Age-related changes of the brain’s arterial network assessed with machine learning segmentation

Age-related changes of the brain’s arterial network assessed with machine learning segmentation | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Age-related changes of the brain’s arterial network assessed with machine learning segmentation View ORCID Profile Jorge T. Gomez , Rachit Saluja , Mert R. Sabuncu , Henning U. Voss doi: https://doi.org/10.1101/2025.09.10.25335518 Jorge T. Gomez a Cornell Tech, New York, NY10044, USA, and Cornell University , Ithaca, NY14853, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jorge T. Gomez Rachit Saluja a Cornell Tech, New York, NY10044, USA, and Cornell University , Ithaca, NY14853, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Mert R. Sabuncu a Cornell Tech, New York, NY10044, USA, and Cornell University , Ithaca, NY14853, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Henning U. Voss b Cornell Magnetic Resonance Imaging Facility, Cornell University , Ithaca, NY14853, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: hv28{at}cornell.edu Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Purpose To provide a tool for the automatic segmentation of an arteriogram of the brain from MRA images and the estimation of arterial tortuosity as a summary marker. Methods A deep learning model was trained and validated on a previously published set of semi-automatically segmented brain arteriograms. We tested whether arterial tortuosity estimated from a large number of age-representative subjects (N = 478) would reproduce previously published statistics of increasing tortuosity with age. Results The tool provides a segmentation of the arteriograms from MRA images of varying resolution and quality. The arterial tortuosity estimated from the automatically segmented brain arteriograms approximately matched their previously published statistics. Further, a highly significant increase of tortuosity with age was observed in the large dataset with 478 subjects (p = 9 ×10 -8 ). Discussion and conclusion The proposed ASN (Angiogram Segmentation Network) algorithm can provide the radiologist with a clean arteriogram of the brain, without the need for manual segmentation by the MR operator. Moreover, it offers arterial tortuosity as an instant quantitative metric that can augment the qualitative visual reading of the MRA. 1. Introduction Tortuosity, defined as the degree of curving, curling, looping, winding, or kinking of blood vessels 1 , 2 , is increasingly recognized as a morphological hallmark of cerebrovascular aging 3 . Both histopathological and imaging studies have shown that the prevalence and severity of vascular tortuosity begin to increase in middle age and continue to progress in older adults 4 - 10 . Similar age-related changes in arterial tortuosity have also been observed in animal models of aging 11 , 12 . Moreover, increased venular tortuosity has been associated with early markers of cerebral small vessel disease and cognitive decline, suggesting a potential role in the pathogenesis of vascular cognitive impairment 13 . Accurately quantifying vascular tortuosity is, therefore, providing important insights into the biological aging of the brain and its related pathologies 1 , 14 - 19 . For instance, tortuous vessels in the white matter are frequently surrounded by perivascular cavities, indicative of parenchymal tissue loss. Additionally, the increased length and curvature of these vessels can lead to reduced kinetic energy and perfusion pressure, thereby increasing the risk of ischemic damage 20 . The mechanical buckling hypothesis 1 , 21 proposes that tortuosity arises as a result of vessel buckling due to elevated luminal pressure or weakened surrounding tissue. This model provides a theoretical link between vascular tortuosity and arterial pulsatility 22 - 28 , although further research is necessary to establish the nature of this relationship. Magnetic resonance angiography (MRA) offers a non-invasive modality to visualize and assess cerebral vasculature in vivo. The most widely used technique, time-of-flight (TOF) MRA, enables high-resolution volume imaging of intracranial arteries at the level of the macrocirculation 29 in the main arteries and veins 30 - 33 . It relies on the signal enhancement of inflowing blood and the simultaneous suppression of stationary signals. Time-of-flight MRA can be tailored to emphasize either arterial or venous structures. When applied to highlight arteries, the technique relies on the inflow of unsaturated blood entering the imaging volume via the carotid and basilar arteries in the neck, producing what is commonly referred to as an arteriogram. The arteriogram is the focus of the present study. Conversely, when TOF-MRA is configured to enhance venous signal, it visualizes blood draining from the superior regions of the brain, resulting in a venogram. While TOF-MRA is capable of producing arteriograms with spatial resolution in the millimeter range or slightly higher, the resulting images are not pure representations of the arterial vasculature. Residual signals from stationary brain parenchyma, extra-cerebral tissues, and venous structures can contaminate the image and obscure arterial detail. In clinical settings, it is common practice for the MRI operator to manually remove these unwanted signals using editing tools on maximum intensity projections (MIPs). However, this manual intervention is time-consuming and subject to inter-operator variability. Automated methods for generating clean arteriograms would not only reduce workload but also improve reproducibility by minimizing dependence on user interaction. In addition, accurate segmentation of cerebral arteries from MRA images is a prerequisite for reliable tortuosity analysis. Traditional segmentation methods often struggle with the complex geometry and variability of cerebral vasculature. However, the advent of deep learning has revolutionized this domain 34 . Convolutional Neural Networks (CNNs), particularly architectures like U-Net 35 and its variants, have demonstrated superior performance in segmenting intricate vascular structures. For example, studies have shown that 3D U-Net models can effectively delineate cerebral vessels, achieving high dice similarity coefficients and enabling the extraction of detailed vascular features 36 . These advancements not only enhance the accuracy of vessel segmentation but also streamline the process, making large-scale analyses of vascular tortuosity feasible. Recent methodologies have employed advanced image processing techniques to extract vessel centerlines and compute tortuosity metrics. For instance, the use of skeletonization algorithms combined with path-finding methods allows for the quantification of vessel curvature and length, facilitating the assessment of age-related vascular changes 37 , 38 . Such quantitative analyses are crucial for establishing correlations between vascular morphology and chronological aging. In this contribution, we provide a tool that leverages an off-the-shelf neural network trained on publicly available segmented MRA data to fully automate cerebral artery segmentation and the estimation of tortuosity. Segmentation is performed for the full field-of-view covered in the MRAs and without manual selection of arterial seed points 7 , 8 . Our study verifies the increase of tortuosity with age in a large sample of MRAs. The automatically estimated tortuosity provides a quantitative summary marker that could be used as an assessment of brain health and as a parameter in further correlation studies that characterize the biological brain age of a subject. 2. Materials and methods 1) Data We use two main data sets, a training and an analysis data set. Training dataset: We trained the ML model using the COSTA data set 39 . COSTA is a large (N=423), multi-center, semi-automatically annotated dataset of 3D TOF-MRA brain images, designed for cerebrovascular segmentation research. To create accurate cerebrovascular segmentations for the COSTA dataset, which are used as a gold standard here, the authors employed a semi-automated annotation pipeline combining deep learning with expert manual correction. The annotation process and review process required approximately 2 to 2.5 hours per volume, underscoring the labor-intensive nature of high-fidelity vascular segmentation. Analysis dataset: To assess the correlation between age and vascular tortuosity, we combined two datasets with publicly available age metadata. The IXI dataset ( https://brain-development.org/ixi-dataset/ ) consists of 570 MRA scans, and the BRAVA ( http://cng.gmu.edu/brava ) dataset includes 56 MRA scans. To prevent data leakage and to ensure that the tortuosity analysis reflects model performance on unseen data, 136 IXI scans that were part of the COSTA training set were excluded from the analysis. This ensures that the resulting tortuosity metrics are applicable for downstream clinical use. Finally, 12 scans had to be removed from the analysis as their age information was missing. In total, the final analysis included 478 unique subjects, with ages ranging from 19 to 86 years. Imaging resolution varied by dataset: IXI-Guys and IXI-HH had a resolution of 0.47 × 0.47 × 0.8 mm, IXI-IOP had a resolution of 0.26 × 0.26 × 0.8 mm, and BRAVA had a resolution of 0.62 × 0.62 × 0.62 mm. From a visual inspection of the data MIPs, IXI-IOP had a superior quality in terms of detail but also in terms of better background signal suppression. 2) Data preprocessing Intensity matching: MRA images often exhibit varying intensity values for the same tissue, largely due to differences in individual scanners, vendors, and imaging protocols. To address this, we applied z-score normalization 40 or histogram matching to reduce intensity variability across datasets 39 . Skull stripping: To improve the performance of machine learning models during training and inference, it is common practice to remove non-brain tissues such as the skull in brain MRI scans. The skull typically contains no relevant information for most neuroimaging tasks, including the one in this work. For this purpose, we used HD-Bet, a deep learning-based brain extraction tool that provides fast and accurate skull tripping across a variety of MRI modalities and scanner types 41 . 3) Training Using the COSTA dataset, we utilized the nnU-Net framework 40 to train a U-Net model. The resulting ASN (Angiogram Segmentation Network) model is an ensemble comprising five distinct models, each trained on a separate fold of the dataset using an 80/20 training-validation split. Each fold applied a series of data augmentation techniques, including Gaussian blurring, Gaussian noise addition, brightness adjustment, low-resolution simulation, gamma transformations, spatial transformations and mirroring. Model training employed a composite loss function combining Dice loss and Cross-Entropy loss, integrated with deep supervision. Optimization was performed using Stochastic Gradient Descent (SGD) with an initial learning rate of 0.01 and a weight decay of 3 × 10 -5 . A polynomial learning rate scheduler was applied throughout training, which was conducted in a patch-wise manner with a patch size of [64 × 224 × 160] and a batch size of 2. The model was trained for 1000 epochs on an NVIDIA A40 GPU. 4) Data analysis postprocessing Before calculating tortuosity, fractal dimension, and branch length, we applied several post-processing steps to clean the vasculature skeletons generated by ASN: Skeleton Dilation: ML-generated vasculature skeletons can sometimes be fragmented, either due to missing vessel segments or minor discontinuities involving just a few voxels. To reconnect vessels split by small gaps, we applied a small 1-iteration binary dilation to the segmentation mask 42 . Skeletonization: After dilation, we skeletonized 43 , 44 the segmentation mask to reduce vessel structures to a one-voxel-wide representation suitable for graph-based analysis. Graph Construction: We constructed a bidirectional graph where each skeleton voxel was connected to its 26 possible neighbors. This enables straightforward detection of endpoints (nodes with one neighbor) and branch points (nodes with three or more neighbors). After graph construction, we retained only the largest connected component. Branch Point and Endpoint Grouping: Due to imperfections in segmentation, particularly at bifurcations, multiple adjacent voxels could be incorrectly classified as separate branch points. To resolve this, we applied DBSCAN 45 clustering to group nearby points and retained only the point closest to the cluster center. Small Vessel Removal: Finally, we removed branches shorter than 9 voxels, as these were likely artifacts or incomplete vessel segments not fully captured by ASN. 5) Tortuosity estimation There are various definitions for arterial tortuosity, some motivated by an easy manual estimation procedure for single or a small number of arteries, others more amenable for an automatic estimation 1 , 2 , 7 , 16 . They all have in common that they intend to measure the same property, namely the elongation of arteries beyond the shortest possible path, and thus have been found to quite generally provide qualitatively similar information (but see Kashyap et al. 46 , Martelli & Giacomozzi 47 ). Beyond the estimation formula, the definition of the vessel section onto which the estimation is being applied to is of importance. For example, from branchpoint to branchpoint (bifurcating), branchpoint or other seed point to endpoint (terminating), or along the whole vessel. The latter is difficult to implement, as vessels are branching, and the selection of the main vessel is somewhat arbitrary. Here we used the branch tortuosity also used in the BRAVA data set, defined by Diedrich et al. 48 , on the post-processed ASN labels. 3. Results 1) Validation of the ASN model An example subject’s MRA, its ground truth COSTA labels, its ASN labels, and skeletonized labels are shown in Fig. 1 . Visual inspection of the complete data set revealed the following. Overall, the ASN labels showed more vascularity than the ground truth data. However, not all of it reflected cerebral arteries; in other words, sporadic venous components or extracerebral blood vessel were visible. The ASN images sometimes showed more occipital arteries that were often not visible in the ground truth. Finally, the ASN model sometimes lacked or only had incomplete internal carotids at the base of the brain and sporadic lack of other vascular tree components. Download figure Open in new tab Figure 1: Comparison of ground truth COSTA labels and ASN labels for one representative subject. The first column shows MIPs of the MR arteriogram, the second column MIPs of the ground truth labels, the third column MIPs of the ASN labels, and the fourth column the skeleton derived from these labels. The skeleton follows from the first three post-processing steps in the Methods section and was dilated for visibility. The three rows correspond to axial, sagittal, and coronal views. For a more quantitative analysis, an additional 61 cases from the COSTA dataset were used to evaluate the model’s performance using the Dice score. On this test set, the ASN model achieved a mean Dice score of 0.95 with a standard deviation of 0.02, and a median Dice score of 0.95. 2) Validation of the ASN model by tortuosity-age law When the ASN model was applied to the unprocessed BRAVA MRA images (N = 56), the relationship between age and tortuosity could be modeled by a linear regression with R = 0.29 (p = 0.03), indicating a statistically significant increase of tortuosity with age ( Fig. 2A ). When terminating and bifurcating branches were analyzed separately, the correlation remained marginally significant for terminating branches (R = 0.28, p = 0.04), and only suggested a weak correlation for bifurcating branches (R = 0.19, p = 0.15) ( Fig. 2B ). For comparison, the semi-manually segmented MRA data reported by Wright et al. 8 showed stronger correlations: R = 0.32 (p < 0.0006) for terminating branches and R = 0.44 (p < 0.0002) for bifurcating branches. Another summary marker, the mean fractal dimension, increased significantly with age ( Fig. 2C ; R = 0.37, p = 0.005). This finding again is in concordance with the results of Wright et. al, who also observed a significant increase of fractal dimension with age. The mean branch length did not change with age ( Fig. 2D ). However, and again in concordance with Wright et al., the bifurcating part of the branch length correlated with age (not shown; R = 0.27, p = 0.047) Download figure Open in new tab Figure 2: Tortuosity for the BRAVA data set in dependence of subject’s age (A) and a comparison of bifurcating vs. terminal branch estimations of tortuosity vs. age (B). Also, fractal dimension (C) and branch length (D) vs. age. 3) The age dependence of tortuosity in a large data set Next, the BRAVA and IXI data sets were combined to a large data set to test the hypothesis of an age dependence of the tortuosity in a larger sample (N = 478). When the ASN model was applied to the unprocessed images of the combined BRAVA-IXI data set, the relationship between age and tortuosity could be modeled by a linear regression with R = 0.24 (p = 9 × 10 -8 ), indicating a statistically highly significant increase of tortuosity with age ( Fig. 3A ). When terminating and bifurcating branches were analyzed separately, the correlation remained significant and positive both for terminating branches (R = 0.14, p = 0.002) and bifurcating branches (R = 0.25, p = 2 × 10 -8 ) ( Fig. 3B ). The mean fractal dimension increased significantly with age ( Fig. 3C ; R = 0.23, p = 2 × 10 -7 ). The fact that the mean branch length did not change with age ( Fig. 3D ; R = 0.02, p = 0.73) is notable and will be discussed later. Download figure Open in new tab Figure 3: Tortuosity for the combined data set in dependence of subject’s age (A) and a comparison of bifurcating vs. terminal branch estimations of tortuosity vs. age (B). Also, fractal dimension (C) and branch length (D) vs. age. 4. Discussion 1) Validation of the ASN model by the ground truth Whereas the ASN labels of cerebral arteries were less accurate than the semi-manually derived ground truth labels, the ASN labels still allowed for a useful rendering of cerebral arteries, with a high Dice similarity coefficient between ASN and ground truth COSTA labels. Notably, ASN is applicable to MRA data of varying image resolution and quality, including those typically encountered in clinical TOF-MRA. The observed tortuosity–age relationship was derived from datasets acquired at different resolutions, demonstrating the method’s robustness across standard imaging protocols. We expect that an application of the trained model to MRAs of different scanner manufacturers and imaging protocols would provide serviceable results. 2) Validation of the ASN model by tortuosity—age law In the comparison of tortuosity from our ASN model segmentation with the tortuosity from the BRAVA data 8 , it turned out that our regression was less significant globally and with respect to the terminating branches, and not significant for the bifurcating branches. A weaker statistic was to be expected, as our fully automated segmentation is less accurate than a semi-manual segmentation of data. Importantly, the general positive correlation of tortuosity could be reproduced even in this relatively small data set. 3) The age dependence of tortuosity We could reproduce the published finding of a positive correlation of cerebroarterial tortuosity in a large data set of MRAs with high significance (p = 9 × 10 -8 ). The fact that the mean branch length did not change with age (R = 0.02, p = 0.73; Fig. 3D ) is interesting, as one would expect that with increasing tortuosity the branch lengths have to increase, too. A more detailed look into bifurcating vs. terminating branches shows that indeed the bifurcating branch lengths suggest an increase with age (R = 0.08, p = 0.07), whereas the terminating branch lengths decrease with age (R = −0.14, p = 0.002). Based on a review of scans from older patients, it seems that the model underperforms in these patients, capturing less information in terminating branches. The reasons for this underperformance might be the model itself, rooted in the physiology of aging, e.g. age-related changes in blood flow or vessel wall thickness, or even in the way the TOF-MRI captures aging blood vessels compared to younger blood vessels. For example, MRI parameters such as flip angles, echo times, and repeat times affect the MRA contrast 49 - 52 , and it might be necessary to actually adjust those to the age of the subjects in order to reduce bias. These and alternative hypotheses have been discussed previously 7 . To eliminate the possibility that the model itself falls short, one could use larger training data sets for older patients; only 15% of the subjects in the IXI-COSTA training dataset are 50 years or older. The scale at which tortuosity was investigated was determined by the visibility of arteries in typical clinical MRAs. Using ultra-high field MRI and contrast agents, significantly smaller scales can be reached 10 . Delineating small arteries at these scales and distinguishing them from veins is a labor-intensive task. The authors hope that, in the future, machine learning approaches such as the one presented here may assist in this important challenge given the potential significance of small arteries to the health of the aging brain 53 - 56 . In a similar way, a possible dependence of arterial tortuosity on the brain region 6 , 48 , 57 has not been investigated here but may be possible with machine learning in the foreseeable future, too. 5. Conclusion The increase of arterial tortuosity with age in the human brain is a phenomenon that has been found in histopathological and imaging studies based on small to intermediate sized data sets. A crucial step in imaging studies is the segmentation of the arterial tree from the images, which are usually not perfect renderings of the vasculature. We are proposing a machine learning based approach for the segmentation step in magnetic resonance angiography images. Our ASN model has been trained and validated on a previously published ground truth data set. In addition, we could approximately reproduce the age dependence law of arterial tortuosity that was previously published in another data set, as well in an application to a large data set (N = 478). The trained ASN model can be used for an instant estimation of arterial tortuosity from individual MRA data, providing a summary marker for the subject’s cerebrovascular status. Data Availability The ASN model for generating a brain arteriogram from data is available on GitHub: https://github.com/rachitsaluja/ASN Declaration of competing interest The authors do not have any competing interests. CRediT authorship contribution statement Jorge T. Gomez: Conceptualization, Data curation, Validation, Writing – original draft, methodology, review & editing Rachit Saluja: Conceptualization, Data curation, Validation, Writing – original draft, methodology, review & editing Mert R. Sabuncu: Supervision, Writing – review & editing Henning U. Voss: Supervision, Conceptualization, Writing – original draft, review & editing Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work the authors used the ChatGPT (May 2025 version) large language model ( https://chat.openai.com/ ) in order to search for additional references and to improve the style of some paragraphs. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication. Acknowledgements The authors thank Armaan K. Tewary for help with manual segmentation of arteriograms, initial literature exploration, and early training and data preprocessing scripts. References 1. ↵ Han HC . Twisted blood vessels: Symptoms, etiology and biomechanical mechanisms . J Vasc Res 2012 ; 49 ( 3 ): 185 – 197 . OpenUrl CrossRef PubMed 2. ↵ Ciurica S , Lopez-Sublet M , Loeys BL , Radhouani I , Natarajan N , Vikkula M et al. Arterial tortuosity novel implications for an old phenotype . Hypertension 2019 ; 73 ( 5 ): 951 – 960 . OpenUrl CrossRef PubMed 3. ↵ Xu XL , Wang B , Ren CH , Hu JN , Greenberg DA , Chen TX et al. Age-related impairment of vascular structure and functions . Aging Dis 2017 ; 8 ( 5 ): 590 – 610 . OpenUrl CrossRef PubMed 4. ↵ Hassler O. Arterial deformities in senile brains - Occurrence of deformities in a large autopsy series and some aspects of their functional significance . Acta Neuropathol 1967 ; 8 ( 3 ): 219 – 229 . OpenUrl CrossRef PubMed 5. Farkas E , Luiten PGM . Cerebral microvascular pathology in aging and Alzheimer’s disease . Prog Neurobiol 2001 ; 64 ( 6 ): 575 – 611 . OpenUrl CrossRef PubMed Web of Science 6. ↵ Thore CR , Anstrom JA , Moody DM , Challa VR , Marion MC , Brown WR . Morphometric analysis of arteriolar tortuosity in human cerebral white matter of preterm, young and aged subjects . J Neuropathol Exp Neurol 2007 ; 66 ( 5 ): 446 – 446 . OpenUrl 7. ↵ Bullitt E , Zeng DL , Mortamet B , Ghosh A , Aylward SR , Lin WL et al. The effects of healthy aging on intracerebral blood vessels visualized by magnetic resonance angiography . Neurobiol Aging 2010 ; 31 ( 2 ): 290 – 300 . OpenUrl CrossRef PubMed 8. ↵ Wright SN , Kochunov P , Mut F , Bergamino M , Brown KM , Mazziotta JC et al. Digital reconstruction and morphometric analysis of human brain arterial vasculature from magnetic resonance angiography . NeuroImage 2013 ; 82 : 170 – 181 . OpenUrl PubMed 9. Kamenskiy AV , Pipinos II , Carson JS , MacTaggart JN , Baxter BT . Age and disease-related geometric and structural remodeling of the carotid artery . J Vasc Surg 2015 ; 62 ( 6 ): 1521 – 1528 . OpenUrl PubMed 10. ↵ Sun Z , Li CY , Wisniewski TW , Haacke EM , Ge YL . Detection of age-related tortuous cerebral small vessels using ferumoxytol-enhanced 7T MRI . Aging Dis 2024 ; 15 ( 4 ): 1913 – 1926 . OpenUrl PubMed 11. ↵ Li YD , Choi WJ , Wei W , Song SZ , Zhang QQ , Liu JL et al. Aging-associated changes in cerebral vasculature and blood flow as determined by quantitative optical coherence tomography angiography . Neurobiol Aging 2018 ; 70 : 148 – 159 . OpenUrl CrossRef PubMed 12. ↵ Lowerison MR , Sekaran NC , Zhang W , Dong ZJ , Chen X , Llano DA et al. Aging-related cerebral microvascular changes visualized using ultrasound localization microscopy in the living mouse . Scientific Reports 2022 ; 12 ( 1 ). 13. ↵ Fulop GA , Tarantini S , Yabluchanskiy A , Molnar A , Prodan C , Kiss T et al. Role of age-related alterations of the cerebral venous circulation in the pathogenesis of vascular cognitive impairment . Am J Physiol-Heart C 2019 ; 316 ( 5 ): H1124 – H1140 . OpenUrl 14. ↵ Spangler KM , Challa VR , Moody DM , Bell MA . Arteriolar tortuosity of the white-matter in aging and hypertension - a microradiographic study . J Neuropathol Exp Neurol 1994 ; 53 ( 1 ): 22 – 26 . OpenUrl PubMed 15. Chu LC , Haroun RR , Beaulieu RJ , Black JH , Dietz HC , Fishman EK . Carotid artery tortuosity index is associated with the need for early aortic root replacement in patients with Loeys-Dietz Syndrome . J Comput Assist Tomo 2018 ; 42 ( 5 ): 747 – 753 . OpenUrl 16. ↵ Klis KM , Krzyzewski RM , Kwinta BM , Lasocha B , Brzegowy P , Stachura K et al. Increased tortuosity of basilar artery might be associated with higher risk of aneurysm development . Eur Radiol 2020 ; 30 ( 10 ): 5625 – 5632 . OpenUrl PubMed 17. Liu JY , Ke XT , Lai QQ . Increased tortuosity of bilateral distal internal carotid artery is associated with white matter hyperintensities . Acta Radiol 2021 ; 62 ( 4 ): 515 – 523 . OpenUrl PubMed 18. van Laarhoven CJHCM , Willemsen SI , Klaassen J , de Vries EE , van der Vliet QMJ , Hazenberg CEVB et al. Carotid tortuosity is associated with extracranial carotid artery aneurysms . Quant Imag Med Surg 2022 ; 12 ( 11 ): 5018 – 5029 . OpenUrl 19. ↵ Khaing ZZ , Chandrasekaran A , Katta A , Reed MJ . The brain and spinal microvasculature in normal aging . J Gerontol a-Biol 2023 ; 78 ( 8 ): 1309 – 1319 . OpenUrl 20. ↵ Brown WR , Thore CR . Cerebral microvascular pathology in ageing and neurodegeneration . Neuropathol Appl Neurobiol 2011 ; 37 ( 1 ): 56 – 74 . OpenUrl CrossRef PubMed Web of Science 21. ↵ Jackson ZS , Dajnowiec D , Gotlieb AI , Langille BL . Partial off-loading of longitudinal tension induces arterial tortuosity . Arterioscl Throm Vas 2005 ; 25 ( 5 ): 957 – 962 . OpenUrl Abstract / FREE Full Text 22. ↵ Tong YJ , Hocke LM , Frederick BD . Short repetition time multiband echo-planar imaging with simultaneous pulse recording allows dynamic imaging of the cardiac pulsation signal . Magn Reson Med 2014 ; 72 ( 5 ): 1268 – 1276 . OpenUrl CrossRef PubMed 23. Voss HU , Dyke JP , Tabelow K , Schiff ND , Ballon DJ . Magnetic resonance advection imaging of cerebrovascular pulse dynamics . J Cereb Blood Flow Metab 2017 ; 37 ( 4 ): 1223 – 1235 . OpenUrl PubMed 24. Lahiri S , Schlick KH , Padrick MM , Rinsky B , Gonzalez N , Jones H et al. Cerebral pulsatility index is elevated in patients with elevated right atrial pressure . J Neuroimaging 2018 ; 28 ( 1 ): 95 – 98 . OpenUrl PubMed 25. Voss HU . Hypersampling of pseudo-periodic signals by analytic phase projection . Comput Biol Med 2018 ; 98 : 159 – 167 . OpenUrl PubMed 26. Aslan S , Hocke L , Schwarz N , Frederick B. Extraction of the cardiac waveform from simultaneous multislice fMRI data using slice sorted averaging and a deep learning reconstruction filter . NeuroImage 2019 ; 198 : 303 – 316 . OpenUrl PubMed 27. Vigen T , Ihle-Hansen H , Lyngbakken MN , Berge T , Thommessen B , Ihle-Hansen H et al. Carotid atherosclerosis is associated with middle cerebral artery pulsatility index . J Neuroimaging 2020 ; 30 ( 2 ): 233 – 239 . OpenUrl CrossRef PubMed 28. ↵ Voss HU , Razlighi QR . Pulsatility analysis of the circle of Willis . Aging Brain 2024 ; 5 . 29. ↵ Caro CG , Parker KH . Mechanics and imaging of the macrocirculation . Magn Reson Med 1990 ; 14 ( 2 ): 179 – 86 . OpenUrl PubMed Web of Science 30. ↵ Alfidi RJ , Masaryk TJ , Haacke EM , Lenz GW , Ross JS , Modic MT et al. MR angiography of peripheral, carotid, and coronary arteries . AJR Am J Roentgenol 1987 ; 149 ( 6 ): 1097 – 109 . OpenUrl CrossRef PubMed Web of Science 31. Haacke EM , Masaryk TJ , Wielopolski PA , Zypman FR , Tkach JA , Amartur S et al. Optimizing blood vessel contrast in fast three-dimensional MRI . Magn Reson Med 1990 ; 14 ( 2 ): 202 – 21 . OpenUrl CrossRef PubMed 32. Parker DL , Yuan C , Blatter DD . MR angiography by multiple thin slab 3D acquisition . Magn Reson Med 1991 ; 17 ( 2 ): 434 – 451 . OpenUrl CrossRef PubMed 33. ↵ Alsop DC , Detre JA . Reduced transit-time sensitivity in noninvasive magnetic resonance imaging of human cerebral blood flow . J Cereb Blood Flow Metab 1996 ; 16 ( 6 ): 1236 – 1249 . OpenUrl CrossRef PubMed Web of Science 34. ↵ Yamada T , Yoshimura T , Ichikawa S , Sugimori H. Improving cerebrovascular imaging with deep learning: Semantic segmentation for time-of-flight magnetic resonance angiography maximum intensity projection image enhancement . Appl Sci-Basel 2025 ; 15 ( 3034 ): 1 – 18 . OpenUrl 35. ↵ Ronneberger O , Fischer P , Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation . Medical Image Computing and Computer-Assisted Intervention, Part III 2015 ; 9351 : 234 – 241 . OpenUrl 36. ↵ de Vos V , Timmins KM , van der Schaaf IC , Ruigrok Y , Velthuis BK , Kuijf HJ . Automatic cerebral vessel extraction in TOF-MRA using deep learning . Medical Imaging 2021: Image Processing 2021 ; 11596 . 37. ↵ Avadiappan S , Payabvash S , Morrison MA , Jakary A , Hess CP , Lupo JM . A fully automated method for segmenting arteries and quantifying vessel radii on magnetic resonance angiography images of varying projection thickness . Front Neurosci 2020 ; 14 . 38. ↵ Yoon HS , Oh J , Kim YC , Corbo D. Assessing machine learning models for predicting age with intracranial vessel tortuosity and thickness information . Brain Sciences 2023 ; 13 ( 11 ). 39. ↵ Mou L , Lin JH , Zhao YF , Liu YH , Ma SD , Zhang J et al. COSTA: A multi-center TOF-MRA dataset and a style self-consistency network for cerebrovascular segmentation . IEEE Trans Med Imaging 2024 ; 43 ( 12 ): 4442 – 4456 . OpenUrl CrossRef PubMed 40. ↵ Isensee F , Jaeger PF , Kohl SAA , Petersen J , Maier-Hein KH . nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation . Nat Methods 2021 ; 18 ( 2 ): 203 – 211 . OpenUrl CrossRef PubMed 41. ↵ Isensee F , Schell M , Pflueger I , Brugnara G , Bonekamp D , Neuberger U et al. Automated brain extraction of multisequence MRI using artificial neural networks . Human Brain Mapping 2019 ; 40 ( 17 ): 4952 – 4964 . OpenUrl CrossRef PubMed 42. ↵ scipy.ndimage.binary_dilation . https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.binary_dilation.html . 43. ↵ Lee TC , Kashyap RL , Chu CN . Building skeleton models via 3-D medial surface axis thinning algorithms . Cvgip-Graph Model Im 1994 ; 56 ( 6 ): 462 – 478 . OpenUrl CrossRef 44. ↵ Bertrand G , Malandain G. A note on building skeleton models via 3-D medial surface axis thinning algorithms . Graph Model Im Proc 1995 ; 57 ( 6 ): 537 – 538 . OpenUrl 45. ↵ Ester M , Kriegel H-P , Sander J , Xu X. A density-based algorithm for discovering clusters in large spatial databases with noise . In: Proceedings of the Second International Conference on Knowledge Discovery and Data Mining . Portland, Oregon : AAAI Press , 1996 . pp 226 – 231 . 46. ↵ Kashyap V , Gharleghi R , Li DD , McGrath-Cadell L , Graham RM , Ellis C et al. Accuracy of vascular tortuosity measures using computational modelling . Scientific Reports 2022 ; 12 ( 1 ). 47. ↵ Martelli F , Giacomozzi C. Tortuosity index calculations in retinal images: Some criticalities arising from commonly used approaches . Information 2021 ; 12 ( 11 ). 48. ↵ Diedrich KT , Roberts JA , Schmidt RH , Kang CK , Cho ZH , Parker DL . Validation of an arterial tortuosity measure with application to hypertension collection of clinical hypertensive patients . BMC Bioinformatics 2011 ; 12 : 1 – 12 . OpenUrl CrossRef PubMed 49. ↵ Axel L. Blood-flow effects in magnetic-resonance imaging . Am J Roentgenol 1984 ; 143 ( 6 ): 1157 – 1166 . OpenUrl CrossRef PubMed Web of Science 50. Bradley WG , Waluch V. Blood-flow-magnetic-resonance imaging . Radiology 1985 ; 154 ( 2 ): 443 – 450 . OpenUrl CrossRef PubMed Web of Science 51. Wehrli FW . Time-of-flight effects in MR imaging of flow . Magn Reson Med 1990 ; 14 ( 2 ): 187 – 193 . OpenUrl CrossRef PubMed 52. ↵ Saloner D. The aapm/rsna physics tutorial for residents - An introduction to MR-angiography . Radiographics 1995 ; 15 ( 2 ): 453 – 465 . OpenUrl CrossRef PubMed 53. ↵ Iadecola C. The pathobiology of vascular dementia . Neuron 2013 ; 80 ( 4 ): 844 – 866 . OpenUrl CrossRef PubMed Web of Science 54. Chiang GC , Hernandez JCC , Kantarci K , Jack CR , Weiner MW , Initi AsDN . Cerebral microbleeds, CSF p-Tau, and cognitive decline: Significance of anatomic distribution . Am J Neuroradiol 2015 ; 36 ( 9 ): 1635 – 1641 . OpenUrl Abstract / FREE Full Text 55. Wang Y , Spincemaille P , Liu Z , Dimov A , Deh K , Li JQ et al. Clinical quantitative susceptibility mapping (QSM): Biometal imaging and its emerging roles in patient care . J Magn Reson Imaging 2017 ; 46 ( 4 ): 951 – 971 . OpenUrl PubMed 56. ↵ Lee K , Mahmud M , Marx D , Yasen W , Sharma O , Ivanidze J et al. Clinical arterial spin-labeling MR imaging to screen for typical and atypical neurodegenerative disease in the new era of Alzheimer treatment . Am J Neuroradiol 2024 ; 45 ( 5 ): 632 – 636 . OpenUrl Abstract / FREE Full Text 57. ↵ Zhou S , Qiao Y , Zhou XW , Wasserman BA , Caughey MC . Detection of dolichoectasia and atherosclerosis by automated MRA tortuosity metrics in a population-based study . J Magn Reson Imaging 2024 ; 59 ( 5 ): 1612 – 1619 . OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted September 12, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Age-related changes of the brain’s arterial network assessed with machine learning segmentation Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Age-related changes of the brain’s arterial network assessed with machine learning segmentation Jorge T. Gomez , Rachit Saluja , Mert R. Sabuncu , Henning U. Voss medRxiv 2025.09.10.25335518; doi: https://doi.org/10.1101/2025.09.10.25335518 Share This Article: Copy Citation Tools Age-related changes of the brain’s arterial network assessed with machine learning segmentation Jorge T. Gomez , Rachit Saluja , Mert R. Sabuncu , Henning U. Voss medRxiv 2025.09.10.25335518; doi: https://doi.org/10.1101/2025.09.10.25335518 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Radiology and Imaging Subject Areas All Articles Addiction Medicine (568) Allergy and Immunology (863) Anesthesia (297) Cardiovascular Medicine (4417) Dentistry and Oral Medicine (443) Dermatology (381) Emergency Medicine (606) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1507) Epidemiology (15209) Forensic Medicine (30) Gastroenterology (1120) Genetic and Genomic Medicine (6578) Geriatric Medicine (666) Health Economics (995) Health Informatics (4516) Health Policy (1366) Health Systems and Quality Improvement (1610) Hematology (538) HIV/AIDS (1264) Infectious Diseases (except HIV/AIDS) (15906) Intensive Care and Critical Care Medicine (1103) Medical Education (620) Medical Ethics (144) Nephrology (667) Neurology (6578) Nursing (345) Nutrition (998) Obstetrics and Gynecology (1139) Occupational and Environmental Health (956) Oncology (3321) Ophthalmology (970) Orthopedics (369) Otolaryngology (420) Pain Medicine (435) Palliative Medicine (129) Pathology (663) Pediatrics (1689) Pharmacology and Therapeutics (691) Primary Care Research (710) Psychiatry and Clinical Psychology (5425) Public and Global Health (9211) Radiology and Imaging (2192) Rehabilitation Medicine and Physical Therapy (1368) Respiratory Medicine (1194) Rheumatology (593) Sexual and Reproductive Health (709) Sports Medicine (529) Surgery (709) Toxicology (99) Transplantation (288) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9fef1e826b5e06e3',t:'MTc3OTMxOTcyMQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();