Chad William Farris
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Chad William Farris has a total of 193 co-authors. Here are the top co-authors by number of shared publications:
- Xin Zhang
Department of Mechanical Engineering Boston University Boston Massachusetts USA
4 shared publications Bruce Rosen
Harvard Medical School, Boston, Massachusetts, USA
3 shared publicationsJanine Barrett
Boston Medical Center, Boston, Massachusetts, USA
3 shared publicationsJesse Mez
Boston University Alzheimer's Disease Research Center, Boston, MA, USA
3 shared publicationsJoseph Sisto
Boston Medical Center, Boston, Massachusetts, USA
3 shared publications- Lena Václavů
Department of Radiology, CJ Gorter MRI Center, Leiden University Medical Center, Leiden, The Netherlands
3 shared publications
Latest Publications
- Open Access
Reliability of TRUST ASL for measuring cerebral blood flow and blood–brain barrier water exchange: a scan–rescan pilot study in healthy subjects and patients with epilepsy
Objective: Water exchange across the blood–brain barrier (BBB) sustains tissue homeostasis; BBB dysfunction can raise neuronal excitability and promote seizures. Noninvasive imaging of water exchange probes BBB function. T2l1lrelaxation underl1lspinl1ltagging (TRUST) arterial spin labeling (ASL) separates intravascular and extravascular labeled water. We tested TRUSTl1lASL feasibility and scan–rescan reliability. Methods: Twelve participants (5 controls, 7 with epilepsy) underwent two 3T MRI sessions 2–4 weeks apart, acquiring T1, FLAIR, and TRUSTl1lASL. At each epilepsy visit we recorded seizure frequency, last seizure details, and medications. ASL images were processed in MATLAB to quantify CBF and water exchange time. Maps were registered to T1 and normalized to MNI space via FreeSurfer. Scan–rescan reliability across gray matter and regions (temporal cortex including amygdala/hippocampus, frontal, occipitoparietal, basal ganglia) used intraclass correlation coefficients. Individual agreement was evaluated with Bland–Altman plots and withinl1lsubject coefficient of variation. Statistical analyses used R. Results: Scan-rescan reliability was good for CBF (ICC = 0.79, 95% CI [0.28 - 0.94]) and excellent for water exchange time (ICC = 0.97, 95%CI [0.90-0.99]) across grey matter. Regional analyses revealed lower reliability in the basal ganglia compared to other subregions. Bland-Altman analysis demonstrated strong agreement between sessions, with mean differences near zero for CBF (limits of agreement: −20 to 20 ml/100g/min), and water exchange time (mean differences 0.17s; limits of agreement: −0.44 to 0.79 s). Significance: Our findings show TRUSTl1lASL reliably measures BBB water exchange. Tex has high relative reliability (ICC) but substantial withinl1lsubject variability (wCV), limiting sensitivity for small individual longitudinal changes. Groupl1llevel reliability is strong, supporting use in longitudinal epilepsy studies. BBB water exchange is a promising noninvasive biomarker of BBB dysfunction in epilepsy. PLAIN LANGUAGE SUMMARY: The blood-brain barrier protects the brain from harmful substances while letting nutrients pass from the blood. In epilepsy, this barrier can become disrupted and may make seizures worse. We tested a noninvasive MRI method (no injections) to measure how well the blood and brain exchange substances. Our results show the technique can clearly visualize these exchanges and can be repeated reliably, offering a safe tool to study barrier function in people with epilepsy.
Open AccessReliability of TRUST ASL for measuring cerebral blood flow and blood–brain barrier water exchange: a scan–rescan pilot study in healthy subjects and patients with epilepsy
Objective: Water exchange across the blood–brain barrier (BBB) sustains tissue homeostasis; BBB dysfunction can raise neuronal excitability and promote seizures. Noninvasive imaging of water exchange probes BBB function. T2l1lrelaxation underl1lspinl1ltagging (TRUST) arterial spin labeling (ASL) separates intravascular and extravascular labeled water. We tested TRUSTl1lASL feasibility and scan–rescan reliability. Methods: Twelve participants (5 controls, 7 with epilepsy) underwent two 3T MRI sessions 2–4 weeks apart, acquiring T1, FLAIR, and TRUSTl1lASL. At each epilepsy visit we recorded seizure frequency, last seizure details, and medications. ASL images were processed in MATLAB to quantify CBF and water exchange time. Maps were registered to T1 and normalized to MNI space via FreeSurfer. Scan–rescan reliability across gray matter and regions (temporal cortex including amygdala/hippocampus, frontal, occipitoparietal, basal ganglia) used intraclass correlation coefficients. Individual agreement was evaluated with Bland–Altman plots and withinl1lsubject coefficient of variation. Statistical analyses used R. Results: Scan-rescan reliability was good for CBF (ICC = 0.79, 95% CI [0.28 - 0.94]) and excellent for water exchange time (ICC = 0.97, 95%CI [0.90-0.99]) across grey matter. Regional analyses revealed lower reliability in the basal ganglia compared to other subregions. Bland-Altman analysis demonstrated strong agreement between sessions, with mean differences near zero for CBF (limits of agreement: −20 to 20 ml/100g/min), and water exchange time (mean differences 0.17s; limits of agreement: −0.44 to 0.79 s). Significance: Our findings show TRUSTl1lASL reliably measures BBB water exchange. Tex has high relative reliability (ICC) but substantial withinl1lsubject variability (wCV), limiting sensitivity for small individual longitudinal changes. Groupl1llevel reliability is strong, supporting use in longitudinal epilepsy studies. BBB water exchange is a promising noninvasive biomarker of BBB dysfunction in epilepsy. PLAIN LANGUAGE SUMMARY: The blood-brain barrier protects the brain from harmful substances while letting nutrients pass from the blood. In epilepsy, this barrier can become disrupted and may make seizures worse. We tested a noninvasive MRI method (no injections) to measure how well the blood and brain exchange substances. Our results show the technique can clearly visualize these exchanges and can be repeated reliably, offering a safe tool to study barrier function in people with epilepsy.
- Open Access
- 1 April 2026
- Published by Frontiers Media SA in Frontiers in Artificial Intelligence
- Vol. 9, 1771088
Few-shot deployment of pretrained MRI transformers in brain imaging tasks
Introduction: Transformer-based deep learning has shown great potential in medical imaging, but its real-world applicability remains limited due to the scarcity of annotated data. This study aims to develop a practical framework for the few-shot deployment of pretrained MRI transformers across diverse brain imaging tasks. Methods: We employ a Masked Autoencoder (MAE) pretraining strategy on a large-scale, multi-cohort brain MRI dataset comprising over 31 million 2D slices to learn transferable representations. For classification tasks, a frozen MAE encoder with a lightweight linear head (MAE-classify) is used. For segmentation, we propose MAE-FUnet, a hybrid architecture that fuses pretrained MAE embeddings with multi-scale CNN features. Extensive evaluations are conducted on multiple datasets, including NACC, ADNI, OASIS, NFBS, SynthStrip, and MRBrainS18, under controlled few-shot settings. Results: The proposed framework achieves state-of-the-art performance in MRI sequence classification, reaching an accuracy of 99.24% with only 6,152 trainable parameters. For segmentation tasks, MAE-FUnet consistently outperforms strong baselines, achieving superior Dice and IoU scores across skull stripping and multi-class anatomical segmentation benchmarks. The model also demonstrates enhanced robustness and stability under data-limited conditions, with lower performance variance compared to competing methods. Discussion: These results highlight the effectiveness of pretrained MAE representations for few-shot medical imaging tasks. The proposed framework enables efficient, scalable, and adaptable deployment of transformer-based models in data-constrained clinical environments. The fusion of global transformer embeddings with local CNN features provides a generalizable design paradigm for a wide range of medical imaging applications.
- 1 April 2026
- Published by Frontiers Media SA in Frontiers in Artificial Intelligence
- Vol. 9, 1771088
Open AccessFew-shot deployment of pretrained MRI transformers in brain imaging tasks
Introduction: Transformer-based deep learning has shown great potential in medical imaging, but its real-world applicability remains limited due to the scarcity of annotated data. This study aims to develop a practical framework for the few-shot deployment of pretrained MRI transformers across diverse brain imaging tasks. Methods: We employ a Masked Autoencoder (MAE) pretraining strategy on a large-scale, multi-cohort brain MRI dataset comprising over 31 million 2D slices to learn transferable representations. For classification tasks, a frozen MAE encoder with a lightweight linear head (MAE-classify) is used. For segmentation, we propose MAE-FUnet, a hybrid architecture that fuses pretrained MAE embeddings with multi-scale CNN features. Extensive evaluations are conducted on multiple datasets, including NACC, ADNI, OASIS, NFBS, SynthStrip, and MRBrainS18, under controlled few-shot settings. Results: The proposed framework achieves state-of-the-art performance in MRI sequence classification, reaching an accuracy of 99.24% with only 6,152 trainable parameters. For segmentation tasks, MAE-FUnet consistently outperforms strong baselines, achieving superior Dice and IoU scores across skull stripping and multi-class anatomical segmentation benchmarks. The model also demonstrates enhanced robustness and stability under data-limited conditions, with lower performance variance compared to competing methods. Discussion: These results highlight the effectiveness of pretrained MAE representations for few-shot medical imaging tasks. The proposed framework enables efficient, scalable, and adaptable deployment of transformer-based models in data-constrained clinical environments. The fusion of global transformer embeddings with local CNN features provides a generalizable design paradigm for a wide range of medical imaging applications.
- 14 March 2026
- Published by Elsevier in Journal Of Neuroradiology
- Vol. 53 (3) , 101548
Arterial spin labeling MRI with multiple post-labeling delays reveals interictal hypo- and hyper-perfusion patterns in epilepsy: A proof of concept
- 14 March 2026
- Published by Elsevier in Journal Of Neuroradiology
- Vol. 53 (3) , 101548
Arterial spin labeling MRI with multiple post-labeling delays reveals interictal hypo- and hyper-perfusion patterns in epilepsy: A proof of concept
- Open Access
Cerebral Blood Flow with multi-delay Arterial Spin Labeling MRI Reveals Interictal Hypo- and Hyper-perfusion in epilepsy: A Proof of Concept
Background and Purpose: Interictal cerebral blood flow (CBF) may be useful for seizure focus localization. However, its accuracy is debatable. Studies suggested interictal hypoperfusion at the seizure focus, yet others suggested hyperperfusion. This study aims to investigate the patterns of interictal perfusion in epilepsy subjects compared to healthy controls using multiple labeling-delays arterial spin labeling MRI, and to explore the accuracy of perfusion estimation using single post-labeling delay arterial-spin labeling MRI by comparing it to multiple post-labeling delays arterial-spin labeling MRI. Materials and Methods: We analyzed CBF in 40 participants, 15 healthy (35.9 ± 9.5 years, 47% women) and 25 epilepsy (41.1 ± 10.5 years; 52% women). We acquired a multiple post-labeling delays arterial spin labeling MRI using Hadamard encoding, and structural MRI. Perfusion quantification was performed using in-house software and analyzed using surface-based method. Z-scores of CBF and its absolute value (|Z-score|) were calculated to evaluate abnormal perfusion. Results: Brain perfusion showed interictal hypo- and/or hyper-perfusion in epilepsy compared to healthy participants involving focal to whole brain alterations. Epilepsy subjects had higher |Z-score| in most cortical regions compared to the healthy group. CBF generated from single post-labeling delay arterial-spin labeling MRI correlated with ones from multiple post-labeling delays in most cortical regions, yet the level of correlation was affected by regional arterial transit time and the labeling scheme. In regions with shorter arterial transit time, correlation peaked at shorter post-labeling delay (1300ms); whereas in regions with longer arterial transit time, correlation peaked at longer post-labeling delay (2250ms). Conclusions: Our findings suggest that the interictal CBF may fluctuate between hypo- and hyper-perfusion, and the involvement of brain regions may extend well beyond the seizure focus. Additionally, while single-post-labeling delay is efficient and clinically feasible, the accuracy of its assessment of perfusion depends on brain regions and the labeling scheme. SUMMARY STATEMENT: We observed interictal hypo-/hyper-perfusion, possibly extending beyond the seizure focus. Caution is warranted when interpreting single-post-labeling-delay perfusion MRI in epilepsy, as its reliability varies by brain region and labeling scheme. Key Results: Epilepsy subjects showed larger fluctuation of CBF, yet the fluctuation may happen in either direction: hypo- or hyper-perfusion. The extent of fluctuation can be focal, regional, hemispheric, or global. CBF derived from single-PLD MRI largely correlates with the ones derived from more reliable multi-PLD method. However, the level of correlation varies spatially (i.e., brain regions) and temporally (i.e., different labeling and post-labeling schemes), and may be explained by regional variability of arterial transit times.
Open AccessCerebral Blood Flow with multi-delay Arterial Spin Labeling MRI Reveals Interictal Hypo- and Hyper-perfusion in epilepsy: A Proof of Concept
Background and Purpose: Interictal cerebral blood flow (CBF) may be useful for seizure focus localization. However, its accuracy is debatable. Studies suggested interictal hypoperfusion at the seizure focus, yet others suggested hyperperfusion. This study aims to investigate the patterns of interictal perfusion in epilepsy subjects compared to healthy controls using multiple labeling-delays arterial spin labeling MRI, and to explore the accuracy of perfusion estimation using single post-labeling delay arterial-spin labeling MRI by comparing it to multiple post-labeling delays arterial-spin labeling MRI. Materials and Methods: We analyzed CBF in 40 participants, 15 healthy (35.9 ± 9.5 years, 47% women) and 25 epilepsy (41.1 ± 10.5 years; 52% women). We acquired a multiple post-labeling delays arterial spin labeling MRI using Hadamard encoding, and structural MRI. Perfusion quantification was performed using in-house software and analyzed using surface-based method. Z-scores of CBF and its absolute value (|Z-score|) were calculated to evaluate abnormal perfusion. Results: Brain perfusion showed interictal hypo- and/or hyper-perfusion in epilepsy compared to healthy participants involving focal to whole brain alterations. Epilepsy subjects had higher |Z-score| in most cortical regions compared to the healthy group. CBF generated from single post-labeling delay arterial-spin labeling MRI correlated with ones from multiple post-labeling delays in most cortical regions, yet the level of correlation was affected by regional arterial transit time and the labeling scheme. In regions with shorter arterial transit time, correlation peaked at shorter post-labeling delay (1300ms); whereas in regions with longer arterial transit time, correlation peaked at longer post-labeling delay (2250ms). Conclusions: Our findings suggest that the interictal CBF may fluctuate between hypo- and hyper-perfusion, and the involvement of brain regions may extend well beyond the seizure focus. Additionally, while single-post-labeling delay is efficient and clinically feasible, the accuracy of its assessment of perfusion depends on brain regions and the labeling scheme. SUMMARY STATEMENT: We observed interictal hypo-/hyper-perfusion, possibly extending beyond the seizure focus. Caution is warranted when interpreting single-post-labeling-delay perfusion MRI in epilepsy, as its reliability varies by brain region and labeling scheme. Key Results: Epilepsy subjects showed larger fluctuation of CBF, yet the fluctuation may happen in either direction: hypo- or hyper-perfusion. The extent of fluctuation can be focal, regional, hemispheric, or global. CBF derived from single-PLD MRI largely correlates with the ones derived from more reliable multi-PLD method. However, the level of correlation varies spatially (i.e., brain regions) and temporally (i.e., different labeling and post-labeling schemes), and may be explained by regional variability of arterial transit times.
- Open Access
- 24 December 2025
- Published by Wiley in Alzheimer's & Dementia
- Vol. 21 (S2) , e100027
Sex differences in traumatic brain injury and later‐life MRI regional brain volume associations in the Framingham Heart Study
Background: Reductions in MRI brain volumes after traumatic brain injury (TBI) have been widely reported, but study samples have been highly selected, brain regions involved vary amongst studies, and sex differences have not been well studied. Method: In this cross‐sectional, community‐based analysis, Framingham Heart Study Generation‐2 cohort participants with a TBI history prior to a 1.5T MRI performed between 2005‐2008 were matched (1:4) on birth year and sex with controls without TBI. Information on TBIs was collected across the lifespan by comprehensive review of medical records, FHS health history updates and exams and self‐report. TBI occurrence and severity were operationalized using modified American Congress of Rehabilitation Medicine and Veterans Affairs/Department of Defense criteria respectively. MRI volumetric analysis using custom in‐house software produced volume measurements for the following supratentorial structures: total intracranial volume (TCV), total brain, total gray matter, total white matter, lateral ventricles, hippocampi, and gray and white matter volume in the frontal, parietal, occipital, and temporal lobes. We assessed TBI group differences in volumes, adjusting for TCV, sex, education, MRI age, APOE e4 status, and TBI‐MRI time differences. Sex‐stratified analyses were also performed. Analyses were false discovery rate corrected for multiple comparisons. Result: Among 795 participants (mean age at MRI: 67.78.6; 59% female), there were 159 participants with TBI (141 mild, 18 moderate‐severe) and 636 participants without TBI. Mean time from TBI to MRI was 11.68.9 years. Participants with TBI had significantly smaller hippocampal volumes than controls (β=‐0.15, p = 0.004) and there was a non‐significant trend towards smaller frontal gray matter volume (β=‐1.66, p = 0.056). Among females, but not males, TBI was significantly associated with smaller hippocampal (β=‐0.25, p < 0.0001), frontal gray matter (β=‐2.67, p = 0.007), and total cerebral gray matter (β=‐5.20, p = 0.030) volumes. Conclusion: In a community‐based cohort of older adults, TBI was associated with reductions in MRI hippocampal, frontal gray matter, and total cerebral gray matter volumes in females, but not males. Sex may portend different risk for later life atrophy after TBI, particularly in brain regions commonly affected by Alzheimer's disease and related dementias. Future work will evaluate TBI associations with longitudinal MRI trajectories.
- 24 December 2025
- Published by Wiley in Alzheimer's & Dementia
- Vol. 21 (S2) , e100027
Open AccessSex differences in traumatic brain injury and later‐life MRI regional brain volume associations in the Framingham Heart Study
Background: Reductions in MRI brain volumes after traumatic brain injury (TBI) have been widely reported, but study samples have been highly selected, brain regions involved vary amongst studies, and sex differences have not been well studied. Method: In this cross‐sectional, community‐based analysis, Framingham Heart Study Generation‐2 cohort participants with a TBI history prior to a 1.5T MRI performed between 2005‐2008 were matched (1:4) on birth year and sex with controls without TBI. Information on TBIs was collected across the lifespan by comprehensive review of medical records, FHS health history updates and exams and self‐report. TBI occurrence and severity were operationalized using modified American Congress of Rehabilitation Medicine and Veterans Affairs/Department of Defense criteria respectively. MRI volumetric analysis using custom in‐house software produced volume measurements for the following supratentorial structures: total intracranial volume (TCV), total brain, total gray matter, total white matter, lateral ventricles, hippocampi, and gray and white matter volume in the frontal, parietal, occipital, and temporal lobes. We assessed TBI group differences in volumes, adjusting for TCV, sex, education, MRI age, APOE e4 status, and TBI‐MRI time differences. Sex‐stratified analyses were also performed. Analyses were false discovery rate corrected for multiple comparisons. Result: Among 795 participants (mean age at MRI: 67.78.6; 59% female), there were 159 participants with TBI (141 mild, 18 moderate‐severe) and 636 participants without TBI. Mean time from TBI to MRI was 11.68.9 years. Participants with TBI had significantly smaller hippocampal volumes than controls (β=‐0.15, p = 0.004) and there was a non‐significant trend towards smaller frontal gray matter volume (β=‐1.66, p = 0.056). Among females, but not males, TBI was significantly associated with smaller hippocampal (β=‐0.25, p < 0.0001), frontal gray matter (β=‐2.67, p = 0.007), and total cerebral gray matter (β=‐5.20, p = 0.030) volumes. Conclusion: In a community‐based cohort of older adults, TBI was associated with reductions in MRI hippocampal, frontal gray matter, and total cerebral gray matter volumes in females, but not males. Sex may portend different risk for later life atrophy after TBI, particularly in brain regions commonly affected by Alzheimer's disease and related dementias. Future work will evaluate TBI associations with longitudinal MRI trajectories.
- Open Access
- 23 December 2025
- Published by Wiley in Alzheimer's & Dementia
- Vol. 21 (S8) , e110230
Sex differences in traumatic brain injury and later‐life MRI regional brain volume associations in the Framingham Heart Study
Background: Reductions in MRI brain volumes after traumatic brain injury (TBI) have been widely reported, but study samples have been highly selected, brain regions involved vary amongst studies, and sex differences have not been well studied. Method: In this cross‐sectional, community‐based analysis, Framingham Heart Study Generation‐2 cohort participants with a TBI history prior to a 1.5T MRI performed between 2005‐2008 were matched (1:4) on birth year and sex with controls without TBI. Information on TBIs was collected across the lifespan by comprehensive review of medical records, FHS health history updates and exams and self‐report. TBI occurrence and severity were operationalized using modified American Congress of Rehabilitation Medicine and Veterans Affairs/Department of Defense criteria respectively. MRI volumetric analysis using custom in‐house software produced volume measurements for the following supratentorial structures: total intracranial volume (TCV), total brain, total gray matter, total white matter, lateral ventricles, hippocampi, and gray and white matter volume in the frontal, parietal, occipital, and temporal lobes. We assessed TBI group differences in volumes, adjusting for TCV, sex, education, MRI age, APOE e4 status, and TBI‐MRI time differences. Sex‐stratified analyses were also performed. Analyses were false discovery rate corrected for multiple comparisons. Result: Among 795 participants (mean age at MRI: 67.78.6; 59% female), there were 159 participants with TBI (141 mild, 18 moderate‐severe) and 636 participants without TBI. Mean time from TBI to MRI was 11.68.9 years. Participants with TBI had significantly smaller hippocampal volumes than controls (β=‐0.15, p = 0.004) and there was a non‐significant trend towards smaller frontal gray matter volume (β=‐1.66, p = 0.056). Among females, but not males, TBI was significantly associated with smaller hippocampal (β=‐0.25, p < 0.0001), frontal gray matter (β=‐2.67, p = 0.007), and total cerebral gray matter (β=‐5.20, p = 0.030) volumes. Conclusion: In a community‐based cohort of older adults, TBI was associated with reductions in MRI hippocampal, frontal gray matter, and total cerebral gray matter volumes in females, but not males. Sex may portend different risk for later life atrophy after TBI, particularly in brain regions commonly affected by Alzheimer’s disease and related dementias. Future work will evaluate TBI associations with longitudinal MRI trajectories.
- 23 December 2025
- Published by Wiley in Alzheimer's & Dementia
- Vol. 21 (S8) , e110230
Open AccessSex differences in traumatic brain injury and later‐life MRI regional brain volume associations in the Framingham Heart Study
Background: Reductions in MRI brain volumes after traumatic brain injury (TBI) have been widely reported, but study samples have been highly selected, brain regions involved vary amongst studies, and sex differences have not been well studied. Method: In this cross‐sectional, community‐based analysis, Framingham Heart Study Generation‐2 cohort participants with a TBI history prior to a 1.5T MRI performed between 2005‐2008 were matched (1:4) on birth year and sex with controls without TBI. Information on TBIs was collected across the lifespan by comprehensive review of medical records, FHS health history updates and exams and self‐report. TBI occurrence and severity were operationalized using modified American Congress of Rehabilitation Medicine and Veterans Affairs/Department of Defense criteria respectively. MRI volumetric analysis using custom in‐house software produced volume measurements for the following supratentorial structures: total intracranial volume (TCV), total brain, total gray matter, total white matter, lateral ventricles, hippocampi, and gray and white matter volume in the frontal, parietal, occipital, and temporal lobes. We assessed TBI group differences in volumes, adjusting for TCV, sex, education, MRI age, APOE e4 status, and TBI‐MRI time differences. Sex‐stratified analyses were also performed. Analyses were false discovery rate corrected for multiple comparisons. Result: Among 795 participants (mean age at MRI: 67.78.6; 59% female), there were 159 participants with TBI (141 mild, 18 moderate‐severe) and 636 participants without TBI. Mean time from TBI to MRI was 11.68.9 years. Participants with TBI had significantly smaller hippocampal volumes than controls (β=‐0.15, p = 0.004) and there was a non‐significant trend towards smaller frontal gray matter volume (β=‐1.66, p = 0.056). Among females, but not males, TBI was significantly associated with smaller hippocampal (β=‐0.25, p < 0.0001), frontal gray matter (β=‐2.67, p = 0.007), and total cerebral gray matter (β=‐5.20, p = 0.030) volumes. Conclusion: In a community‐based cohort of older adults, TBI was associated with reductions in MRI hippocampal, frontal gray matter, and total cerebral gray matter volumes in females, but not males. Sex may portend different risk for later life atrophy after TBI, particularly in brain regions commonly affected by Alzheimer’s disease and related dementias. Future work will evaluate TBI associations with longitudinal MRI trajectories.
- Open Access
Few-Shot Deployment of Pretrained MRI Transformers in Brain Imaging Tasks
Machine learning using transformers has shown great potential in medical imaging, but its real-world applicability remains limited due to the scarcity of annotated data. In this study, we propose a practical framework for the few-shot deployment of pretrained MRI transformers in diverse brain imaging tasks. By utilizing the Masked Autoencoder (MAE) pretraining strategy on a large-scale, multi-cohort brain MRI dataset comprising over 31 million slices, we obtain highly transferable latent representations that generalize well across tasks and datasets. For high-level tasks such as classification, a frozen MAE encoder combined with a lightweight linear head achieves state-of-the-art accuracy in MRI sequence identification with minimal supervision. For low-level tasks such as segmentation, we propose MAE-FUnet, a hybrid architecture that fuses multiscale CNN features with pretrained MAE embeddings. This model consistently outperforms other strong baselines in both skull stripping and multi-class anatomical segmentation under data-limited conditions. With extensive quantitative and qualitative evaluations, our framework demonstrates efficiency, stability, and scalability, suggesting its suitability for low-resource clinical environments and broader neuroimaging applications.
Open AccessFew-Shot Deployment of Pretrained MRI Transformers in Brain Imaging Tasks
Machine learning using transformers has shown great potential in medical imaging, but its real-world applicability remains limited due to the scarcity of annotated data. In this study, we propose a practical framework for the few-shot deployment of pretrained MRI transformers in diverse brain imaging tasks. By utilizing the Masked Autoencoder (MAE) pretraining strategy on a large-scale, multi-cohort brain MRI dataset comprising over 31 million slices, we obtain highly transferable latent representations that generalize well across tasks and datasets. For high-level tasks such as classification, a frozen MAE encoder combined with a lightweight linear head achieves state-of-the-art accuracy in MRI sequence identification with minimal supervision. For low-level tasks such as segmentation, we propose MAE-FUnet, a hybrid architecture that fuses multiscale CNN features with pretrained MAE embeddings. This model consistently outperforms other strong baselines in both skull stripping and multi-class anatomical segmentation under data-limited conditions. With extensive quantitative and qualitative evaluations, our framework demonstrates efficiency, stability, and scalability, suggesting its suitability for low-resource clinical environments and broader neuroimaging applications.
- 21 May 2025
- Published by SAGE Publications in Interventional Neuroradiology
Endovascular treatment of sigmoid Sinus diverticulum in patients with pulsatile tinnitus
Background: Sigmoid sinus diverticula (SSD) are increasingly recognized among the most common causes of pulsatile tinnitus (PT). However, questions remain regarding their etiopathogenesis and optimal management. This study aims to review and compare management strategies and outcomes of SSD treated endovascularly in patients with PT. Methods: We retrospectively reviewed all patients with refractory PT who were found to have SSD and were treated endovascularly between September 2020 and January 2024. Clinical features, radiological findings, and endovascular strategies were analyzed. Outcomes were evaluated at 24 hours, 3 months, and 1 year. Results: A total of 26 SSDs were found in 23 patients. Diverticula were categorized as saccular in 76.9% (20/26) of cases and as fusiform in 23.1% (6/26) of cases. Their median dimensions were 5.1 mm [IQR 3.8–6.7 mm] in the transverse axis. Transverse sinus stenosis was present in all cases (26/26) with a median pre-procedural pressure gradient of 8 mmHg ([4–11 mmHg]; n = 23). Idiopathic intracranial hypertension was present in 65.4% (17/26) of cases. Ipsilateral sigmoid wall dehiscence was present in 84.6% (22/26) of cases. Stenting alone was performed in 14 cases (53.8%), stent-assisted coiling in 9 cases (34.6), and coiling alone was performed in 3 cases (11.5%). The primary outcome of treatment success, defined as complete resolution of PT at 3 months, occurred in 88.5% (23/26) of cases. Partial resolution of PT at 3 months occurred in an additional 11.5% (3/26) of cases. There was no difference in outcome between the different endovascular treatments at any time point (24 hours p = 0.58; 3 months p = 1.00; 1 year p = 1). Conclusion: Endovascular treatment is safe and effective in resolving PT in patients with sigmoid sinus diverticulum. Stenting alone was as effective as stent-assisted coiling, indicating that direct treatment of the diverticulum may not be necessary as long as the associated stenosis is addressed. Further studies are needed to validate these findings and to refine the therapeutic options in PT patients with venous sinus anomalies.
- 21 May 2025
- Published by SAGE Publications in Interventional Neuroradiology
Endovascular treatment of sigmoid Sinus diverticulum in patients with pulsatile tinnitus
Background: Sigmoid sinus diverticula (SSD) are increasingly recognized among the most common causes of pulsatile tinnitus (PT). However, questions remain regarding their etiopathogenesis and optimal management. This study aims to review and compare management strategies and outcomes of SSD treated endovascularly in patients with PT. Methods: We retrospectively reviewed all patients with refractory PT who were found to have SSD and were treated endovascularly between September 2020 and January 2024. Clinical features, radiological findings, and endovascular strategies were analyzed. Outcomes were evaluated at 24 hours, 3 months, and 1 year. Results: A total of 26 SSDs were found in 23 patients. Diverticula were categorized as saccular in 76.9% (20/26) of cases and as fusiform in 23.1% (6/26) of cases. Their median dimensions were 5.1 mm [IQR 3.8–6.7 mm] in the transverse axis. Transverse sinus stenosis was present in all cases (26/26) with a median pre-procedural pressure gradient of 8 mmHg ([4–11 mmHg]; n = 23). Idiopathic intracranial hypertension was present in 65.4% (17/26) of cases. Ipsilateral sigmoid wall dehiscence was present in 84.6% (22/26) of cases. Stenting alone was performed in 14 cases (53.8%), stent-assisted coiling in 9 cases (34.6), and coiling alone was performed in 3 cases (11.5%). The primary outcome of treatment success, defined as complete resolution of PT at 3 months, occurred in 88.5% (23/26) of cases. Partial resolution of PT at 3 months occurred in an additional 11.5% (3/26) of cases. There was no difference in outcome between the different endovascular treatments at any time point (24 hours p = 0.58; 3 months p = 1.00; 1 year p = 1). Conclusion: Endovascular treatment is safe and effective in resolving PT in patients with sigmoid sinus diverticulum. Stenting alone was as effective as stent-assisted coiling, indicating that direct treatment of the diverticulum may not be necessary as long as the associated stenosis is addressed. Further studies are needed to validate these findings and to refine the therapeutic options in PT patients with venous sinus anomalies.
- Top Cited PapersOpen Access
- 4 July 2024
- Published by Springer Nature in Nature Medicine
- Vol. 30 (10) , 2977-2989
AI-based differential diagnosis of dementia etiologies on multimodal data
Differential diagnosis of dementia remains a challenge in neurology due to symptom overlap across etiologies, yet it is crucial for formulating early, personalized management strategies. Here, we present an artificial intelligence (AI) model that harnesses a broad array of data, including demographics, individual and family medical history, medication use, neuropsychological assessments, functional evaluations and multimodal neuroimaging, to identify the etiologies contributing to dementia in individuals. The study, drawing on 51,269 participants across 9 independent, geographically diverse datasets, facilitated the identification of 10 distinct dementia etiologies. It aligns diagnoses with similar management strategies, ensuring robust predictions even with incomplete data. Our model achieved a microaveraged area under the receiver operating characteristic curve (AUROC) of 0.94 in classifying individuals with normal cognition, mild cognitive impairment and dementia. Also, the microaveraged AUROC was 0.96 in differentiating the dementia etiologies. Our model demonstrated proficiency in addressing mixed dementia cases, with a mean AUROC of 0.78 for two co-occurring pathologies. In a randomly selected subset of 100 cases, the AUROC of neurologist assessments augmented by our AI model exceeded neurologist-only evaluations by 26.25%. Furthermore, our model predictions aligned with biomarker evidence and its associations with different proteinopathies were substantiated through postmortem findings. Our framework has the potential to be integrated as a screening tool for dementia in clinical settings and drug trials. Further prospective studies are needed to confirm its ability to improve patient care.
- 4 July 2024
- Published by Springer Nature in Nature Medicine
- Vol. 30 (10) , 2977-2989
Top Cited PapersOpen AccessAI-based differential diagnosis of dementia etiologies on multimodal data
Differential diagnosis of dementia remains a challenge in neurology due to symptom overlap across etiologies, yet it is crucial for formulating early, personalized management strategies. Here, we present an artificial intelligence (AI) model that harnesses a broad array of data, including demographics, individual and family medical history, medication use, neuropsychological assessments, functional evaluations and multimodal neuroimaging, to identify the etiologies contributing to dementia in individuals. The study, drawing on 51,269 participants across 9 independent, geographically diverse datasets, facilitated the identification of 10 distinct dementia etiologies. It aligns diagnoses with similar management strategies, ensuring robust predictions even with incomplete data. Our model achieved a microaveraged area under the receiver operating characteristic curve (AUROC) of 0.94 in classifying individuals with normal cognition, mild cognitive impairment and dementia. Also, the microaveraged AUROC was 0.96 in differentiating the dementia etiologies. Our model demonstrated proficiency in addressing mixed dementia cases, with a mean AUROC of 0.78 for two co-occurring pathologies. In a randomly selected subset of 100 cases, the AUROC of neurologist assessments augmented by our AI model exceeded neurologist-only evaluations by 26.25%. Furthermore, our model predictions aligned with biomarker evidence and its associations with different proteinopathies were substantiated through postmortem findings. Our framework has the potential to be integrated as a screening tool for dementia in clinical settings and drug trials. Further prospective studies are needed to confirm its ability to improve patient care.
- Open Access
AI-based differential diagnosis of dementia etiologies on multimodal data
Differential diagnosis of dementia remains a challenge in neurology due to symptom overlap across etiologies, yet it is crucial for formulating early, personalized management strategies. Here, we present an AI model that harnesses a broad array of data, including demographics, individual and family medical history, medication use, neuropsychological assessments, functional evaluations, and multimodal neuroimaging, to identify the etiologies contributing to dementia in individuals. The study, drawing on 51, 269 participants across 9 independent, geographically diverse datasets, facilitated the identification of 10 distinct dementia etiologies. It aligns diagnoses with similar management strategies, ensuring robust predictions even with incomplete data. Our model achieved a micro-averaged area under the receiver operating characteristic curve (AUROC) of 0.94 in classifying individuals with normal cognition, mild cognitive impairment and dementia. Also, the micro-averaged AUROC was 0.96 in differentiating the dementia etiologies. Our model demonstrated proficiency in addressing mixed dementia cases, with a mean AUROC of 0.78 for two cooccurring pathologies. In a randomly selected subset of 100 cases, the AUROC of neurologist assessments augmented by our AI model exceeded neurologist-only evaluations by 26.25%. Furthermore, our model predictions aligned with biomarker evidence and its associations with different proteinopathies were substantiated through postmortem findings. Our framework has the potential to be integrated as a screening tool for dementia in various clinical settings and drug trials, with promising implications for person-level management.
Open AccessAI-based differential diagnosis of dementia etiologies on multimodal data
Differential diagnosis of dementia remains a challenge in neurology due to symptom overlap across etiologies, yet it is crucial for formulating early, personalized management strategies. Here, we present an AI model that harnesses a broad array of data, including demographics, individual and family medical history, medication use, neuropsychological assessments, functional evaluations, and multimodal neuroimaging, to identify the etiologies contributing to dementia in individuals. The study, drawing on 51, 269 participants across 9 independent, geographically diverse datasets, facilitated the identification of 10 distinct dementia etiologies. It aligns diagnoses with similar management strategies, ensuring robust predictions even with incomplete data. Our model achieved a micro-averaged area under the receiver operating characteristic curve (AUROC) of 0.94 in classifying individuals with normal cognition, mild cognitive impairment and dementia. Also, the micro-averaged AUROC was 0.96 in differentiating the dementia etiologies. Our model demonstrated proficiency in addressing mixed dementia cases, with a mean AUROC of 0.78 for two cooccurring pathologies. In a randomly selected subset of 100 cases, the AUROC of neurologist assessments augmented by our AI model exceeded neurologist-only evaluations by 26.25%. Furthermore, our model predictions aligned with biomarker evidence and its associations with different proteinopathies were substantiated through postmortem findings. Our framework has the potential to be integrated as a screening tool for dementia in various clinical settings and drug trials, with promising implications for person-level management.