MRI-based surface reconstruction and cortical thickness estimation of the human brain: Benchmarking deep-learning based morphometry tools

MRI-based surface reconstruction and cortical thickness estimation of the human brain: Benchmarking deep-learning based morphometry tools | 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 MRI-based surface reconstruction and cortical thickness estimation of the human brain: Benchmarking deep-learning based morphometry tools Victor B. B. Mello , Richard McKinley , Roland Wiest , Christian Rummel doi: https://doi.org/10.1101/2025.03.22.25324274 Victor B. B. Mello 1 Support Center for Advanced Neuroimaging (SCAN) University Institute of Diagnostic and Interventional Neuroradiology University of Bern, Inselspital, Bern University Hospital , Bern, Switzerland . Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: vbraga{at}pos.if.ufrj.br Richard McKinley 1 Support Center for Advanced Neuroimaging (SCAN) University Institute of Diagnostic and Interventional Neuroradiology University of Bern, Inselspital, Bern University Hospital , Bern, Switzerland . Find this author on Google Scholar Find this author on PubMed Search for this author on this site Roland Wiest 1 Support Center for Advanced Neuroimaging (SCAN) University Institute of Diagnostic and Interventional Neuroradiology University of Bern, Inselspital, Bern University Hospital , Bern, Switzerland . Find this author on Google Scholar Find this author on PubMed Search for this author on this site Christian Rummel 2 European Campus Rottal-Inn Technische Hochschule Deggendorf Max-Breiherr-Straße 32 , D-84347 Pfarrkirchen, Germany 1 Support Center for Advanced Neuroimaging (SCAN) University Institute of Diagnostic and Interventional Neuroradiology University of Bern, Inselspital, Bern University Hospital , Bern, Switzerland . Find this author on Google Scholar Find this author on PubMed Search for this author on this site Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Establishing reliable and time efficient pipelines for structural MRI segmentation, parcellation and surface reconstruction, is essential to explore the potential clinical applications of research-grade morphometry tools. The integration between deep-learning based methods for fast whole-brain segmentation and the well known surface reconstruction algorithms is a viable alternative to perform this task. In this work, we applied this idea with three deep-learning based cortical parcellation models, DeepSCAN, FastsurferCNN and QuickNAT. With a 11 min surface reconstruction pipeline, we evaluated the performance of each segmentation beyond the voxel-based approaches and dice coefficient comparison between the generated parcellation and Freesurfer’s established silver standard. To prove the concept, we performed a direct comparison between the morphological variables obtained by our methodology and Freesurfer. Using a synthetic dataset, we benchmark each reconstruction pipeline based on the similarity to the ground-truth surface and reproduction of the expected surface-based metrics. The most robust pipeline across the human dataset and closer to the synthetic ground truth was based on DeepSCAN segmentation, producing a reliable morphometric tool with a processing time realistic for clinical applications like diagnostics support in individuals. Introduction Quantitative analysis of structural Magnetic Resonance Imaging (MRI) has been available for more than 20 years [ 1 – 3 ], serving as a powerful tool for studying the healthy brain and different neurological disorders across the human lifespan [ 4 ]. Whole-brain reconstruction from a structural MRI is typically performed by time consuming processing by one of a handful of standardly accepted tools: voxel-based morphometry (VBM), as provided by SPM [ 5 ], ANTs [ 6 ], and FSL [ 7 ], or cortical surface reconstruction for surface-based morphometry (SBM), as performed by Freesurfer [ 8 ], BrainSuite [ 9 ] and CIVET [ 10 ]. The complexity of the cortical geometry, the high biological variability across subjects and the memory requirements of the data are the main reasons that whole-brain reconstruction is so computationally demanding. From a clinical perspective, structural MRI provides a non-invasive visual assessment of the brain, with well established protocols for diagnosing and monitoring a wide range of neurological conditions [ 11 – 13 ]. Furthermore, various forms of brain morphometry derived from these images have significantly contributed to a deeper understanding of healthy brain development, aging, and the mechanisms underlying disease manifestation. For this reason, it would be beneficial to translate the morphometry tools produced within the research context to auxiliary quantitative assessments that can used in clinical practice [ 14 – 17 ]. In order to achieve this goal, it is essential to devise new reconstruction pipelines that are not only capable of fully reconstructing a structural MRI in a time efficient manner but also are robust and accurate against the inherent variability in the reality of clinical applications, e.g., multiple MRI machines, different imaging contrast and subjects. Deep-learning (DL) [ 18 ] based methods have become the first choice for fast whole-brain segmentation and parcellation. In the last few years, the field has rapidly evolved from architectures that generate brain segmentation with a runtime of 15-60 min [ 19 – 24 ] to models with a runtime of less than a minute [ 25 – 28 ]. However, there is much progress to be made in the use of DL for brain surface reconstruction. Only a few models are available [ 29 – 31 ] and, although they can be very fast (less than a 1 min runtime), they still generate surfaces with self-intersecting regions. The state-of-the-art model available [ 31 ] reported an improvement on the accuracy and regularity of DL-based surfaces, with a reduction from (1.47 ± 0.52)% to (0.24 ± 0.17)% of self-intersecting faces on the reconstructed pial. Still, there is no guarantee that the reconstructed surfaces will have the required topological properties. This makes Fastsurfer [ 28 ], which reconstructs a cortical surface by combining DL-based segmentation and Freesurfer in about 90 min, the fastest whole-brain reconstruction pipeline described in the literature so far with the certainty of generating topologically correct surfaces. The advent of novel structural MRI reconstruction techniques makes the development of methods to probe their accuracy imperative. However, a major obstacle to such evaluation is the lack of datasets with their indisputable ground truth reconstruction. The manual production of such golden standard by experts [ 32 ] is a time-consuming task that requires highly-specialized anatomical knowledge and may present significant variability across different experts [ 33 ]. To overcome this problem using real structural MRI, it is common to use Freesurfer reconstructions as a silver standard given the long standing trustworthiness on its results. An alternative is using multiple MRI acquisitions of the same cortex to perform a test-retest reliability assessment [ 28 , 34 ]. There are some caveats, with the former leading to an evaluation bias [ 34 ] and the latter being only a indirect indication of reconstruction accuracy. Another possible solution is the use of a synthetic ground-truth dataset generated by a controlled modification on a real brain MRI. As cortical thickness is a metric routinely used to study neurodegenerative and neurological conditions, this approach typically focuses on inducing synthetic atrophy by shrinking the cortex. By producing a topology-preserving deformation field from the pial surface to the white matter (WM) – grey matter (GM) interface, the atrophy is generated by moving the pial along this field, in the direction of the WM. The available models to create a synthetic cortical atrophy can be based on the Jacobian of the transformation [ 35 – 37 ], biomechanical models [ 38 – 40 ], binary morphological operations [ 41 ] or DL [ 42 , 43 ]. To extend the analysis of the relative performance of DL-based segmentation tools beyond the dice coefficient between the alternative cortical parcellation and a silver standard ground truth as provided by Freesurfer, in this work we aim to examine also the quality of morphometric variables extracted from these segmentations. To maximize compatibility with fast clinical deployment, we developed a fast surface reconstruction pipeline deeply inspired by the approach presented in [ 28 ]. This yields a reliable pipeline that can be applied to any high-quality segmentation method, reducing processing time for whole-brain reconstruction down to minutes. In this work we evaluate three state-of-the-art DL-based tools for whole-brain MRI segmentation for morphometry, DeepSCAN [ 27 ], FastsurferCNN [ 28 ] and QuickNAT [ 26 ] To prove the concept, we performed a direct comparison between the morphological variables obtained from DL-based segmentations by our proposed fast surface reconstruction pipeline and Freesurfer for a wide variety of subjects and MRI sequences. Then, using the synthetic dataset generated by [ 43 ], we benchmarked each reconstruction pipeline based on the similarity to the ground-truth surface and reproduction of the expected surface-based metrics. Materials and Methods Imaging data We study morphometric pipelines applied to both real T1w human brain MRI (obtained from open-access databases) and synthetic data (the DL-based cortical atrophy phantom generated by [ 43 ]). The human dataset aims to provide a proof of concept for the fast surface reconstruction pipeline using DL-based whole-brain segmentation. In order to demonstrate the generalizability of the fast surface methodology, the human dataset was compiled with a high variability regarding age, pathology and MRI scanning parameters for both male and female subjects. The dataset combines: (i) a random sample of 1000 young subjects from the Adolescent Brain Cognitive Development (ABCD) study [ 44 ]; The healthy young adults from the Human Connectome Project (HCP) [ 45 ]; (iii) The Phantom of Bern (PoB) [ 46 ], a longitudinal experiment with two healthy adults in which the MRI sequence parameters inversion time (TI) and repetition time (TR) are systematically changed to control the contrast between the WM and GM signal intensities; (iv) The elderly subjects, both healthy and with some degree of cognitive impairment, from the OASIS3 dataset [ 47 ]. A summary description of the human MRI dataset is given in Table 1 . Both HCP and OASIS3 dataset have subjects with more than one MRI scan taken at the same session, allowing a reproducibility test-retest analysis. With the exception of the PoB, which was designed for a controlled change on the MRI scanning parameters, all datasets were obtained in studies with more than one scanner and distinct acquisition parameters. View this table: View inline View popup Download powerpoint Table 1: Summary of the human MRI dataset. Abbreviations: CTL, healthy control, AD, Alzheimer’s disease. The synthetic dataset was generated using 20 heathy controls from the Alzheimer Disease Neuroimaging Initiative (ADNI) [ 48 , 49 ] by exploiting the inherent one-to-one correspondence between vertices of the WM surface and vertices of the pial surface on Freesurfer’s reconstructed meshes. A uniform controlled atrophy across the GM was introduced moving the pial surface in the direction of the WM with a fixed distance. Then, based on these deformed meshes, a synthetic MRI was generated using a generative adversarial network (GAN) model [ 50 ]. The synthetic dataset provides an expected value for the cortical thickness and ground-truth surfaces for the pial and white matter, allowing a direct comparison to benchmark each reconstruction methodology. The values of atrophies provided by the authors range from zero (no modification of the surfaces) to 1 mm. The processing pipeline with different segmentation/surface reconstruction methods All datasets were first processed using (A) Freesurfer v6.0 [ 8 ] reconstruction pipeline following the Desikan Killiany (DK) protocol atlas parcellation [ 51 ]. No manual corrections were performed [ 52 ]. These reconstructions for the human dataset were used as a silver standard comparison baseline given the long standing reliability of Freesurfer reconstructions (despite the inherent bias towards Freesurfer-based methods) For reconstruction from DL-based segmentation, we focused on three different methods: (B) DeepSCAN [ 27 ], (C) FastsurferCNN [ 28 ] and (D) QuickNAT [ 26 ]. FastsurferCNN and QuickNAT can be regarded as variants of the same architecture, with the former being inspired by the later. All human datasets were processed by the three DL-based segmentation models. Since (D) QuickNAT cannot process skull-stripped imaging, it was not applied to the synthetic CTh phantom. The surfaces for each DL-based segmentation method can be obtained with minimal modifications on the pipeline presented in [ 28 ]. Given a proper conformation of the input segmentation and a labeling map consistent with the number of labels that each DL model can predict, the surfaces are obtained with a significant speed-up compared to Freesurfer. Heavily inspired by this, the fast surface pipeline suggested in this work is a simplified version of the one suggested by [ 28 ]. Table 2 summarizes the steps omitted to produce the fast surface reconstruction pipeline suggested here. Based on the segmentation provided by each DL model, the WM surface is reconstructed using a topology correcting fast marching algorithm [ 53 ]. This surface encloses the cerebral WM and sub-cortical structures such as the ventricles, thalamus, caudate nucleus, putamen, pallidum, hippocampus, amygdala, accumbens nucleus and ventral diencephalon. Only the pial surface is reconstructed using the Freesurfer engines, requiring functionalities that normalize the MRI to ensure a WM signal with about 110 in gray-scale, modify the WM segmentation to avoid edges or corners and position the tessellation of the cortical surface [ 1 ]. Finally, each vertex of the tesselation is mapped on the labels provided by the segmentation with no spatial smoothing applied on the labels, allowing a region-wise analysis. View this table: View inline View popup Download powerpoint Table 2: Summary of the modifications made to the Fastsurfer reconstruction pipeline. To ensure that our simplification of the Fastsurfer pipeline does not affect the surfaces and results presented in this paper, the full Fastsurfer pipeline was used to clarify cases with large discrepancy in comparison to Freesurfer. These results are summarized in the supplementary material A. All the codes necessary to run the fast pipeline reconstruction are available at https://github.com/bragamello . The performance achieved in this paper was obtained using a computer with 8 Intel Xeon(R) E5 1630 v3 3.70 GHz Central Processing Unit (CPU) and a 8 Gb GeForce GTX 1080 Graphical Processing Unit (GPU). Data analysis From each reconstruction, we derived five common surface-based metrics: the volume enclosed by the WM and pial surfaces, their respective surface areas, and the cortical thickness. Cortical thickness was calculated from the reconstructed WM and pial surfaces using Freesurfer’s definition [ 54 ]. This method averages the closest distance measured from WM to pial surface and from pial to WM surface to generate a vertex-wise thickness estimate. The morphological variables were compared both by their values per hemisphere and regionally according to the DK atlas. In the case of FastsurferCNN predictions, the cortex parcellation follows the Desikan-Killiany-Tourville (DKT) Atlas [ 55 ]. To account for this difference, the bankssts, temporal pole and frontal pole were omitted in our region-wise comparisons. Given the small size of these regions, the effect of this difference has no impact on our general observations. As the synthetic dataset was conceived to benchmark thickness estimation methods, we included comparisons with a registration-based cortical thickness measurement (DiReCT) [ 56 ] for this particular dataset. This methodology not only was suggested [ 27 ] to detect subtle changes in cortical thickness from a MRI together with their segmentation model but also showed good performance with the synthetic dataset [ 43 ]. Table 3 summarizes all the processing pipelines and datasets available in this work. View this table: View inline View popup Download powerpoint Table 3: Summary of the processing pipelines used in each dataset. With the exception of QuickNAT, all models produce a brain parcellation allowing the assignment of vertex-wise labels to the pial surface reconstruction. For whole-hemisphere comparisons, we performed linear regression on the surface-based metrics derived from Freesurfer and our alternative method to estimate the slope and offset with a 95% confidence interval (C.I.). In the region-wise analysis, we calculated the median percentage deviation of the regional surface-based metrics relative to Freesurfer values. Additionally, to evaluate test-retest reliability, we used the repeated MRI scans from the HCP dataset. Using all possible combinations of repeated scans, we calculate de root mean square (RMS) of the difference in the surface-based metrics, reporting the mean value for each region. Although repeated MRI scans were also available for OASIS3, the cohort with young healthy adults were preferred in this analysis to avoid introducing extra confounder, such as repeated subjects due to bad quality images or extreme cases of atrophy. The relationship between surface reconstruction and the contrast between the WM and GM signal intensities on the MRI was examined using the PoB dataset by determining the correlation between all surface-based metrics and the measured WM/GM contrast. Following Freesurfer’s standard, the local WM/GM contrast was calculated based on the reconstructed WM surface, according to the ratio 2 × ( I WM − I GM ) / ( I WM + I GM ), by sampling I WM tissue intensity values at 1 mm inside the WM and I GM reaching 30% of the cortical thickness inside the GM. Regarding the synthetic dataset, the similarity of the surface reconstructions to the ground-truth surface was computed using the Hausdorff distance and the relative error in the measurement of the surface areas and volumes. For benchmarking the thickness estimation methodologies, we directly compared the synthetically imposed atrophy with the estimated one. Results Surface reconstruction A general overview of the surfaces obtained by the fast pipeline and the cortical parcellation provided by its respective DL-based model is given in Figure 1 using a scan from the PoB dataset with a mid-value contrast as an example. Further examples can be found at the Supplementary Material 12 Download figure Open in new tab Figure 1: Example of the results obtained from the fast surface reconstruction pipeline based on (B) DeepSCAN, (C) FastsurferCNN and (D) QuickNAT segmentation in comparison with the standard (A) Freesurfer reconstruction. The raw MRI is displayed together with the reconstructed WM (blue) and pial surfaces (red) for the same coronal slice to facilitate a direct comparison between surfaces. Apart from occasional segmentation failures, the surfaces are similar. A systematic exception is the hippocampus, which is not enclosed inside the WM for (A) Freesurfer. The 3D reconstruction of the pial surface showing the region labeling provided by each DL-based segmentation model allows a neuro-anatomical comparison with the DK Atlas. The subject from the PoB dataset is identified for reproducibility and the anatomical visualization was kept consistent to facilitate comparison between reconstructions. We retained the same coloring and naming presented in [ 51 ]. The distribution of runtimes required to reconstruct a healthy young adult structural MRI with the fast surface pipeline is shown in Figure 2 with a direct comparison with Fastsurfer and Freesurfer. Importantly, our fast surface pipeline is a simplification of Fastsurfer, which generates only the surfaces, the vertex-wise thickness, curvature and annotation as output but not the many other metrics that FreeSurfer and FastSurfer provide. This runtime reflects the minimum amount of time required to generate a surface and minimal outputs that can be further manipulated according to the desired application. Download figure Open in new tab Figure 2: The runtime distribution for the fast surface reconstruction, Fastsurfer and Freesurfer pipelines. Due to the significant difference in the time scale from minutes to hours, the results are shown in log scale. The analysis per hemisphere for subjects of the ABCD, HCP and OASIS3 datasets is shown in Figure 3 . In the supplementary information C, the Table 5 summarizes the values of the slope and offset with 95% C.I. obtained by the linear regression between surface-based metrics based on Freesurfer and the alternative fast surface reconstruction method. Download figure Open in new tab Figure 3: Comparison between the SBM metrics for a whole hemisphere obtained using the fast surface reconstruction based on (B) DeepSCAN, (C) FastsurferCNN and (D) QuickNAT segmentation models (y axes) and the standard (A) Freesurfer reconstruction (x axes). Each data point represents a subject, with the metric value obtained by the average between left and right hemispheres. The identity mapping is indicated by a dotted diagonal line. The regional median relative difference to Freesurfer’s pial area and cortical thickness calculated region-wise is shown in Figure 4 following the DK atlas. Since (D) QuickNAT does not provide a parcellation of the cortex, method (D) is excluded here. A test-retest reliability assessment in HCP subjects with repeated scans is shown in Figure 5 by the mean RMS of difference in the metrics calculated using repeated MRI scans. Download figure Open in new tab Figure 4: Median relative difference in the regional pial area, WM area and mean cortical thickness between Freesurfer reconstruction and the alternative fast surface reconstruction method based on (B) DeepSCAN, (C) FastsurferCNN segmentation. Negative values (blue hues) imply that values smaller than for Freesurfer were obtained. The DK atlas was used for parcellation. Download figure Open in new tab Figure 5: Test-retest reliability assessment of the pial surface area (top), WM surface area (middle) and cortical thickness (bottom) using HCP subjects with repeated MRI scans. Following the DK atlas, we determine the mean RMS of each metric using all rescan pairs of the same subject for (A) Freesurfer reconstruction and the fast surface pipeline based on DeepSCAN and (C) FastsurferCNN. The impact of different MRI WM/GM contrast on the reconstructed surface and the derived surface-based metrics is shown in Figure 6 for the PoB dataset. The volumes enclosed by the pial and WM surfaces were used to determine if the surfaces are shrinking or increasing as a function of the contrast. Table 4 summarizes the correlation coefficient and the statistical significance of the results. Download figure Open in new tab Figure 6: Correlation between the surface reconstruction and the morphological variables obtained from it with the MRI WM/GM contrast for (A) Freesurfer reconstruction and the fast surface pipeline based on (B) DeepSCAN, (C) FastsurferCNN and (D) QuickNAT. The average between left and right hemispheres were used to calculate the metrics for each data point. View this table: View inline View popup Download powerpoint Table 4: The Pearson correlation coefficient between the surface-based metrics and WM/GM contrast. The symbol * indicates the statistically significant results with p < 0.01. Benchmarking reconstruction pipelines The synthetic dataset provides an expected value for the cortical thickness and ground-truth surfaces for the pial and white matter. Figure 7 shows the Hausdorff distance between the reconstructed surface and the ground-truth. The relative error obtained in the respective morphological variable is shown in Figure 8 . The level of atrophy estimated by each reconstruction pipeline, including measurement with DiReCT instead of FreeSurfer, is presented in Figure 9 . For all segmentation methods studied, the Freesurfer definition of cortical thickness fails to reproduce the synthetic atrophy. The DiReCT methodology improves the atrophy estimation showing a better performance when combined with DeepSCAN segmentation. Download figure Open in new tab Figure 7: The Hausdorff distance calculated between the reconstructed surface and the ground-truth for the white matter surface (left) and the pial surface (right) using the (A) Freesurfer reconstruction and our fast surface pipeline based on (B) DeepSCAN and (C) FastsurferCNN to the synthetic ground truth surface. Download figure Open in new tab Figure 8: The relative error in the morphological variables derived from the surfaces reconstructed by (A) Freesurfer and the fast surface pipeline based on (B) DeepSCAN, (C) FastsurferCNN, using the synthetic ground-truth surface for comparison. Download figure Open in new tab Figure 9: Estimated vs induced atrophy in the synthetic phantom, using both DiReCT and Freesurfer methodology [ 54 ]. The black line represents the identity. Discussion There is evidence that DL-based methods are a promising approach for a fast whole-brain segmentation [ 25 – 28 ]. The results presented here extend previous work done in [ 28 ], not only adding information about the performance of another state of the art whole-brain segmentation model (DeepSCAN [ 27 ]) but also going beyond the voxel-based evaluation and dice coefficient comparison between the DL-based segmentation and Freesurfer’s silver standard. Using a methodology inspired by [ 28 ], we produced a fast surface reconstruction pipeline to be used with any parcellation of a structural MRI into cortical regions and WM. With a processing runtime of 11 minutes ( Figure 2 ), this pipeline reflects the minimum amount of time required to generate a surface to be further manipulated according to the desired application. The general overview presented in Figure 1 indicates that the surfaces obtained from the fast reconstruction pipeline are qualitatively equivalent to the ones obtained by Freesurfer. While there is no dissimilarity compared to Freesurfer in the (B) DeepSCAN based reconstructions, small deviations are apparent for (C) FastsurferCNN in the WM surface, only for the ABCD dataset. More significant differences are present for (D) QuickNAT in all reconstructions displayed. The inaccurate surface reconstructions are caused by a systematic underestimation of the WM volume by either missing regions and/or while delimiting the WM/GM boundary. Further examples are given in the supplementary material B and D. Quantitatively, some differences between methodologies are revealed in both whole-hemisphere and region-wise analysis. For the whole-hemisphere, Figure 3 and the slope coefficients shown in Table 5 demonstrate that the surfaces based on (B) DeepSCAN are the closest to (A) Freesurfer pseudo ground-truth for all tested datasets. This observation is maintained in the region-wise analysis presented in Figure 4 for which (B) DeepSCAN shows less deviation than the other models. More, in the test-retest analysis shown in Figure 5 , (B) DeepSCAN displays a level of reproducibility higher than both (A) Freesurfer and (C) FastsurferCNN, a result in agreement with the observation of [ 27 ]. The reconstructions based on (C) FastsurferCNN fail to reproduce Freesurfer in the ABCD dataset. As shown in the supplementary material A and D, these discrepancies are due to a systematic underestimation of the WM/GM boundary and are maintained even for the full Fastsurfer reconstruction. Although (C) FastsurferCNN metrics derived from the whole hemisphere are close to (A) Freesurfer pseudo ground-truth, region-wise there is significant differences. The reconstructions based on (D) QuickNAT have the most significant differences to the (A) Freesurfer pseudo ground-truth for all datasets due to an inaccurate WM/GM boundary delimitation. Based on the level of robustness across a highly heterogeneous test dataset and the remarkably accelerated runtime of 11 minutes, considering the usage of the fast surface reconstruction pipeline based on combination (B) DeepSCAN+DiReCT becomes realistic for potential future clinical routine applications like diagnostic support [ 14 – 17 ]. In this context, the significant correlation between the resulting surface metrics with the WM/GM contrast of the used MRI sequence (see Figure 6 and Table 4 , and already noticed in previous studies [ 57 , 58 ]), is of great importance. The structural MRI acquired for clinical purposes presents an inherent variability regarding not only subjects and MRI machines but also sequence parameters. These effects should be taken into account in longitudinal and cross sectional comparisons. This stands as one of the major challenges to overcome for clinical routine applications. The major difference in the morphometry derived from the fast surface reconstruction for both (B) DeepSCAN and (C) FastsurferCNN is observed for cortical thickness. As both pial and WM volumes reconstructed on the human dataset are in agreement with Freesurfer, meaning that they enclose similar regions, this suggests an instability on Freesurfer thickness estimation methodology, which is better delineated with the synthetic dataset. The synthetic dataset, as shown in Figures 7 and 8 revealed a general tendency of all reconstructions, including the silver standard Freesurfer, to underestimate the sizes of the pial and WM surface. Among the reconstructions, DeepSCAN showed a better performance in reconstructing the ground truth, presenting about 1.5-2 times less relative error than the others in terms of pial area, pial volume and WM area and nearly zero relative error in the WM volume estimation. The absence of outliers in the Hausdorff distance for the DeepSCAN pipeline, as illustrated in Figure 7 , even when analyzing synthetic atrophies exceeding 0.7 mm, indicates that no implausible surfaces were detected for this pipeline. This contrasts with the results observed for Freesurfer and Fastsurfer. It is important to mention that, even though the synthetic dataset is a powerful tool to benchmark the reconstruction methods, it must be considered only as a model of the evolution of the cortical atrophy neurodegenerative. Values of synthetic atrophy as large as 0.5 mm or even more are rather unrealistic. However, the robustness in the reconstruction of these highly implausible brains can be interpreted as a indication of robustness of the pipeline itself. In terms of cortical thickness estimation, the Freesurfer methodology is unable to reproduce the introduced atrophy. For this particular metric, DiReCT provided results closer to the ground truth: this is in agreement with previously published results [ 27 , 43 ]. In conclusion, our results suggests that, as long as the segmentation model is able provide a good WM delimitation, the fast surface reconstruction pipeline will reconstruct reliable surfaces with a 11 minute runtime. The methodology presented here is scalable to any segmentation map, opening the possibility of using updated versions of the current models, new DL-based segmentation, specialized models for contrast enhanced sequences or specific diseases [ 59 – 61 ]. In particular, contrast agnostic models such as [ 61 ] can be used to overcome the contrast dependency on the metrics. For the three state-of-the-art segmentation models examined in this work, the most robust pipeline across the human dataset and closer to the synthetic ground truth was based on (B) DeepSCAN segmentation and DiReCT estimation of cortical thickness. This reconstruction methodology stands out as a possibility to bridge the gap between research and clinical application with a reliable and robust morphometry tool. Data Availability All data and codes used in the present study are public available. Supplementary Information A Full Fastsurfer reconstruction The fast surface reconstructions based on (C) FastsurferCNN fail to reproduce Freesurfer in the ABCD dataset, see top row of Fig. 3 . To ensure this effect is not due to our simplification of the pipeline, we have run the full Fastsurfer pipeline in a sub-sample of 100 random subjects. Figure 10 shows that the same dissimilarity is observed in the full Fastsurfer pipeline. Download figure Open in new tab Figure 10: Comparison between the SBM of obtained using the full FastsurferCNN pipeline and the simplified pipeline with the standard (A) Freesurfer reconstruction. Additionally, the surface obtained by the fast methodology does not displays significant regional differences to the Full Fastsurfer as shown in Figure 11 . The median relative difference between the morphological variables derived from the full Fastsurfer and the fast reconstruction pipeline further indicates no effect on the final surface due to the pipeline simplification. Download figure Open in new tab Figure 11: The Median relative difference between the surface based morphological variables derived from the full Fastsurfer reonstruction and the fast reconstruction pipeline. B General overview of the fast surface reconstruction Download figure Open in new tab Figure 12: Example of the results obtained from the fast surface reconstruction pipeline based on (B) DeepSCAN, (C) FastsurferCNN and (D) QuickNAT segmentation in comparison with the standard (A) Freesurfer reconstruction. The raw MRI is displayed together with the reconstructed WM (blue) and pial surfaces (red) in coronal view for the same cross-section of the brain to facilitate a direct comparison between surfaces. Apart from occasional segmentation failures, the surfaces are similar except in the hippocampus, which is not enclosed inside the WM for (A) Freesurfer. The 3D pial surface showing the region labeling provided by each DL-based segmentation model allows a neuro-anatomical comparison with the DK Atlas. The subjects are identified for reproducibility and the anatomical visualization was kept consistent to facilitate comparison between reconstructions. We retained the same coloring and naming presented in [ 51 ]. C Table with the linear regression coefficients View this table: View inline View popup Download powerpoint Table 5: Slope and offset obtained by the linear regression between the Freesurfer’s metrics and the one obtained by the alternative methods (see Figure 3 ). D General overview of the WM segmentation As a supplement to Figure 3 , a general overview of the WM segmentation provided by the three DL-based methods is shown in Figures 13 to 15 . The representative subjects were chosen using the first (close to Freesurfer), second (median behavior) and third (far from Freesurfer) quartiles of the percentage difference between the WM volume reconstructed and Freesurfer’s silver standard. Download figure Open in new tab Figure 13: Examples of the WM segmentation provided by (D) QuickNAT. There is a systematic underestimation of the WM either missing regions and/or in the WM/GM boundary delimitation. This explains the most significant differences observed when comparing (D) QuickNAT with Freesurfer reconstruction, for all datasets. This systematic tendency to underestimate the WM volume can also be seen on (C) FastsurferCNN prediction for the ABCD dataset only. Figures 14 and 15 shows a general overview of the WM segmentation for both (B) DeepSCAN and (C) FastsurferCNN. Download figure Open in new tab Figure 14: Examples of the WM segmentation provided by (C) FastsurferCNN. Download figure Open in new tab Figure 15: Examples of the WM segmentation provided by both (B) DeepSCAN. Acknowledgements This work is supported by the Swiss National Science Foundation (SNF, grant 204593, “ScanOMetrics” project). C.R. and R.W. acknowledge other support by the SNF (grant CRSII5 180365, Swiss First project) and the NIH (grant R01NS107513, Enigma Parkinson). We are thankful to the research team and volunteers for participating in this research project. We also want to acknowledge the importance of the public data sharing initiative followed by the authors of ABDC, HCP, OASIS, Synthetic atrophy and Phantom of Bern studies. Without it, this work would not be possible. References [1]. ↵ Anders M. Dale , Bruce Fischl , and Martin I. Sereno . Cortical surface-based analysis: I. segmentation and surface reconstruction . NeuroImage , 9 ( 2 ): 179 – 194 , 1999 . OpenUrl CrossRef PubMed Web of Science [2]. Bruce Fischl , Martin I. Sereno , and Anders M. Dale . Cortical surface-based analysis: Ii: Inflation, flattening, and a surface-based coordinate system . NeuroImage , 9 ( 2 ): 195 – 207 , 1999 . OpenUrl CrossRef PubMed Web of Science [3]. ↵ John Ashburner and Karl J. Friston . Voxel-based morphometry—the methods . NeuroImage , 11 ( 6 ): 805 – 821 , 2000 . OpenUrl CrossRef PubMed Web of Science [4]. ↵ Masami Goto , Osamu Abe , Akifumi Hagiwara , Shohei Fujita , Koji Kamagata , Masaaki Hori , Shigeki Aoki , Takahiro Osada , Seiki Konishi , Yoshitaka Masutani , Hajime Sakamoto , Yasuaki Sakano , Shinsuke Kyogoku , and Hiroyuki Daida . Advantages of using both voxel– and surface-based morphometry in cortical morphology analysis: A review of various applications . Magnetic Resonance in Medical Sciences , 21 ( 1 ): 41 – 57 , 2022 . OpenUrl PubMed [5]. ↵ William D Penny , Karl J Friston , John T Ashburner , Stefan J Kiebel , and Thomas E Nichols , editors. Statistical parametric mapping: The analysis of functional brain images . Academic Press , San Diego, CA , November 2006 . [6]. ↵ Hongzhi Wang and Paul Yushkevich . Multi-atlas segmentation with joint label fusion and corrective learning—an open source implementation . Frontiers in Neuroinformatics , 7 , 2013 . [7]. ↵ Mark Jenkinson , Christian F Beckmann , Timothy E J Behrens , Mark W Woolrich , and Stephen M Smith . FSL . Neuroimage , 62 ( 2 ): 782 – 790 , August 2012 . OpenUrl CrossRef PubMed Web of Science [8]. ↵ Bruce Fischl , David H Salat , Evelina Busa , Marilyn Albert , Megan Dieterich , Christian Haselgrove , Andre van der Kouwe , Ron Killiany , David Kennedy , Shuna Klaveness , Albert Montillo , Nikos Makris , Bruce Rosen , and Anders M Dale . Whole brain segmentation . Neuron , 33 ( 3 ): 341 – 355 , January 2002 . OpenUrl CrossRef PubMed Web of Science [9]. ↵ David W Shattuck and Richard M Leahy . BrainSuite: an automated cortical surface identification tool . Med. Image Anal ., 6 ( 2 ): 129 – 142 , June 2002 . OpenUrl CrossRef PubMed Web of Science [10]. ↵ D MacDonald , N Kabani , D Avis , and A C Evans . Automated 3-D extraction of inner and outer surfaces of cerebral cortex from MRI . Neuroimage , 12 ( 3 ): 340 – 356 , September 2000 . OpenUrl CrossRef PubMed Web of Science [11]. ↵ Keith A Johnson , Nick C Fox , Reisa A Sperling , and William E Klunk . Brain imaging in alzheimer disease . Cold Spring Harb. Perspect. Med ., 2 ( 4 ): a006213 , April 2012 . [12]. A. Traboulsee , J.H. Simon , L. Stone , E. Fisher , D.E. Jones , A. Malhotra , S.D. Newsome , J. Oh , D.S. Reich , N. Richert , K. Rammohan , O. Khan , E.-W. Radue , C. Ford , J. Halper , and D. Li . Revised recommendations of the consortium of ms centers task force for a standardized mri protocol and clinical guidelines for the diagnosis and follow-up of multiple sclerosis . American Journal of Neuroradiology , 37 ( 3 ): 394 – 401 , 2016 . OpenUrl Abstract / FREE Full Text [13]. ↵ Jörg Wellmer , Carlos M. Quesada , Lars Rothe , Christian E. Elger , Christian G. Bien , and Horst Urbach . Proposal for a magnetic resonance imaging protocol for the detection of epileptogenic lesions at early outpatient stages . Epilepsia , 54 ( 11 ): 1977 – 1987 , 2013 . OpenUrl CrossRef PubMed [14]. ↵ Serena Verdi , Seyed Mostafa Kia , Keir XX Yong , Duygu Tosun , Jonathan M Schott , Andre F Marquand , and James H Cole . Revealing individual neuroanatomical heterogeneity in alzheimer disease using neuroanatomical normative modeling . Neurology , 100 ( 24 ): e2442 – e2453 , 2023 . OpenUrl PubMed [15]. Christian Rummel , Nedelina Slavova , Andrea Seiler , Eugenio Abela , Martinus Hauf , Yuliya Burren , Christian Weisstanner , Serge Vulliemoz , Margitta Seeck , Kaspar Schindler , et al. Personalized structural image analysis in patients with temporal lobe epilepsy . Scientific reports , 7 ( 1 ): 10883 , 2017 . OpenUrl PubMed [16]. Christian Rummel , Fabian Aschwanden , Richard McKinley , Franca Wagner , Anke Salmen , Andrew Chan , and Roland Wiest . A fully automated pipeline for normative atrophy in patients with neurodegenerative disease . Frontiers in neurology , 8 : 727 , 2018 . OpenUrl PubMed [17]. ↵ David Romascano , Michael Rebsamen , Piotr Radojewski , Timo Blattner , Richard McKinley , Roland Wiest , and Christian Rummel . Cortical thickness and grey-matter volume anomaly detection in individual mri scans: Comparison of two methods . NeuroImage: Clinical, page 103624 , 2024 . [18]. ↵ Yann LeCun , Yoshua Bengio , and Geoffrey Hinton . Deep learning . Nature , 521 ( 7553 ): 436 – 444 , May 2015 . OpenUrl CrossRef PubMed [19]. ↵ Alexandre de Brébisson and Giovanni Montana . Deep neural networks for anatomical brain segmentation . In 2015 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , pages 20 – 28 , 2015 . [20]. Pim Moeskops , Max A. Viergever , Adriënne M. Mendrik , Linda S. de Vries , Manon J. N. L. Benders , and Ivana Išgum . Automatic segmentation of mr brain images with a convolutional neural network . IEEE Transactions on Medical Imaging , 35 ( 5 ): 1252 – 1261 , 2016 . OpenUrl CrossRef PubMed [21]. Rachana Mehta and Keyur Rana . Evolution of singular value decomposition in recommendation systems: a review . International Journal of Business Intelligence and Data Mining , 14 ( 4 ): 528 – 547 , 2019 . OpenUrl [22]. Christian Wachinger , Martin Reuter , and Tassilo Klein . Deepnat: Deep convolutional neural network for segmenting neuroanatomy . NeuroImage , 170 : 434 – 445 , 2018 . Segmenting the Brain. OpenUrl PubMed [23]. Yuankai Huo , Zhoubing Xu , Yunxi Xiong , Katherine Aboud , Prasanna Parvathaneni , Shunxing Bao , Camilo Bermudez , Susan M. Resnick , Laurie E. Cutting , and Bennett A. Landman . 3d whole brain segmentation using spatially localized atlas network tiles . NeuroImage , 194 : 105 – 119 , 2019 . OpenUrl PubMed [24]. ↵ Dinggang Shen , Tianming Liu , Terry M. Peters , Lawrence H. Staib , Caroline Essert , Sean Zhou , Pew-Thian Yap , and Ali Khan Pierrick Coupé , Boris Mansencal , Michaël Clément , Rémi Giraud , Baudouin Denis de Senneville , Vinh-Thong Ta , Vincent Lepetit , and José V. Manjon . Assemblynet: A novel deep decision-making process for whole brain mri segmentation . In Dinggang Shen , Tianming Liu , Terry M. Peters , Lawrence H. Staib , Caroline Essert , Sean Zhou , Pew-Thian Yap , and Ali Khan , editors, Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 , pages 466 – 474 , Cham, 2019 . Springer International Publishing . [25]. ↵ Maxime Descoteaux Abhijit Guha Roy , Sailesh Conjeti , Debdoot Sheet , Amin Katouzian , Nassir Navab , and Christian Wachinger . Error corrective boosting for learning fully convolutional networks with limited data . In Maxime Descoteaux , Lena Maier-Hein , Alfred Franz , Pierre Jannin , D. Louis Collins , and Simon Duchesne , editors, Medical Im-age Computing and Computer Assisted Intervention – MICCAI 2017 , pages 231 – 239 , Cham, 2017 . Springer International Publishing . [26]. ↵ Abhijit Guha Roy , Sailesh Conjeti , Nassir Navab , and Christian Wachinger . Quicknat: A fully convolutional network for quick and accurate segmentation of neuroanatomy . NeuroImage , 186 : 713 – 727 , 2019 . OpenUrl CrossRef PubMed [27]. ↵ Michael Rebsamen , Christian Rummel , Mauricio Reyes , Roland Wiest , and Richard McKinley . Direct cortical thickness estimation using deep learning-based anatomy segmentation and cortex parcellation . Human Brain Mapping , 41 ( 17 ): 4804 – 4814 , 2020 . OpenUrl CrossRef PubMed [28]. ↵ Leonie Henschel , Sailesh Conjeti , Santiago Estrada , Kersten Diers , Bruce Fischl , and Martin Reuter . Fastsurfer – a fast and accurate deep learning based neuroimaging pipeline . NeuroImage , 219 : 117012 , 2020 . OpenUrl CrossRef PubMed [29]. ↵ Léo Lebrat , Rodrigo Santa Cruz , Frédéric de Gournay , Darren Fu , Pierrick Bourgeat , Jurgen Fripp , Clinton Fookes , and Olivier Salvado . Corticalflow: A diffeomorphic mesh deformation module for cortical surface reconstruction , 2022 . [30]. Qiang Ma , Emma C. Robinson , Bernhard Kainz , Daniel Rueckert , and Amir Alansary . Pialnn: A fast deep learning framework for cortical pial surface reconstruction , 2021 . [31]. ↵ Rodrigo Santa Cruz , Léo Lebrat , Darren Fu , Pierrick Bourgeat , Jurgen Fripp , Clinton Fookes , and Olivier Salvado . Corticalflow++: Boosting cortical surface reconstruction accuracy, regularity, and interoperability , 2022 . [32]. ↵ E Geuze , E Vermetten , and J D Bremner . MR-based in vivo hippocampal volumetrics: 1. review of methodologies currently employed . Molecular Psychiatry , 10 ( 2 ): 147 – 159 , February 2005 . OpenUrl CrossRef PubMed Web of Science [33]. ↵ Loretxu Bergouignan , Marie Chupin , Yvonne Czechowska , Serge Kinkingnéhun , Cédric Lemogne , Guillaume Le Bastard , Martin Lepage , Line Garnero , Olivier Colliot , and Philippe Fossati . Can voxel based morphometry, manual segmentation and automated segmentation equally detect hippocampal volume differences in acute depression? NeuroImage , 45 ( 1 ): 29 – 37 , 2009 . OpenUrl CrossRef PubMed Web of Science [34]. ↵ Michael Rebsamen , Yannick Suter , Roland Wiest , Mauricio Reyes , and Christian Rummel . Brain morphometry estimation: From hours to seconds using deep learning . Frontiers in Neurology , 11 , 2020 . [35]. ↵ B. Karacali and C. Davatzikos . Simulation of tissue atrophy using a topology preserving transformation model . IEEE Transactions on Medical Imaging , 25 ( 5 ): 649 – 652 , 2006 . OpenUrl CrossRef PubMed Web of Science [36]. P. Pieperhoff , M. Südmeyer , L. Hömke , K. Zilles , A. Schnitzler , and K. Amunts . Detection of structural changes of the human brain in longitudinally acquired mr images by deformation field morphometry: Methodological analysis, validation and application . NeuroImage , 43 ( 2 ): 269 – 287 , 2008 . OpenUrl PubMed [37]. ↵ S. Sharma , V. Noblet , F. Rousseau , F. Heitz , L. Rumbach , and J.-P. Armspach . Evaluation of brain atrophy estimation algorithms using simulated ground-truth data . Medical Image Analysis , 14 ( 3 ): 373 – 389 , 2010 . OpenUrl CrossRef PubMed [38]. ↵ M. Sonka Andrew D Castellano Smith , William R. Crum , Derek L G Hill , Neil A. Thacker , and Paul A. Bromiley . Biomechanical simulation of atrophy in mr images . In M. Sonka and J.M. Fitzpatrick , editors, Proceedings of SPIE – The International Society for Optical Engineering|Proc SPIE Int Soc Opt Eng , volume 5032, pages 481 – 490 , United States , 2003 . SPIE . Medical Imaging 2003: Image Processing; Conference date: 01-07-2003. [39]. O. Camara , M. Schweiger , R.I. Scahill , W.R. Crum , B.I. Sneller , J.A. Schnabel , G.R. Ridgway , D.M. Cash , D.L.G. Hill , and N.C. Fox . Phenomenological model of diffuse global and regional atrophy using finite-element methods . IEEE Transactions on Medical Imaging , 25 ( 11 ): 1417 – 1430 , 2006 . OpenUrl CrossRef PubMed [40]. ↵ Bishesh Khanal , Marco Lorenzi , Nicholas Ayache , and Xavier Pennec . A biophysical model of brain deformation to simulate and analyze longitudinal mris of patients with alzheimer’s disease . NeuroImage , 134 : 35 – 52 , 2016 . OpenUrl PubMed [41]. ↵ Kathleen E. Larson and Ipek Oguz . Synthetic atrophy for longitudinal surface-based cortical thickness measure-ment . In Ivana Išgum and Bennett A. Landman , editors, Medical Imaging 2021: Image Processing , volume 11596, page 115963K . International Society for Optics and Photonics, SPIE , 2021 . [42]. ↵ Jose Bernal , Sergi Valverde , Kaisar Kushibar , Mariano Cabezas , Arnau Oliver , Xavier Lladó , and The Alzheimer’s Disease Neuroimaging Initiative. Generating longitudinal atrophy evaluation datasets on brain magnetic resonance images using convolutional neural networks and segmentation priors . Neuroinformatics , 19 ( 3 ): 477 – 492 , Jul 2021 . OpenUrl PubMed [43]. ↵ Filip Rusak , Rodrigo Santa Cruz , Léo Lebrat , Ondrej Hlinka , Jurgen Fripp , Elliot Smith , Clinton Fookes , Andrew P. Bradley , and Pierrick Bourgeat . Quantifiable brain atrophy synthesis for benchmarking of cortical thickness estimation methods . Medical Image Analysis , 82 : 102576 , 2022 . [44]. ↵ B.J. Casey , Tariq Cannonier , May I. Conley , Alexandra O. Cohen , Deanna M. Barch , Mary M. Heitzeg , Mary E. Soules , Theresa Teslovich , Danielle V. Dellarco , Hugh Garavan , Catherine A. Orr , Tor D. Wager , Marie T. Banich , Nicole K. Speer , Matthew T. Sutherland , Michael C. Riedel , Anthony S. Dick , James M. Bjork , Kathleen M. Thomas , Bader Chaarani , Margie H. Mejia , Donald J. Hagler , M. Daniela Cornejo , Chelsea S. Sicat , Michael P. Harms , Nico U.F. Dosenbach , Monica Rosenberg , Eric Earl , Hauke Bartsch , Richard Watts , Jonathan R. Polimeni , Joshua M. Kuperman , Damien A. Fair , and Anders M. Dale . The adolescent brain cognitive development (abcd) study: Imaging acquisition across 21 sites . Developmental Cognitive Neuroscience , 32 : 43 – 54 , 2018 . The Adolescent Brain Cognitive Development (ABCD) Consortium: Rationale, Aims, and Assessment Strategy. OpenUrl CrossRef PubMed [45]. ↵ Matthew F. Glasser , Stamatios N. Sotiropoulos , J. Anthony Wilson , Timothy S. Coalson , Bruce Fischl , Jesper L. Andersson , Junqian Xu , Saad Jbabdi , Matthew Webster , Jonathan R. Polimeni , David C. Van Essen , and Mark Jenkinson . The minimal preprocessing pipelines for the human connectome project . NeuroImage , 80 : 105 – 124 , 2013 . Mapping the Connectome. OpenUrl CrossRef PubMed Web of Science [46]. ↵ Michael Rebsamen , Milena Capiglioni , Robert Hoepner , Anke Salmen , Roland Wiest , Piotr Radojewski , and Christian Rummel . Growing importance of brain morphometry analysis in the clinical routine: The hidden impact of mr sequence parameters . Journal of Neuroradiology , 51 ( 1 ): 5 – 9 , 2024 . OpenUrl PubMed [47]. ↵ Pamela J. LaMontagne , Sarah Keefe , Wallace Lauren , Chengjie Xiong , Elizabeth A. Grant , Krista L. Moul-der , John C. Morris , Tammie L.S. Benzinger , and Daniel S. Marcus . P3-083: Oasis-3: Longitudinal neu-roimaging, clinical, and cognitive dataset for normal aging and alzheimer’s disease . Alzheimer’s & Dementia , 14 ( 7S_Part_20 ): P1097 – P1097 , 2018 . OpenUrl [48]. ↵ Clifford R. Jack Jr. , Matt A. Bernstein , Nick C. Fox , Paul Thompson , Gene Alexander , Danielle Harvey , Bret Borowski , Paula J. Britson , Jennifer L. Whitwell , Chadwick Ward , Anders M. Dale , Joel P. Felmlee , Jeffrey L. Gunter , Derek L.G. Hill , Ron Killiany , Norbert Schuff , Sabrina Fox-Bosetti , Chen Lin , Colin Studholme , Charles S. DeCarli , Gunnar Krueger , Heidi A. Ward , Gregory J. Metzger , Katherine T. Scott , Richard Mallozzi , Daniel Blezek , Joshua Levy , Josef P. Debbins , Adam S. Fleisher , Marilyn Albert , Robert Green , George Bartzokis , Gary Glover , John Mugler , and Michael W. Weiner . The alzheimer’s disease neuroimaging initiative (adni): Mri methods . Journal of Magnetic Resonance Imaging , 27 ( 4 ): 685 – 691 , 2008 . OpenUrl CrossRef PubMed Web of Science [49]. ↵ Michael W. Weiner , Dallas P. Veitch , Paul S. Aisen , Laurel A. Beckett , Nigel J. Cairns , Robert C. Green , Danielle Harvey , Clifford R. Jack , William Jagust , John C. Morris , Ronald C. Petersen , Jennifer Salazar , Andrew J. Saykin , Leslie M. Shaw , Arthur W. Toga , and John Q. Trojanowski . The alzheimer’s disease neuroimaging initiative 3: Continued innovation for clinical trial improvement . Alzheimer’s & Dementia , 13 ( 5 ): 561 – 571 , 2017 . OpenUrl [51]. ↵ Filip Rusak , Rodrigo Santa Cruz , Elliot Smith , Jurgen Fripp , Clinton Fookes , Pierrick Bourgeat , and Andrew Bradley . Detail matters: High-frequency content for realistic synthetic mri generation . In David Svoboda , Ninon Burgos , Jelmer M. Wolterink , and Can Zhao , editors, Simulation and Synthesis in Medical Imaging , pages 3 – 13 , Cham, 2021 . Springer International Publishing . [51]. ↵ Rahul S. Desikan , Florent Ségonne , Bruce Fischl , Brian T. Quinn , Bradford C. Dickerson , Deborah Blacker , Randy L. Buckner , Anders M. Dale , R. Paul Maguire , Bradley T. Hyman , Marilyn S. Albert , and Ronald J. Killiany . An automated labeling system for subdividing the human cerebral cortex on mri scans into gyral based regions of interest . NeuroImage , 31 ( 3 ): 968 – 980 , 2006 . OpenUrl CrossRef PubMed Web of Science [52]. ↵ Christopher S. McCarthy , Avinash Ramprashad , Carlie Thompson , Jo-Anna Botti , Ioana L. Coman , and Wendy R. Kates . A comparison of FreeSurfer-generated data with and without manual intervention . Frontiers in Neuro-science , 9 , 2015 . [53]. ↵ Pierre-Louis Bazin and Dzung L. Pham . Topology correction of segmented medical images using a fast marching algorithm . Computer Methods and Programs in Biomedicine , 88 ( 2 ): 182 – 190 , 2007 . OpenUrl CrossRef PubMed [54]. ↵ Bruce Fischl and Anders M. Dale . Measuring the thickness of the human cerebral cortex from magnetic resonance images . Proceedings of the National Academy of Sciences , 97 ( 20 ): 11050 – 11055 , 2000 . OpenUrl Abstract / FREE Full Text [55]. ↵ Arno Klein and Jason Tourville . 101 labeled brain images and a consistent human cortical labeling protocol . Frontiers in Neuroscience , 6 , 2012 . [56]. ↵ Sandhitsu R. Das , Brian B. Avants , Murray Grossman , and James C. Gee . Registration based cortical thickness measurement . NeuroImage , 45 ( 3 ): 867 – 879 , 2009 . OpenUrl CrossRef PubMed Web of Science [57]. ↵ Sven Haller , Pavel Falkovskiy , Reto Meuli , Jean-Philippe Thiran , Gunnar Krueger , Karl-Olof Lovblad , Tobias Kober , Alexis Roche , and Bénédicte Marechal . Basic MR sequence parameters systematically bias automated brain volume estimation . Neuroradiology , 58 ( 11 ): 1153 – 1160 , November 2016 . OpenUrl CrossRef PubMed [58]. ↵ Michael Rebsamen , Milena Capiglioni , Robert Hoepner , Anke Salmen , Roland Wiest , Piotr Radojewski , and Christian Rummel . Growing importance of brain morphometry analysis in the clinical routine: The hidden impact of mr sequence parameters . Journal of Neuroradiology , 2023 . [59]. ↵ Michael Rebsamen , Richard McKinley , Piotr Radojewski , Maximilian Pistor , Christoph Friedli , Robert Hoepner , Anke Salmen , Andrew Chan , Mauricio Reyes , Franca Wagner , Roland Wiest , and Christian Rummel . Reliable brain morphometry from contrast-enhanced t1w-mri in patients with multiple sclerosis . Human Brain Mapping , 44 ( 3 ): 970 – 979 , 2023 . OpenUrl CrossRef PubMed [60]. Richard McKinley , Rik Wepfer , Fabian Aschwanden , Lorenz Grunder , Raphaela Muri , Christian Rummel , Rajeev Verma , Christian Weisstanner , Mauricio Reyes , Anke Salmen , et al. Simultaneous lesion and brain segmentation in multiple sclerosis using deep neural networks . Scientific reports , 11 ( 1 ): 1087 , 2021 . OpenUrl [61]. ↵ Benjamin Billot , Douglas N. Greve , Oula Puonti , Axel Thielscher , Koen Van Leemput , Bruce Fischl , Adrian V. Dalca , and Juan Eugenio Iglesias . Synthseg: Segmentation of brain mri scans of any contrast and resolution without retraining . Medical Image Analysis , 86 : 102789 , 2023 . OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted March 23, 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 MRI-based surface reconstruction and cortical thickness estimation of the human brain: Benchmarking deep-learning based morphometry tools 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 MRI-based surface reconstruction and cortical thickness estimation of the human brain: Benchmarking deep-learning based morphometry tools Victor B. B. Mello , Richard McKinley , Roland Wiest , Christian Rummel medRxiv 2025.03.22.25324274; doi: https://doi.org/10.1101/2025.03.22.25324274 Share This Article: Copy Citation Tools MRI-based surface reconstruction and cortical thickness estimation of the human brain: Benchmarking deep-learning based morphometry tools Victor B. B. Mello , Richard McKinley , Roland Wiest , Christian Rummel medRxiv 2025.03.22.25324274; doi: https://doi.org/10.1101/2025.03.22.25324274 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 (299) Cardiovascular Medicine (4423) Dentistry and Oral Medicine (443) Dermatology (382) Emergency Medicine (607) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1507) Epidemiology (15219) Forensic Medicine (30) Gastroenterology (1123) Genetic and Genomic Medicine (6587) Geriatric Medicine (667) Health Economics (997) Health Informatics (4524) Health Policy (1368) Health Systems and Quality Improvement (1612) Hematology (540) HIV/AIDS (1264) Infectious Diseases (except HIV/AIDS) (15910) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (145) Nephrology (667) Neurology (6588) Nursing (345) Nutrition (998) Obstetrics and Gynecology (1143) Occupational and Environmental Health (956) Oncology (3331) Ophthalmology (970) Orthopedics (369) Otolaryngology (420) Pain Medicine (435) Palliative Medicine (129) Pathology (663) Pediatrics (1690) Pharmacology and Therapeutics (691) Primary Care Research (710) Psychiatry and Clinical Psychology (5438) Public and Global Health (9218) Radiology and Imaging (2195) Rehabilitation Medicine and Physical Therapy (1369) Respiratory Medicine (1195) Rheumatology (593) Sexual and Reproductive Health (709) Sports Medicine (529) Surgery (709) Toxicology (99) Transplantation (289) 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:'9ff8201ba9e34807',t:'MTc3OTQxNDE1OA=='};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())}}}})();