for Automated Carotid Intima-Media Thickness Measurement on CUBS v1