Deep Bayesian networks for uncertainty estimation and adversarial resistance of white matter hyperintensity segmentation

Forooshani, Parisa Mojiri; Biparva, Mahdi; Ntiri, Emmanuel E.; Ramı́rez, Joel; Boone, Lyndon; Holmes, Melissa F.; Adamo, Sabrina; Gao, Fuqiang; Ozzoude, Miracle; Scott, Christopher J.; Dowlatshahi, Dar; Lawrence‐Dewar, Jane M.; Kwan, Donna; Lang, Anthony E.; Marcotte, Karine; Léonard, Carol; Rochon, Elizabeth; Heyn, Chris; Bartha, Robert; Strother, Stephen C.; Tardif, Jean‐Claude; Symons, Sean; Masellis, Mario; Swartz, Richard H.; Moody, Alan R.; Black, Sandra E.; Goubran, Maged

doi:10.1002/hbm.25784

Cited by 30 publications

(38 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This pipeline outperforms other automated WMH segmentation procedures in quantifying deep and small WMH. 10 For illustrations of white matter lesions, WMH masks were co-registered to MNI standard space using the FLIRT tool (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/ FLIRT) in FSL6.0.1. Subsequently, a WMH probability map was created for baseline data of carriers to visualize the spatial pattern of lesions.…”

Section: Image Acquisition and Processingmentioning

confidence: 99%

Progressive White Matter Injury in Preclinical Dutch Cerebral Amyloid Angiopathy

Shirzadi

Yau

Schultz

et al. 2022

Annals of Neurology

Self Cite

View full text Add to dashboard Cite

Autosomal‐dominant, Dutch‐type cerebral amyloid angiopathy (D‐CAA) offers a unique opportunity to develop biomarkers for pre‐symptomatic cerebral amyloid angiopathy (CAA). We hypothesized that neuroimaging measures of white matter injury would be present and progressive in D‐CAA prior to hemorrhagic lesions or symptomatic hemorrhage. In a longitudinal cohort of D‐CAA carriers and non‐carriers, we observed divergence of white matter injury measures between D‐CAA carriers and non‐carriers prior to the appearance of cerebral microbleeds and >14 years before the average age of first symptomatic hemorrhage. These results indicate that white matter disruption measures may be valuable cross‐sectional and longitudinal biomarkers of D‐CAA progression. ANN NEUROL 2022;92:358–363

show abstract

Section: Image Acquisition and Processingmentioning

confidence: 99%

Progressive White Matter Injury in Preclinical Dutch Cerebral Amyloid Angiopathy

Shirzadi

Yau

Schultz

et al. 2022

Annals of Neurology

Self Cite

View full text Add to dashboard Cite

show abstract

“…WMH were delineated based on our automated segmentation tool ‘HyperMapper’ that uses T1, FLAIR, and Bayesian 3D Convolutional Neuronal Networks (CNN) [33]. Subcortical lacunar infarcts were masked before WMH segmentation [34].…”

Section: Methodsmentioning

confidence: 99%

Vascular burden and cognition: Mediating roles of neurodegeneration and amyloid-PET

Ottoy

Ozzoude

Zukotynski

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

INTRODUCTION: It remains unclear to which extent vascular burden promotes neurodegeneration and cognitive dysfunction in a cohort spanning low-to-severe small vessel disease (SVD) and amyloid-beta pathology. METHODS: In 120 subjects, we investigated 1) whether vascular burden, quantified as total or lobar white matter hyperintensity (WMH) volumes, is associated with different cognitive domains; and 2) whether the total WMH effect on cognition is mediated by amyloid (18F-AV45-PET), glucose metabolism (18F-FDG-PET), and/or cortical atrophy. RESULTS: Increased total WMH volume was associated with poorer performance in all cognitive domains tested, with the strongest effects observed for semantic fluency. These relationships were mediated mainly through cortical atrophy, particularly in the temporal lobe, and to a lesser extent through amyloid and metabolism. WMH volumes differentially impacted cognition depending on lobar location and amyloid status. DISCUSSION: Our study suggests mainly an amyloid-dependent pathway in which vascular burden affects cognitive impairment through temporal lobe atrophy.

show abstract

“…FLAIR contrast significantly improves the detection of WMHs over T1-weighted scans (Bakshi et al, 2001). The datasets for these tasks have been described in detail in previously published work Ntiri et al, 2021;Mojiri Forooshani et al, 2022). Each dataset comprises patient scans from multiple studies.…”

Section: Tasks and Datasetsmentioning

confidence: 99%

ROOD-MRI: Benchmarking the robustness of deep learning segmentation models to out-of-distribution and corrupted data in MRI

Boone¹,

Biparva²,

Forooshani³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Deep artificial neural networks (DNNs) have moved to the forefront of medical image analysis due to their success in classification, segmentation, and detection challenges. A principal challenge in large-scale deployment of DNNs in neuroimage analysis is the potential for shifts in signal-to-noise ratio, contrast, resolution, and presence of artifacts from site to site due to variances in scanners and acquisition protocols. DNNs are famously susceptible to these distribution shifts in computer vision. Currently, there are no benchmarking platforms or frameworks to assess the robustness of new and existing models to specific distribution shifts in MRI, and accessible multi-site benchmarking datasets are still scarce or task-specific. To address these limitations, we propose ROOD-MRI: a novel platform for benchmarking the Robustness of DNNs to Out-Of-Distribution (OOD) data, corruptions, and artifacts in MRI. This flexible platform provides modules for generating benchmarking datasets using transforms that model distribution shifts in MRI, implementations of newly derived benchmarking metrics for image segmentation, and examples for using the methodology with new models and tasks. We apply our methodology to hippocampus, ventricle, and white matter hyperintensity segmentation in several large studies, providing the hippocampus dataset as a publicly available benchmark. By evaluating modern DNNs on these datasets, we demonstrate that they are highly susceptible to distribution shifts and corruptions in MRI. We show that while data augmentation strategies can substantially improve robustness to OOD data for anatomical segmentation tasks, modern DNNs using augmentation still lack robustness in more challenging lesion-based segmentation tasks. We finally benchmark U-Nets and transformer-based models, finding consistent differences in robustness to particular classes of transforms across architectures. The presented open-source platform enables generating new benchmarking datasets and comparing across models to study model design that results in improved robustness to OOD data and corruptions in MRI.

show abstract

Deep Bayesian networks for uncertainty estimation and adversarial resistance of white matter hyperintensity segmentation

Abstract: White matter hyperintensities (WMHs) are frequently observed on structural neuroimaging of elderly populations and are associated with cognitive decline and Sandra E. Black and Maged Goubran are co-senior authors.

Cited by 30 publications

References 48 publications

Progressive White Matter Injury in Preclinical Dutch Cerebral Amyloid Angiopathy

Progressive White Matter Injury in Preclinical Dutch Cerebral Amyloid Angiopathy

Vascular burden and cognition: Mediating roles of neurodegeneration and amyloid-PET

ROOD-MRI: Benchmarking the robustness of deep learning segmentation models to out-of-distribution and corrupted data in MRI

Contact Info

Product

Resources

About