Gillian N. Miller
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Gillian N. Miller has a total of 84 co-authors. Here are the top co-authors by number of shared publications:
- Alexander Li Cohen
Department of Neurology, Boston Children’s Hospital, Harvard Medical School, Boston, MA, United States
6 shared publications Alberto Castro Palacin
Department of Neurology, Boston Children's Hospital, Boston, MA, USA
2 shared publications- Shreya Tripathy
Department of Neurology, Boston Children’s Hospital, Harvard Medical School , Boston, MA 02115
2 shared publications - Simon Keith Warfield
Department of Radiology Boston Children's Hospital Boston Massachusetts USA
2 shared publications Aaron D. Boes
Departments of Pediatrics, Neurology, Psychiatry, Carver College of Medicine, University of Iowa, Iowa City, IA, United States
1 shared publication- Abdelhakim Ouaalam
Department of Radiology, Boston Children's Hospital Harvard Medical School Boston Massachusetts USA
1 shared publication
Latest Publications
- Open Access
- 10 June 2026
- Published by Springer Nature in Nature
- Vol. 655 (8123) , 769-779
A prognostic human brain network for diffuse midline glioma
Diffuse midline gliomas (DMGs) are near-universally lethal tumours of the childhood central nervous system 1,2 . In animal models, DMGs form brain-wide integrated networks through neuron-to-glioma synapses 3–6 and glioma-to-glioma gap junctional coupling 3 . This extensive connectivity robustly promotes the growth and invasion of DMG 3–9 and other glial malignancies 10–12 through paracrine mechanisms and direct neuron-to-glioma synapses. However, the organization and clinical implications of these connections in the living human brain remain to be elucidated. Here, we develop tumour network mapping to compute the brain-wide connectivity profile of DMG, defining a conserved brain network across pontine and thalamic DMG associated with patient short-term survival (DMG network). Tumour functional connectivity with the DMG network was independently predictive of patient overall survival across two external validation cohorts. Tumour growth mapped to DMG network-specific trajectories and peak in-network neurometabolic changes across development spatiotemporally aligned with the peak age incidence of DMG. Analyses of single-nucleus RNA sequencing data confirmed diverse synaptic gene enrichment in high-connectivity DMG. Strikingly, incidental surgical resection of high-connectivity thalamic DMG tissue conferred a significant survival advantage. Collectively, these data define a conserved and prognostically important brain network in children with DMG, consistent with the hypothesis that DMGs exploit otherwise healthy brain circuits to promote tumour growth.
- 10 June 2026
- Published by Springer Nature in Nature
- Vol. 655 (8123) , 769-779
Open AccessA prognostic human brain network for diffuse midline glioma
Diffuse midline gliomas (DMGs) are near-universally lethal tumours of the childhood central nervous system 1,2 . In animal models, DMGs form brain-wide integrated networks through neuron-to-glioma synapses 3–6 and glioma-to-glioma gap junctional coupling 3 . This extensive connectivity robustly promotes the growth and invasion of DMG 3–9 and other glial malignancies 10–12 through paracrine mechanisms and direct neuron-to-glioma synapses. However, the organization and clinical implications of these connections in the living human brain remain to be elucidated. Here, we develop tumour network mapping to compute the brain-wide connectivity profile of DMG, defining a conserved brain network across pontine and thalamic DMG associated with patient short-term survival (DMG network). Tumour functional connectivity with the DMG network was independently predictive of patient overall survival across two external validation cohorts. Tumour growth mapped to DMG network-specific trajectories and peak in-network neurometabolic changes across development spatiotemporally aligned with the peak age incidence of DMG. Analyses of single-nucleus RNA sequencing data confirmed diverse synaptic gene enrichment in high-connectivity DMG. Strikingly, incidental surgical resection of high-connectivity thalamic DMG tissue conferred a significant survival advantage. Collectively, these data define a conserved and prognostically important brain network in children with DMG, consistent with the hypothesis that DMGs exploit otherwise healthy brain circuits to promote tumour growth.
- Open Access
- 2 January 2026
- Published by Oxford University Press (OUP) in Brain Communications
- Vol. 8 (1)
Lesion network mapping of focal injury-related aggression finds two distinct network injury patterns
Aggressive behaviour is prevalent across neurodevelopmental and neuropsychiatric conditions and is related to poor outcomes. Yet, despite extensive neuroimaging studies, a consistent set of brain networks where dysfunction is consistently related to aggressive behaviour remains unclear. Many studies are correlational in nature, while causal studies, such a lesion-location-based studies, are often limited to injury of the frontal lobes. Here, we analyse 61 focal brain lesions, identified in the current medical literature, that are associated with new-onset aggression and related behaviours. Lesions were traced onto a standardized brain atlas and used as seeds for a functional connectivity analysis, leveraging resting-state data from 1000 healthy individuals. These maps, representing likely pre-injury connectivity, were grouped using a data-driven hierarchical clustering approach. Then, the lesion networks of the identified clusters were separately compared with the connectivity of 716 lesions causing other symptoms. This data-driven approach identified two distinct lesion subgroups that both appear to manifest aggression through the co-occurrence of disrupted functional connectivity to networks involved in emotional expression and cognitive control. Both ‘aggression networks’ demonstrated sensitivity and specificity when compared with lesions causing a wide variety of other neuropsychiatric symptoms. The first subgroup network involved connectivity to the anterior cingulate cortex and was correlated with the connectivity of lesions causing akinetic mutism. The second subgroup network involved connectivity to the ventromedial prefrontal cortex and contained a notable subset of cases (n = 25) that had reported criminal behaviour, supporting a role of self-control in this subgroup and implying separable influences towards criminality within our identified aggression networks. Alterations within the first group aligned with motivation networks, while the second group aligned with disinhibition networks, indicating these as the potential underlying factors in aggression in the two subgroups, respectively. This characterization was supported by previous work on the atrophy network mapping of behavioural variant frontotemporal dementia. Overall, this study suggests that aggressive behaviour may be more likely after injury to distinct brain networks, which may be related to distinct behavioural factors, and has implications for potential targeted therapeutic interventions such as focused neuromodulation.
- 2 January 2026
- Published by Oxford University Press (OUP) in Brain Communications
- Vol. 8 (1)
Open AccessLesion network mapping of focal injury-related aggression finds two distinct network injury patterns
Aggressive behaviour is prevalent across neurodevelopmental and neuropsychiatric conditions and is related to poor outcomes. Yet, despite extensive neuroimaging studies, a consistent set of brain networks where dysfunction is consistently related to aggressive behaviour remains unclear. Many studies are correlational in nature, while causal studies, such a lesion-location-based studies, are often limited to injury of the frontal lobes. Here, we analyse 61 focal brain lesions, identified in the current medical literature, that are associated with new-onset aggression and related behaviours. Lesions were traced onto a standardized brain atlas and used as seeds for a functional connectivity analysis, leveraging resting-state data from 1000 healthy individuals. These maps, representing likely pre-injury connectivity, were grouped using a data-driven hierarchical clustering approach. Then, the lesion networks of the identified clusters were separately compared with the connectivity of 716 lesions causing other symptoms. This data-driven approach identified two distinct lesion subgroups that both appear to manifest aggression through the co-occurrence of disrupted functional connectivity to networks involved in emotional expression and cognitive control. Both ‘aggression networks’ demonstrated sensitivity and specificity when compared with lesions causing a wide variety of other neuropsychiatric symptoms. The first subgroup network involved connectivity to the anterior cingulate cortex and was correlated with the connectivity of lesions causing akinetic mutism. The second subgroup network involved connectivity to the ventromedial prefrontal cortex and contained a notable subset of cases (n = 25) that had reported criminal behaviour, supporting a role of self-control in this subgroup and implying separable influences towards criminality within our identified aggression networks. Alterations within the first group aligned with motivation networks, while the second group aligned with disinhibition networks, indicating these as the potential underlying factors in aggression in the two subgroups, respectively. This characterization was supported by previous work on the atrophy network mapping of behavioural variant frontotemporal dementia. Overall, this study suggests that aggressive behaviour may be more likely after injury to distinct brain networks, which may be related to distinct behavioural factors, and has implications for potential targeted therapeutic interventions such as focused neuromodulation.
- Open Access
- 5 December 2025
- Published by MIT Press in Imaging Neuroscience
- Vol. 3
Comparison of brain normalization software and lesion compensation techniques in chronic perinatal stroke imaging
Neuroimaging research depends on registration, the alignment of patients’ brains to a template or standard space, to enable accurate comparisons across individuals. Decades of work have advanced our ability to accurately register typical brains, but registering atypical brains, such as those with injury or highly distorted anatomy, remains a challenge. In particular, registration of perinatal stroke imaging is often complicated by delayed injury identification, which results in imaging obtained during the chronic stage where secondary structural impacts are evident. While analyses in native space can be valuable for subject-specific investigations, group-level studies require registration to a common template space, which enables between-subject comparisons of lesion locations and their network correlates. Although many registration algorithms exist, as do various compensation techniques for focal lesions, it is unclear how effective they are when applied to the highly distorted anatomy often present in this chronic perinatal stroke imaging. Here, we quantitatively and qualitatively compared the performance of three registration algorithms (FNIRT, ANTs, EasyReg) in registering eleven variably distorted brains with perinatal stroke to a standard template using their default lesion-compensation techniques. We also assessed the impact of “brain grafting”, that is, inserting a healthy tissue mask in place of the defined lesion area prior to registration. Our findings show that ANTs and EasyReg are significantly more accurate than FNIRT for chronic perinatal stroke imaging, although all three software packages have marked difficulty with large lesions. Notably, brain grafting significantly improved the lesion mask normalization performance of FNIRT. In light of these comparisons, the recently released EasyReg appears to be an appropriate starting point for registering cohorts with chronic perinatal strokes, but we still emphasize the necessity of consistent visual inspection of registered brains.
- 5 December 2025
- Published by MIT Press in Imaging Neuroscience
- Vol. 3
Open AccessComparison of brain normalization software and lesion compensation techniques in chronic perinatal stroke imaging
Neuroimaging research depends on registration, the alignment of patients’ brains to a template or standard space, to enable accurate comparisons across individuals. Decades of work have advanced our ability to accurately register typical brains, but registering atypical brains, such as those with injury or highly distorted anatomy, remains a challenge. In particular, registration of perinatal stroke imaging is often complicated by delayed injury identification, which results in imaging obtained during the chronic stage where secondary structural impacts are evident. While analyses in native space can be valuable for subject-specific investigations, group-level studies require registration to a common template space, which enables between-subject comparisons of lesion locations and their network correlates. Although many registration algorithms exist, as do various compensation techniques for focal lesions, it is unclear how effective they are when applied to the highly distorted anatomy often present in this chronic perinatal stroke imaging. Here, we quantitatively and qualitatively compared the performance of three registration algorithms (FNIRT, ANTs, EasyReg) in registering eleven variably distorted brains with perinatal stroke to a standard template using their default lesion-compensation techniques. We also assessed the impact of “brain grafting”, that is, inserting a healthy tissue mask in place of the defined lesion area prior to registration. Our findings show that ANTs and EasyReg are significantly more accurate than FNIRT for chronic perinatal stroke imaging, although all three software packages have marked difficulty with large lesions. Notably, brain grafting significantly improved the lesion mask normalization performance of FNIRT. In light of these comparisons, the recently released EasyReg appears to be an appropriate starting point for registering cohorts with chronic perinatal strokes, but we still emphasize the necessity of consistent visual inspection of registered brains.
Convolutional neural networks for automatic tuber segmentation and quantification of tuber burden in tuberous sclerosis complex
Objective: This study was undertaken to develop a fully automated algorithm for tuber segmentation and quantification of tuber volume that performs similarly to the gold standard human neuroradiologist. Methods: We used brain magnetic resonance imaging (MRI) from patients with tuberous sclerosis complex (TSC) to train and validate a convolutional neural network (CNN), which was evaluated on segmentation with the Dice–Sørensen similarity coefficient (DSSC) and on tuber burden quantification with Spearman correlation coefficient against a neuroradiologist's gold standard in the test set. Results: We collected 263 MRIs from 196 patients (57% males) with median (25th percentile–75th percentile) age of 4.3 (3.0–10.1) years: 176 MRIs in the train set, 39 in the validation set, and 48 in the test set. The final model achieved in the test set a DSSC of .820 (95% confidence interval [CI] = .799–.840) in the whole brain and in the different lobes the following: .831 (95% CI = .804–.850) in left frontal, .827 (95% CI = .799–.853) in right frontal, .817 (95% CI = .779–.842) in left temporal, .834 (95% CI = .812–.849) in right temporal, .821 (95% CI = .783–.856) in left parietal, .840 (95% CI = .810–.865) in right parietal, .832 (95% CI = .808–.851) in left occipital, and .856 (95% CI = .838–.871) in right occipital. CNN tuber volume quantification nearly perfectly correlated (Spearman correlation coefficient) with the neuroradiologist's across the whole brain (.984, 95% CI = .971–.991) and in the different lobes: .966 (95% CI = .940–.981) in left frontal, .973 (95% CI = .952–.985) in right frontal, .936 (95% CI = .888–.964) in left temporal, .967 (95% CI = .942–.982) in right temporal, .989 (95% CI = .980–.994) in left parietal, .983 (95% CI = .970–.990) in right parietal, .992 (95% CI = .985–.995) in left occipital, and .982 (95% CI = .968–.990) in right occipital (all p < .00001). Significance: We generated, trained, validated, and made publicly available a CNN that achieves a near‐perfect correlation with a neuroradiologist gold standard quantification of tuber burden, allows for objective tuber segmentation, and increases rigor and reproducibility in TSC research across institutions.
Convolutional neural networks for automatic tuber segmentation and quantification of tuber burden in tuberous sclerosis complex
Objective: This study was undertaken to develop a fully automated algorithm for tuber segmentation and quantification of tuber volume that performs similarly to the gold standard human neuroradiologist. Methods: We used brain magnetic resonance imaging (MRI) from patients with tuberous sclerosis complex (TSC) to train and validate a convolutional neural network (CNN), which was evaluated on segmentation with the Dice–Sørensen similarity coefficient (DSSC) and on tuber burden quantification with Spearman correlation coefficient against a neuroradiologist's gold standard in the test set. Results: We collected 263 MRIs from 196 patients (57% males) with median (25th percentile–75th percentile) age of 4.3 (3.0–10.1) years: 176 MRIs in the train set, 39 in the validation set, and 48 in the test set. The final model achieved in the test set a DSSC of .820 (95% confidence interval [CI] = .799–.840) in the whole brain and in the different lobes the following: .831 (95% CI = .804–.850) in left frontal, .827 (95% CI = .799–.853) in right frontal, .817 (95% CI = .779–.842) in left temporal, .834 (95% CI = .812–.849) in right temporal, .821 (95% CI = .783–.856) in left parietal, .840 (95% CI = .810–.865) in right parietal, .832 (95% CI = .808–.851) in left occipital, and .856 (95% CI = .838–.871) in right occipital. CNN tuber volume quantification nearly perfectly correlated (Spearman correlation coefficient) with the neuroradiologist's across the whole brain (.984, 95% CI = .971–.991) and in the different lobes: .966 (95% CI = .940–.981) in left frontal, .973 (95% CI = .952–.985) in right frontal, .936 (95% CI = .888–.964) in left temporal, .967 (95% CI = .942–.982) in right temporal, .989 (95% CI = .980–.994) in left parietal, .983 (95% CI = .970–.990) in right parietal, .992 (95% CI = .985–.995) in left occipital, and .982 (95% CI = .968–.990) in right occipital (all p < .00001). Significance: We generated, trained, validated, and made publicly available a CNN that achieves a near‐perfect correlation with a neuroradiologist gold standard quantification of tuber burden, allows for objective tuber segmentation, and increases rigor and reproducibility in TSC research across institutions.
- Open Access
- 30 October 2025
- Published by Oxford University Press (OUP) in Brain Communications
- Vol. 7 (6) , fcaf463
Network localization of altered auditory and somatosensory sensitivity based on causal brain lesions
Sensory processing as a neurological construct is the perception and interpretation of sensory information from both the body and the environment. Disruptions to sensory processing adversely impact daily functioning. One type of disruption that is particularly interesting is altered sensory modulation, leading to hypersensitivity or hyposensitivity, which are common in many neurodevelopmental and psychiatric conditions such as autism spectrum disorder, attention-deficit/hyperactivity disorder, Alzheimer’s, and schizophrenia. Here we aim to identify modality-specific and cross-modality brain networks involved in altered auditory and somatosensory processing. A systematic review identified 61 patients with new-onset sensory alterations following focal brain injury. Lesions were traced and combined with resting-state data from 1000 healthy controls to generate normative lesion connectivity maps. The specificity of our cohort’s lesion-connectivity compared to lesions associated with 22 other neuropsychiatric symptoms was assessed with voxel-wise two-sample t-tests performed with the FSL Permutation Analysis of Linear Models tool (family-wise error P < 0.05). A conjunction analysis against lesions associated with hallucination was conducted by binarizing and multiplying one-sample T-test maps to identify common lesion-connections between the conditions. Modality-specific networks were connected to their unimodal cortices and the cerebellum. Overall, lesions associated with cross-modality sensitivity changes had convergent connections to the substantia nigra, medial orbitofrontal cortex, and cerebellum (Lobule V, medial Lobule VIIIa). Subgroup analysis by direction revealed that lesions causing decreased sensitivity were connected to lobule X, in addition to the aforementioned cerebellar regions, while those causing increased sensitivity were only connected to medial V and bilateral V. Regardless of directionality, 90% of lesions exhibited connections to bilateral Lobule V. Conjunction analysis with hallucinations revealed common lesion-connections to cerebellar vermis and frontal pole. Our analysis identified significant lesion-connections to the substantia nigra, medial orbitofrontal cortex, and cerebellum—highlighting these key regions in cross-modality sensory processing. These findings emphasize the role of higher cognitive functions in sensory integration and suggest potential targets for neuromodulation to improve sensory processing.
- 30 October 2025
- Published by Oxford University Press (OUP) in Brain Communications
- Vol. 7 (6) , fcaf463
Open AccessNetwork localization of altered auditory and somatosensory sensitivity based on causal brain lesions
Sensory processing as a neurological construct is the perception and interpretation of sensory information from both the body and the environment. Disruptions to sensory processing adversely impact daily functioning. One type of disruption that is particularly interesting is altered sensory modulation, leading to hypersensitivity or hyposensitivity, which are common in many neurodevelopmental and psychiatric conditions such as autism spectrum disorder, attention-deficit/hyperactivity disorder, Alzheimer’s, and schizophrenia. Here we aim to identify modality-specific and cross-modality brain networks involved in altered auditory and somatosensory processing. A systematic review identified 61 patients with new-onset sensory alterations following focal brain injury. Lesions were traced and combined with resting-state data from 1000 healthy controls to generate normative lesion connectivity maps. The specificity of our cohort’s lesion-connectivity compared to lesions associated with 22 other neuropsychiatric symptoms was assessed with voxel-wise two-sample t-tests performed with the FSL Permutation Analysis of Linear Models tool (family-wise error P < 0.05). A conjunction analysis against lesions associated with hallucination was conducted by binarizing and multiplying one-sample T-test maps to identify common lesion-connections between the conditions. Modality-specific networks were connected to their unimodal cortices and the cerebellum. Overall, lesions associated with cross-modality sensitivity changes had convergent connections to the substantia nigra, medial orbitofrontal cortex, and cerebellum (Lobule V, medial Lobule VIIIa). Subgroup analysis by direction revealed that lesions causing decreased sensitivity were connected to lobule X, in addition to the aforementioned cerebellar regions, while those causing increased sensitivity were only connected to medial V and bilateral V. Regardless of directionality, 90% of lesions exhibited connections to bilateral Lobule V. Conjunction analysis with hallucinations revealed common lesion-connections to cerebellar vermis and frontal pole. Our analysis identified significant lesion-connections to the substantia nigra, medial orbitofrontal cortex, and cerebellum—highlighting these key regions in cross-modality sensory processing. These findings emphasize the role of higher cognitive functions in sensory integration and suggest potential targets for neuromodulation to improve sensory processing.
- Open Access
- 11 March 2025
- Published by Wiley in Annals of the Child Neurology Society
- Vol. 3 (2) , 91-104
Coordinate Network Mapping of Focal Brain Volume Differences in ADHD Reveals Common Patterns That Lack Specificity: A Systematic Review
Objective: Attention‐deficit/hyperactivity disorder (ADHD) has been associated with decreased regional brain volume, yet no consistent localization has emerged across studies. This discrepancy has been attributed to ADHD's diagnostic heterogeneity; however, one alternative is that ADHD is associated with alterations of brain networks, not individual regions. To test this hypothesis, we compared a traditional anatomic likelihood estimate (ALE) with a “coordinate network mapping” (CNM) approach using data from 38 studies comparing regional brain volumes in ADHD versus healthy controls.Methods: We performed an ALE analysis, determining above‐chance convergence between experiments. Next, we calculated the overlap with the putamen and default mode network, defined a priori. We then applied CNM, generating connectivity maps for each study and statistically comparing these maps to identify common areas of connectivity across studies. Finally, we compared the network map of ADHD with several control groups of neuropsychiatric disorders and with randomly generated coordinates.Results: ALE identified no significant spatial convergence between experiments. We also found only limited spatial overlap with the default mode network and weak functional connectivity with the putamen. Conversely, CNM revealed that the heterogenous coordinates fell within a consistent brain network characterized by connectivity with the reward and cingulo‐opercular “action mode” networks. However, we could not differentiate this network from the CNM‐derived networks in control groups.Interpretation: Although this network is biologically plausible and consistent with ADHD symptoms, the findings suggest that this network is not specific to ADHD and may reflect large‐scale brain networks. Although this meta‐analysis adds to the literature on the neurobiology of ADHD, the nonspecific findings convey the importance of studying ADHD at the symptom level.
- 11 March 2025
- Published by Wiley in Annals of the Child Neurology Society
- Vol. 3 (2) , 91-104
Open AccessCoordinate Network Mapping of Focal Brain Volume Differences in ADHD Reveals Common Patterns That Lack Specificity: A Systematic Review
Objective: Attention‐deficit/hyperactivity disorder (ADHD) has been associated with decreased regional brain volume, yet no consistent localization has emerged across studies. This discrepancy has been attributed to ADHD's diagnostic heterogeneity; however, one alternative is that ADHD is associated with alterations of brain networks, not individual regions. To test this hypothesis, we compared a traditional anatomic likelihood estimate (ALE) with a “coordinate network mapping” (CNM) approach using data from 38 studies comparing regional brain volumes in ADHD versus healthy controls.Methods: We performed an ALE analysis, determining above‐chance convergence between experiments. Next, we calculated the overlap with the putamen and default mode network, defined a priori. We then applied CNM, generating connectivity maps for each study and statistically comparing these maps to identify common areas of connectivity across studies. Finally, we compared the network map of ADHD with several control groups of neuropsychiatric disorders and with randomly generated coordinates.Results: ALE identified no significant spatial convergence between experiments. We also found only limited spatial overlap with the default mode network and weak functional connectivity with the putamen. Conversely, CNM revealed that the heterogenous coordinates fell within a consistent brain network characterized by connectivity with the reward and cingulo‐opercular “action mode” networks. However, we could not differentiate this network from the CNM‐derived networks in control groups.Interpretation: Although this network is biologically plausible and consistent with ADHD symptoms, the findings suggest that this network is not specific to ADHD and may reflect large‐scale brain networks. Although this meta‐analysis adds to the literature on the neurobiology of ADHD, the nonspecific findings convey the importance of studying ADHD at the symptom level.