On a two-truths phenomenon in spectral graph clustering

Priebe, Carey E.; Park, Youngser; Vogelstein, Joshua T.; Conroy, John M.; Lyzinski, Vince; Tang, Minh; Athreya, Avanti; Cape, Joshua; Bridgeford, Eric

doi:10.1073/pnas.1814462116

Cited by 57 publications

(66 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, in functional connectomics, increasing data acquisitions from ~5 to 30 minutes/subject has been shown to dramatically increase reliability, even if combining data across differing fMRI scan types (e.g., task, rest) (104,105). Different solutions can be equally meaningful, with each capturing a distinct aspect of the data (109). Thus, a common within-sample strategy for validation is to demonstrate that identified neurosubtypes explain variation in measures other than the features employed for subtyping.…”

Section: Diagnostic Samplesmentioning

confidence: 99%

Toward Neurosubtypes in Autism

Hong

Vogelstein

Gozzi

et al. 2020

Biological Psychiatry

Self Cite

121

View full text Add to dashboard Cite

Section: Diagnostic Samplesmentioning

confidence: 99%

Toward Neurosubtypes in Autism

Hong

Vogelstein

Gozzi

et al. 2020

Biological Psychiatry

Self Cite

121

View full text Add to dashboard Cite

“…Different solutions can be equally meaningful, with each capturing a distinct aspect of the data (109). Thus, a common within-sample strategy for validation is to demonstrate that identified neurosubtypes explain variation in measures other than the features employed for subtyping.…”

Section: Validationmentioning

confidence: 99%

Towards Neurosubtypes in Autism

Hong¹,

Vogelstein²,

Gozzi³

et al. 2019

Preprint

View full text Add to dashboard Cite

There is a general consensus that substantial heterogeneity underlies the neurobiology in autism spectrum disorder (ASD). As such, it has become increasingly clear that a dissection of variation at the molecular-, cellular-, and system-level domains is a prerequisite for identifying biomarkers and developing more targeted therapeutic strategies in ASD. Advances in neuroimaging approaches to characterizing atypical brain patterns have recently motivated their application as viable tools to delineate more homogenous ASD subgroups at the level of brain structure and function -i.e., neurosubtyping. This review assesses and critically discusses the current datadriven neurosubtyping in ASD. It breaks this pursuit into key methodological steps: the selection of diagnostic samples, neuroimaging features, algorithm and validation approaches. For each step, we appraise the current literature in terms of progress, as well as remaining challenges and potential solutions. Convergence across findings is discussed and biological implications are highlighted. Although preliminary and with limited methodological overlap, results from this literature illustrate the feasibility of neurosubtyping. Across studies, there is general agreement that distinct neurosubtypes exist, but the exact number and their definitions vary depending on the specific features and approach utilized in a given study. Results also suggest the utility of subtypes in predicting symptom severity and diagnostic labels above and beyond group-average comparison designs. This review concludes with a discussion of future avenues towards a comprehensive understanding of the mechanisms underlying ASD heterogeneity.

show abstract

“…Since this map visualizes similarity between two regions in two atlases, the information provided by the Dice map can be used quantify which regions in a given atlas are most similar to regions in another atlas. This method has proven valuable for performing inference with parcellations lacking anatomical annotation, as it allows conclusions realized at the parcel level to be inferred at the anatomical level [43].…”

Section: Dice Coefficientmentioning

confidence: 99%

Standardizing Human Brain Parcellations

Myers

Arvapalli

Ramachandran

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

Using brain atlases to localize regions of interest is a required for making neuroscientifically valid statistical inferences. These atlases, represented in volumetric or surface coordinate spaces, can describe brain topology from a variety of perspectives. Although many human brain atlases have circulated the field over the past fifty years, limited effort has been devoted to their standardization and specification. The purpose of standardization and specification is to ensure consistency and transparency with respect to orientation, resolution, labeling scheme, file storage format, and coordinate space designation. Consequently, researchers are often confronted with limited knowledge about a given atlas's organization, which make analytic comparisons more difficult. To fill this gap, we consolidate an extensive selection of popular human brain atlases into a single, curated open-source library, where they are stored following a standardized protocol with accompanying metadata. We propose that this protocol serves as the basis for storing future atlases. To demonstrate the utility of storing and standardizing these atlases following a common protocol, we conduct an experiment using data from the Healthy Brain Network whereby we quantify the statistical dependence of each atlas label on several key phenotypic variables. The repository containing the atlases, the specification, as well as relevant transformation functions is available at https://neurodata.io/mriBackground & Summary Understanding the brain's organization is one of the key challenges in human neuroscience [1] and is critical for clinical translation [2]. Parcellation of the brain into functionally and structurally distinct regions has seen impressive advances in recent years [3], and has grown the field of network neuroscience [4,5]. Through a range of techniques such as clustering [6-9], multivariate decomposition [10,11] , gradient based connectivity [1,[12][13][14][15][16], and multimodal neuroimaging [1], parcellations have enabled fundamental insights into the brain's topological organization and network properties [17]. In turn, these properties have allowed researchers to investigate brain-behavioral associations with developmental [18,19], cognitive [20,21], and clinical phenotypes [22][23][24].More recently, researchers interested in understanding brain organization are presented with a variety of brain atlases that can be used to define nodes of network-based analyses [25]. While this variety is a boon to researchers, the use of different parcellations across studies makes assessing reproducibility of brain-behavior relationships difficult (e.g. comparing across parcellations with different organizations and numbers of nodes; [5]). Amalgamating multiple brain parcellations into a single, standardized, curated list would offer researchers a valuable resource for evaluating replication of neuroimaging studies.Some efforts to consolidate these atlases is already underway. For example, Nilearn is a popular Python package that provides machine-...

show abstract

On a two-truths phenomenon in spectral graph clustering

Cited by 57 publications

References 22 publications

Toward Neurosubtypes in Autism

Toward Neurosubtypes in Autism

Towards Neurosubtypes in Autism

Standardizing Human Brain Parcellations

Contact Info

Product

Resources

About