Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Proc SPIE Int Soc Opt Eng . Author manuscript; available in PMC: 2026 Apr 11. Published in final edited form as: Proc SPIE Int Soc Opt Eng. 2025 Apr 11;13406:1340608. doi: 10.1117/12.3045799 Search in PMC Search in PubMed View in NLM Catalog Add to search Approximate diffusion tractography from FLAIR MRI and anatomical context using recurrent neural networks Zhiyuan Li Zhiyuan Li a Dept. of Electrical and Computer Engineering, Vanderbilt University, Nashville, TN, USA Find articles by Zhiyuan Li a , Michael E Kim Michael E Kim b Dept. of Computer Science, Vanderbilt University, Nashville, TN, USA Find articles by Michael E Kim b , Tian Yu Tian Yu b Dept. of Computer Science, Vanderbilt University, Nashville, TN, USA Find articles by Tian Yu b , Praitayini Kanakaraj Praitayini Kanakaraj b Dept. of Computer Science, Vanderbilt University, Nashville, TN, USA Find articles by Praitayini Kanakaraj b , Tianyuan Yao Tianyuan Yao b Dept. of Computer Science, Vanderbilt University, Nashville, TN, USA Find articles by Tianyuan Yao b , Chenyu Gao Chenyu Gao a Dept. of Electrical and Computer Engineering, Vanderbilt University, Nashville, TN, USA Find articles by Chenyu Gao a , Susan M Resnick Susan M Resnick c Laboratory of Behavioral Neuroscience, National Institute on Aging, Baltimore, MD, USA Find articles by Susan M Resnick c , Lori L Beason-Held Lori L Beason-Held c Laboratory of Behavioral Neuroscience, National Institute on Aging, Baltimore, MD, USA Find articles by Lori L Beason-Held c , Mohamad Habes Mohamad Habes d Neuroimage Analytics Laboratory and Glenn Biggs Institute Neuroimaging Core, Glenn Biggs Institute for Neurodegenerative Diseases, University of Texas Health San Antonio, San Antonio, TX, USA Find articles by Mohamad Habes d , Leon Y Cai Leon Y Cai e Dept. of Internal Medicine, Johns Hopkins Bayview Medical Center, Baltimore, MD, USA Find articles by Leon Y Cai e , Bennett A Landman Bennett A Landman a Dept. of Electrical and Computer Engineering, Vanderbilt University, Nashville, TN, USA b Dept. of Computer Science, Vanderbilt University, Nashville, TN, USA f Dept. of Radiology & Radiological Sciences, Vanderbilt University Medical Center, Nashville, TN, USA Find articles by Bennett A Landman a, b, f Author information Article notes Copyright and License information a Dept. of Electrical and Computer Engineering, Vanderbilt University, Nashville, TN, USA b Dept. of Computer Science, Vanderbilt University, Nashville, TN, USA c Laboratory of Behavioral Neuroscience, National Institute on Aging, Baltimore, MD, USA d Neuroimage Analytics Laboratory and Glenn Biggs Institute Neuroimaging Core, Glenn Biggs Institute for Neurodegenerative Diseases, University of Texas Health San Antonio, San Antonio, TX, USA e Dept. of Internal Medicine, Johns Hopkins Bayview Medical Center, Baltimore, MD, USA f Dept. of Radiology & Radiological Sciences, Vanderbilt University Medical Center, Nashville, TN, USA Issue date 2025 Feb. PMC Copyright notice PMCID: PMC12726966 NIHMSID: NIHMS2049470 PMID: 41445918 The publisher's version of this article is available at Proc SPIE Int Soc Opt Eng Abstract Diffusion MRI (dMRI) tractography methods provide a valuable method for in-vivo estimation of whole-brain white matter pathways that are commonly assumed to rely on microstructure models derived from dMRI. However, recent pioneering works have demonstrated that the accuracy of white-matter measurements computed from T1-weighted (T1w) MRI tractography is on a similar level to scan-rescan variability in dMRI tractography. This revelation raises new questions about understanding tractography: Is it primarily a dMRI microstructural phenomenon, and how different can it be when estimated from other imaging modalities? In this study, we propose a framework to approximate tractography from fluid-attenuated inversion recovery (FLAIR) MRI and examine its performance compared to tractography based on diffusion and T1w MRI. We adapt the teacher-student recurrent neural network (RNN) model from existing work on T1w tractography. Additionally, we use brain segmentation maps as the anatomical context. We conduct white matter bundle analysis and compare various metrics with those from T1w tractography and the traditional dMRI tractography. FLAIR tractography achieved significant different performance compared to T1 tractography evaluated by Dice similarity coefficient (p=0.004) and bundle adjacency streamlines distance (p=0.012). An average absolute difference of 23% was observed in eight bundle shape measurements between FLAIR tractography and traditional dMRI tractography. Both qualitative and quantitative results suggest that tractography based on FLAIR MRI is feasible and underscore the need for comprehensive research to understand tractography in the broader context of multi-modality brain MRI. Keywords: Tractography, recurrent neural networks, FLAIR MRI, diffusion MRI, white matter bundles 1. INTRODUCTION Diffusion magnetic resonance imaging (dMRI) is a well-established in-vivo and non-invasive approach for studying the microstructure and connectivity of brain white matter 1 – 4 . Typically, dMRI scans acquire multiple volumes to measure the self-diffusion of water molecules in multiple directions, thereby revealing the structures of brain tissues, such as axon fiber populations, at the millimeter scale 5 . Over the last few decades, microstructure models have been developed to provide condensed representations of the high-dimensional raw dMRI signals and the underlying brain tissue properties, including the tensor model 6 and fiber orientation distribution 7 . These models have initiated the development of a core dMRI technique, streamline tractography, which has demonstrated widespread usage in providing insights into whole brain neural connections 8 , 9 and advancing the understanding of neurological health and disease development 10 – 12 . Despite the great success and potential of dMRI tractography, its real-world clinical use is limited by a major drawback of dMRI: the long acquisition time required for high-quality images. To robustly model the complex fiber populations within a voxel and facilitate more accurate downstream tractography representations, high angular resolution diffusion imaging (HARDI) 13 is required or highly preferred 14 , 15 , which is time-consuming and can therefore suffer from artifacts and noise 16 – 18 . Aside from researching more rapid HARDI acquisition techniques, another under-explored yet important research question is: What would be an affordable yet efficient alternative way to estimate tractography using information from other modalities of brain MRI? Recent pioneering work has demonstrated that tractography can be approximated by T1-weighted (T1w) MRI, achieving a performance level in a range similar to the scan-rescan variability in dMRI tractography 19 – 21 . These findings suggest that estimating tractography from modalities other than dMRI is possible and then call for advanced research efforts to gain a comprehensive understanding of tractography: Is it primarily a dMRI microstructural phenomenon, and how different can it be when estimated from other imaging modalities? To facilitate a fast and intuitive follow-up investigation, we propose a fluid-attenuated inversion recovery (FLAIR) tractography framework that adapts the convolutional recurrent neural networks proposed in T1-weighted tractography. We train a teacher-student model so that the dMRI features learned from the teacher model can guide the training of the student model. This enables the student model to independently approximate tractography from FLAIR MRI and its associated anatomical context. To better resemble clinical scans, we train and evaluate the proposed framework using a cohort of 15 subjects on the Baltimore Longitudinal Study of Aging (BLSA) dataset 22 . The results indicate that FLAIR tractography is significantly different from T1 tractography, as demonstrated by the worse Dice similarity coefficient and bundle adjacency streamlines distance. Additionally, an average absolute difference of 23% is observed in eight bundle shape measurements between FLAIR tractography and traditional dMRI tractography. Still, our findings suggest that FLAIR tractography is both possible and promising. We also discuss the implications and call for future research attentions in comprehensive understanding of whole-brain tractography. 2. METHOD 2.1. Dataset and Data preprocessing We randomly selected a cohort of 16 subjects from the Baltimore Longitudinal Study of Aging (BLSA) dataset, as in our previous study, for better resemblance to clinical scans 20 . Each subject had a group of paired diffusion MRI scans (T1-weighted MRI scans and FLAIR MRI scans) in each session. These scans were acquired on a 3T Philips scanner. The dMRI scans were acquired in 32 directions with a b-value of 700 s/mm 2 and voxel resolutions of 0.8125 x 0.8125 x 2.2 mm 3 . The T1-weighted and FLAIR scans were acquired with voxel resolutions of 1.2 x 1 x 1 mm 3 and 0.75 x 0.75 x 3 mm 3 , respectively. We divided this cohort into four groups: a training set of 5 subjects (8 sessions), a validation set comprising 1 subject (3 sessions), a testing set consisting of 9 subjects (18 sessions in total, including a pair of scan-rescan sessions for each subject), and finally, a single subject (1 session) used as a template for anatomical context in testing cases. The mean and standard deviation of the patients’ ages are shown in Table 1 . Table 1. Age distribution of patients included in this study. Sex Age Training 5 subjects with 8 sessions 2 females and 3 males 69.35 ± 7.76 Validation 1 subject with 3 sessions 1 male 88.13 ± 2.39 Testing 9 subjects with 18 sessions 1 female and 8 males 78.48 ± 7.85 Open in a new tab First, all dMRI scans were resampled to 0.8125 x 0.8125 x 0.8125 mm 3 and then preprocessed using PreQual 23 for denoising and correcting for motion, susceptibility, and eddy current-induced image artifacts. Quality check was performed on the PreQual reports and the preprocessed images to verify successful data preprocessing. Second, N4 bias correction 24 was applied for T1w images, which were then resampled to 1.0 x 1.0 x 1.0 mm 3 . Three groups of anatomical contexts were computed using the 1.0 x 1.0 x 1.0 mm 3 T1w images, including tissue-type masks using MRtrix3’s “5ttgen” command 25 , whole brain segmentation maps using SLANT 26 , and white matter bundle probability maps using white matter learning (WML) 27 , as in previous works 21 . All images in the 1.0 x 1.0 x 1.0 mm 3 T1w space were then registered to Montreal Neurological Institute (MNI) common space at 2.0 x 2.0 x 2.0 mm 3 resolution 28 , to prepare them as input for the student network. Finally, the FLAIR images were normalized by setting their 99.9th percentile as the maximum value and 0 as the minimum value and then were registered to the 2.0 x 2.0 x 2.0 mm 3 MNI space. To prepare the tractography streamlines for training and evaluations, fiber orientation distribution (FODs) was derived using constrained spherical deconvolution (CSD) 7 , 14 from the given dMRI images. Then, one million streamlines were computed from the FOD using the MRtrix3’s SDStream algorithm 25 . These streamlines were also registered to the 2mm MNI space and served as the ground truth streamlines for each session. During training and testing of the neural network, operation of sampling the grid using trilinear interpolation (SAMP) 21 was utilized for interpolating tractography features at any arbitrary location. 2.2. The teacher-student framework We adapt the idea of the teacher-student framework, as proposed in the T1w tractography method 21 , to encourage the neural network to approximate tractography from FLAIR images and their associated anatomical contexts ( Figure 1 ). Initially, a teacher model is trained on dMRI features to approximate traditional diffusion tractography. Subsequently, the learned features and weights of the teacher model are transferred to guide the training of the student model. By using the fixed parameters of the teacher model, the student model is encouraged to mimic the learned features of the teacher model, but by extracting them from FLAIR images and the associated anatomical contexts, instead of from dMRI used for the teacher model. Finally, the student model is optimized by minimizing the loss function between its approximated tractography and the ground truth tractography along with a regularization function that ensures the student model mimics the features learned by the teacher model. Figure 1. Open in a new tab In our proposed framework for approximating tractography from FLAIR MRI, first, a teacher model is trained on diffusion MRI with its derived fiber orientation distribution (FOD). Next, a student model is trained to learn and transform features from FLAIR MRI and its anatomical contexts to mimic the FOD features learned in teacher model. The weights of recurrent neural network that predicts the streamline directions is transferred and frozen from the teacher model to the student model. In implementation, we adapt the convolutional-recurrent neural network (CoRNN) as the backbone neural network architecture 21 . For training the teacher, first a multilayer perceptron (MLP) is applied to extract features from FOD information queried with SAMP at training streamline point locations. These features are then fed to two stacked gated-recurrent-unit (GRU) layers. The outputs of the GRU are then concatenated with its previous conditioned feature maps to predict a unit vector, d ¯ pred , that represents the direction of the next streamline point in spherical coordinates θ and φ, using another MLP layer. For encouraging the teacher model to accurately approximate the propagation of the ground truth streamlines, the cosine similarity loss is applied between the prediction and the ground truth label, d ¯ label as: ℒ cos T = 1 − 〈 d ¯ pred , d ¯ label 〉 ‖ d ¯ pred ‖ ‖ d ¯ label ‖ (1) Similarly, the student model follows the same training pipeline as the teacher model, except that instead of extracting features from FOD information, the student model extracts features from the FLAIR feature maps, which are previously computed from the FLAIR images and the associated anatomical contexts using a convolutional layer (CONV). To encourage the student model to mimic the features extracted by the teacher model—which will later be consistently utilized to predict the next streamline points—cosine similarity loss is also applied between the input features of the GRU as learned in both the teacher and the student models. Therefore, the final loss for the student model is as follows: ℒ cos S = ( 1 − 〈 d ¯ pred , d ¯ label 〉 ‖ d ¯ pred ‖ ‖ d ¯ label ‖ ) + ( 1 − 〈 f ¯ T , f ¯ S 〉 ‖ f ¯ T ‖ ‖ f ¯ S ‖ ) , (2) where f ¯ T and f ¯ S are the feature vector of each streamline point in the input layer of the GRU. During training, the anatomical contexts for the student model are the subject’s own contexts, computed from their own T1w image. During testing, the model will use the template’s anatomical contexts that are registered to the subject’s space. The whole framework is implemented on Ubuntu 20.04 with Python 3.9.17 and PyTorch 2.0.1. The training and evaluation are performed on an Nvidia RTX A6000. The Adam optimizer is used with a constant learning rate of 0.001, and training is stopped if there is no improvement in validation loss after 200 epochs. The model checkpoints with the lowest validation loss are used for testing and evaluation. 2.3. Analysis To evaluate the performance of FLAIR compared with other methods, we generate tractograms in four scenarios for each test subject. First, we apply the MRTrix3 SDStream tractography for the dMRI scan and rescan sessions, labeled as (1) dMRI scan SDStream and (2) dMRI rescan SDStream, respectively. Additionally, we produce (3) T1w tractography approximated tractograms and (4) FLAIR tractography approximated tractograms. Initially, this allows us to compare (1) and (2) to assess dMRI scan-rescan variability. Next, we compare (3) and (4) with (1) to evaluate how tractography approximated from other MRI modalities differs from traditional dMRI tractography. Finally, we compare (3) and (4) to determine if there are differences in tractography approximated by different MRI modalities other than dMRI. For each method’s tractogram, 39 white matter bundles are identified using RecoBundlesX 29 . We then investigate bundle-wise geometric agreement using metrics such as the bundle adjacency streamlines distance and the Dice similarity coefficient 30 – 32 Additionally, we examine the absolute percent difference for eight major bundle shape measurements, including volume, length, span, surface area, diameter, elongation, curl, and irregularity 32 . MI-Brain 33 is used to visualize streamlines with T1w or FLAIR images as the background reference. 3. RESULTS Through qualitative visualization of the whole brain streamlines ( Figure 2 ), we observe variations in tractograms generated from the four different tractography methods. Across all methods, the overall distribution of the whole-brain streamlines appears visually similar, demonstrating consistent coverage and agreement with the overall brain anatomical structure. However, a closer examination reveals that the streamlines approximated by T1w and FLAIR MRI lose some of the detailed bundle structures, especially at the middle part of the left ILF. Notably, the FLAIR MRI tractography produces a bundle that is visually larger than those approximated by T1w MRI and the reference dMRI scan, and it tends to produce more straight streamlines at the end of bundles. Figure 2. Open in a new tab Representative samples from different tractography methods: (a) a left view of whole brain streamlines, and (b) a left view of the left inferior longitudinal fasciculus (ILF_L). In general, the whole brain streamlines are visually very similar in terms of the overall distribution of the streamlines for all methods. Streamlines approximated by T1w and FLAIR MRI appear to lose some detailed bundle structures. The bundle approximated by FLAIR MRI is visually larger than both the T1w approximated bundle and the ground truth. Regarding geometric agreement, on average, T1w tractography maintains a level of accuracy similar to that observed in the dMRI scan-rescan variability, as demonstrated by comparable values of bundle adjacency streamlines distance and the Dice similarity coefficient. In contrast, FLAIR tractography exhibits significantly poorer performance, as evidenced by lower Dice similarity coefficients (p=0.004) and higher bundle adjacency streamlines distances (p=0.012) ( Figure 3 , left). The boxplots for each bundle further support the consistent performance of T1w tractography across almost all bundles ( Figure 3 , right). On the other hand, the performance of FLAIR tractography is notably inferior, evidenced by the large difference of median values. Interestingly, both T1w and FLAIR tractography appear to achieve smaller variations in geometric agreement metrics for most of the bundles than the dMRI scan-rescan variability. Figure 3. Open in a new tab Geometric agreement metrics for 39 white matter bundles. On average (left two figures), T1w tractography achieves a similar accuracy level within the range of dMRI scan-rescan variability, while FLAIR tractography performs significantly worse than the other methods. Bundle-wise plots (right two figures) corroborate these observations, as demonstrated by the consistent performance across almost all bundles. * represents p<=0.05 and *** represents p<=5E-4. The absolute percent difference varies across the eight major bundle shape measurements ( Figure 4 ). For measurements of length, span, curl, and irregularity, all methods achieved high agreement, as demonstrated by the small difference range from 5% to 15%. On the other hand, the differences are notably higher for measurements that are closely related to the size of the bundle, such as diameter, surface area, and elongation, which difference range from 15% to 30%. These differences further amplify the large disparities observed in volume measurements, which can up to 55%. Additionally, increasing error is observed from SDStream upon rescan to the T1w tractography to the FLAIR tractography on these four size-related measurements. Figure 4. Open in a new tab Absolute percent difference for eight major bundle shape measurements. The differences for some measurements, such as length and span, are small across all methods, while larger differences observed in surface area and diameter further amplify the significant disparities noted in the volume measurements. 4. DISCUSSION Our study provides further validation that T1w tractography is a reliable alternative for approximating diffusion tractography, as it consistently achieves accurate estimations with error level similar to that observed in dMRI scan-rescan variability. This finding was initially suggested in previous works of T1w tractography 21 , which were trained and evaluated using the high-quality HARDI dMRI data from the Human Connectome Project (HCP) dataset 34 . In this study, the experimental results suggest that T1w tractography is also reliable and achieves similar performance when trained and evaluated using the BLSA dataset, which better resembles clinical scans. These results underscore the potential of T1w tractography in future clinical settings, providing a viable option for facilities tractography analysis without access to dMRI scans. Although FLAIR tractography does not perform as well as T1w tractography, our experimental results show that FLAIR tractography is possible and promising. This is demonstrated by the highly similar visualization of the whole brain tractograms compared with both T1w tractography and the traditional dMRI tractography. Loss of detailed bundle structures is observed in both T1w and FLAIR tractography and is worth further investigation. Additionally, the FLAIR tractography achieves an average of 0.37 for Dice similarity coefficient and 5mm of bundle adjacency streamlines distance, with the Dice similarity coefficient being as large as 0.6 for some bundles. These results all suggest the FLAIR tractography can approximate reasonable results and is promising for future development. The difference of performance between FLAIR and T1w tractography can be caused by that the anatomical contexts is derived from T1w MRI only. We reason that a more compressive whole-brain segmentation maps derived from FLAIR or multi-modal MRI should be helpful to improve the performance of FLAIR tractography. Additionally, pathological patterns, such as brain tumors that may change the structural anatomy, can impact the performance of our method as confounding factors. We believe this presents an intriguing direction for future investigation. Our studies suggest that approximated tractography from modalities other than dMRI still exhibits differences from traditional dMRI, and different modalities may have varying capabilities for approximating diffusion tractography. This raises many open research questions. For example, what is the most critical information captured in brain MRI that facilitates the estimation of tractography? What would be the best method of integrating information from multi-modality brain MRI to accurately estimate tractography? We are excited about the significant potential contributions these research directions may offer and emphasize the urgent need for comprehensive research to understand diffusion tractography more fully. Continued exploration in this field could lead to enhanced diagnostic tools and more effective treatment strategies in neurology and related disciplines. 5. CONCLUSION The findings from this work underscore the viability of FLAIR MRI for approximating diffusion tractography, presenting a potential alternative to traditional dMRI methodologies. Our research reveals that tractography derived from FLAIR MRI is not only visually appealing but also achieves reasonable geometric agreement with traditional diffusion tractography. These results emphasize the potential of researching alternatives to traditional dMRI tractography through investigating multiple brain MRI modalities. ACKNOWLEDGEMENTS This research is supported by NSF CAREER 1452485, NIH 1R01EB017230, and the ADSP Phenotype Harmonization Consortium (ADSP-PHC) that is funded by NIA (U24 AG074855, U01 AG068057 and R01 AG059716). Mohamad Habes, PhD is supported in part by 1R01AG080821. This study was supported in part using the resources of the Advanced Computing Center for Research and Education (ACCRE) at Vanderbilt University, Nashville, TN (NIH S10OD023680). The imaging dataset(s) used for this research were obtained with the support of ImageVU, a research resource supported by the Vanderbilt Institute for Clinical and Translational Research (VICTR) and Vanderbilt University Medical Center institutional funding. The VICTR is funded by the National Center for Advancing Translational Sciences (NCATS) Clinical Translational Science Award (CTSA) Program, Award Number 5UL1TR002243-03. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. The BLSA is supported by the Intramural Research Program, National Institute on Aging, NIH. REFERENCES 1. Basser PJ, Mattiello J, and LeBihan D, “MR diffusion tensor spectroscopy and imaging,” Biophys J 66(1), 259–267, Elsevier; (1994). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Alexander DC et al. , “Imaging brain microstructure with diffusion MRI: practicality and applications,” NMR Biomed 32(4), e3841, Wiley Online Library; (2019). [ DOI ] [ PubMed ] [ Google Scholar ] 3. Basser PJ and Pierpaoli C, “Microstructural and physiological features of tissues elucidated by quantitative-diffusion-tensor MRI,” Journal of magnetic resonance 213(2), 560–570, Elsevier; (2011). [ DOI ] [ PubMed ] [ Google Scholar ] 4. Johansen-Berg H and Behrens TEJ, Diffusion MRI: from quantitative measurement to in vivo neuroanatomy, Academic Press; (2013). [ Google Scholar ] 5. Jones DK, Diffusion MRI: theory, methods, and application, Oxford University Press; (2010). [ Google Scholar ] 6. Pierpaoli C et al. , “Diffusion tensor MR imaging of the human brain.,” Radiology 201(3), 637–648 (1996). [ DOI ] [ PubMed ] [ Google Scholar ] 7. Toumier J-D et al. , “Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution,” Neuroimage 23(3), 1176–1185, Elsevier; (2004). [ DOI ] [ PubMed ] [ Google Scholar ] 8. Jeurissen B et al. , “Diffusion MRI fiber tractography of the brain,” NMR Biomed 32(4), e3785, Wiley Online Library; (2019). [ DOI ] [ PubMed ] [ Google Scholar ] 9. Yang JY-M et al. , “Diffusion MRI tractography for neurosurgery: the basics, current state, technical reliability and challenges,” Phys Med Biol 66(15), 15TR01, IOP Publishing; (2021). [ Google Scholar ] 10. Cetin-Karayumak S et al. , “White matter abnormalities across the lifespan of schizophrenia: a harmonized multi-site diffusion MRI study,” Mol Psychiatry 25(12), 3208–3219, Nature Publishing Group UK London; (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Harrison JR et al. , “Imaging Alzheimer’s genetic risk using diffusion MRI: A systematic review,” Neuroimage Clin 27, 102359, Elsevier; (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Schilling KG et al. , “Aging and white matter microstructure and macrostructure: a longitudinal multi-site diffusion MRI study of 1218 participants,” Brain Struct Funct 227(6), 2111–2125, Springer; (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Descoteaux M, “High angular resolution diffusion imaging (HARDI),” Wiley encyclopedia of electrical and electronics engineering, 1–25, Wiley Online Library; (1999). [ Google Scholar ] 14. Tournier J-D, Calamante F, and Connelly A, “Robust determination of the fibre orientation distribution in diffusion MRI: non-negativity constrained super-resolved spherical deconvolution,” Neuroimage 35(4), 1459–1472, Elsevier; (2007). [ DOI ] [ PubMed ] [ Google Scholar ] 15. Alexander DC and Seunarine KK, “Mathematics of crossing fibers,” Diffusion MRI: theory, methods, and application. Oxford, 451–464, Oxford University Press; New York: (2010). [ Google Scholar ] 16. Jones DK and Cercignani M, “Twenty-five pitfalls in the analysis of diffusion MRI data,” NMR Biomed 23(7), 803–820, Wiley Online Library; (2010). [ DOI ] [ PubMed ] [ Google Scholar ] 17. Elhabian S et al. , “Subject–motion correction in HARDI acquisitions: choices and consequences,” Front Neurol 5, 240, Frontiers Media SA; (2014). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Zhou X et al. , “Quantitative quality assurance in a multicenter HARDI clinical trial at 3 T,” Magn Reson Imaging 35, 81–90, Elsevier; (2017). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Cai LY et al. , “Tractography from T1-weighted MRI: empirically exploring the clinical viability of streamline propagation without diffusion MRI,” Imaging Neuroscience, MIT Press 255 Main Street, 9th Floor, Cambridge, Massachusetts 02142, USA~; … (2024). [ Google Scholar ] 20. Yu T et al. , “Tractography with T1-weighted MRI and associated anatomical constraints on clinical quality diffusion MRI,” in Medical Imaging 2024: Image Processing, 12926 pp. 576–582 (2024). [ Google Scholar ] 21. Cai LY et al. , “Convolutional-recurrent neural networks approximate diffusion tractography from T1-weighted MRI and associated anatomical context,” bioRxiv, Cold Spring Harbor Laboratory Preprints; (2023). [ Google Scholar ] 22. Shock NW, Normal human aging: The Baltimore longitudinal study of aging, US Department of Health and Human Services, Public Health Service, National~…; (1984). [ Google Scholar ] 23. Cai LY et al. , “PreQual: An automated pipeline for integrated preprocessing and quality assurance of diffusion weighted MRI images,” Magn Reson Med 86(1), 456–470, Wiley Online Library; (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Tustison NJ et al. , “N4ITK: improved N3 bias correction,” IEEE Trans Med Imaging 29(6), 1310–1320, IEEE; (2010). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Tournier J-D et al. , “MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation,” Neuroimage 202, 116137, Elsevier; (2019). [ DOI ] [ PubMed ] [ Google Scholar ] 26. Huo Y et al. , “3D whole brain segmentation using spatially localized atlas network tiles,” Neuroimage 194, 105–119, Elsevier; (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Yang Q et al. , “Learning white matter subject-specific segmentation from structural MRI,” Med Phys 49(4), 2502–2513, Wiley Online Library; (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Grabner G et al. , “Symmetric atlasing and model based segmentation: an application to the hippocampus in older adults,” in Medical Image Computing and Computer-Assisted Intervention–MICCAI 2006: 9th International Conference, Copenhagen, Denmark, October 1-6, 2006. Proceedings, Part II 9, pp. 58–66 (2006). [ Google Scholar ] 29. Garyfallidis E et al. , “Recognition of white matter bundles using local and global streamline-based registration and clustering,” Neuroimage 170, 283–295, Elsevier; (2018). [ DOI ] [ PubMed ] [ Google Scholar ] 30. Rheault F, “Analyse et reconstruction de faisceaux de la matière blanche,” Université de Sherbrooke Computer Science (2020). [ Google Scholar ] 31. Schilling KG et al. , “Tractography dissection variability: What happens when 42 groups dissect 14 white matter bundles on the same dataset?,” Neuroimage 243, 118502, Elsevier; (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Yeh F-C, “Shape analysis of the human association pathways,” Neuroimage 223, 117329, Elsevier; (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Rheault F et al. , “MI-Brain, a software to handle tractograms and perform interactive virtual dissection,” in Proceedings of the ISMRM Diffusion study group workshop, Lisbon; (2016). [ Google Scholar ] 34. Van Essen DC et al. , “The WU-Minn human connectome project: an overview,” Neuroimage 80, 62–79, Elsevier; (2013). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] ACTIONS View on publisher site PDF (800.9 KB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top