FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2896397176> ?p ?o ?g. }

• W2896397176 abstract "Safety protocols in trials for Alzheimer's Disease therapies include surveillance for cerebral microbleeds (CMBs) on MRI. CMBs are focal hemosiderin deposits detected as focal signal reduction on T2*-weighted (T2*w) MRI. Normal anatomy such as blood vessels can also lead to focal dark spots. CMBs may be detected with high sensitivity but low specificity using conventional image processing. We present a two-stage method wherein conventional image processing is followed by artificial intelligence in the form of a convolutional neural network (CNN) providing specificity. 1896 T2*GRE and T1-weighted (T1w) pairs from the Mayo Clinic Study of Aging (MCSA) were used for training and test data. The Alzheimer's Disease Neuroimaging Initiative (ADNI) provided 4718 pairs used solely for validation. Visual CMB assessment of all T2*w images was carried out providing 1314 observed CMBs in Mayo and 2577 in ADNI. T1w and segmentation images were resampled to match T2*GRE images. Normalized cross-correlation of a bulls-eye shaped kernel highlighted small circular signal voids on T2*w images. “Candidate CMBs” were defined as a local maximum in a cluster with correlation >0.3 in the brain containing less than 30 pixels and with T2*w signal <70% of the mean white matter T2*w intensity. Scans had ∼1000 candidate CMB's using these sensitive criteria. A CNN using 32x32x3 voxel image blocks around candidate CMB locations was trained on 90% of the MCSA visually identified CMBs with 200-fold data augmentation (flips, rotations, scalings) and a 5x match of false candidates. Five input channels (T2w, correlation, T1w, WM and CSF) were used (Figure 1). Example inputs presented to the CNN demonstrate the challenge of CMB detection. Full slices are from a single subject T2*GRE image. A “true CMB” is shown in the left panel. A “false” candidate CMB's are shown middle and right. The candidate in the middle image is through-plane vessel. The candidate in the right panel is near the intensity and correlation threshold. Below each image are the five sub-images from the respective image slice that are presented as channels to the network. The network receives image patches immediately above and below the slices shown providing additional contextual information. The trained network correctly classified these candidate CMBs. (Figure 2.) Testing on the 10% held-out MCSA CMBs (N=140 without augmentation) and 700 non-CMB candidates yielded good separation, AUC= 0.974 (95% CI 0.953-0.992). From ADNI 2577 “true” and 12855 “false” CMBs were used for validation. On ADNI, a multi-center, multi-MR-vendor study, similar performance was observed: AUC 0.985 (95% CI 0.982-0.987). Rejection of false CMBs at the ∼98% level is feasible while retaining ∼98% of true CMBs. ROC curves for rejection of false candidate CMB's are shown. Performance in the 10% hold-out set from MCSA images and the completely independent ADNI validation set are similar indicating that the model is stable in the face of variability encountered in a large multi-center, multi-vendor environment. A reasonable operating point allows 98% of “true” CMB's to be kept with around 98% rejection of “false” CMBs. Manual assessment of CMBs is laborious and subject to inter-rater variability. This method is sufficiently accurate and generalizable to reduce, if not eventually eliminate, the visual assessment process ." @default.
• W2896397176 created "2018-10-26" @default.
• W2896397176 creator A5005099850 @default.
• W2896397176 creator A5005572080 @default.
• W2896397176 creator A5013894575 @default.
• W2896397176 creator A5021517944 @default.
• W2896397176 creator A5031990281 @default.
• W2896397176 creator A5045246745 @default.
• W2896397176 creator A5059027860 @default.
• W2896397176 creator A5077933873 @default.
• W2896397176 creator A5079202443 @default.
• W2896397176 date "2018-07-01" @default.
• W2896397176 modified "2023-10-16" @default.
• W2896397176 title "P4‐232: AUTOMATING CEREBRAL MICROBLEED DETECTION IN SUPPORT OF ALZHEIMER'S DISEASE TRIALS USING A CONVOLUTIONAL NEURAL NETWORK AI" @default.
• W2896397176 doi "https://doi.org/10.1016/j.jalz.2018.07.053" @default.
• W2896397176 hasPublicationYear "2018" @default.
• W2896397176 type Work @default.
• W2896397176 sameAs 2896397176 @default.
• W2896397176 citedByCount "1" @default.
• W2896397176 countsByYear W28963971762022 @default.
• W2896397176 crossrefType "journal-article" @default.
• W2896397176 hasAuthorship W2896397176A5005099850 @default.
• W2896397176 hasAuthorship W2896397176A5005572080 @default.
• W2896397176 hasAuthorship W2896397176A5013894575 @default.
• W2896397176 hasAuthorship W2896397176A5021517944 @default.
• W2896397176 hasAuthorship W2896397176A5031990281 @default.
• W2896397176 hasAuthorship W2896397176A5045246745 @default.
• W2896397176 hasAuthorship W2896397176A5059027860 @default.
• W2896397176 hasAuthorship W2896397176A5077933873 @default.
• W2896397176 hasAuthorship W2896397176A5079202443 @default.
• W2896397176 hasConcept C118552586 @default.
• W2896397176 hasConcept C126838900 @default.
• W2896397176 hasConcept C142724271 @default.
• W2896397176 hasConcept C143409427 @default.
• W2896397176 hasConcept C153180895 @default.
• W2896397176 hasConcept C154945302 @default.
• W2896397176 hasConcept C2778373026 @default.
• W2896397176 hasConcept C2779134260 @default.
• W2896397176 hasConcept C2781192897 @default.
• W2896397176 hasConcept C2989005 @default.
• W2896397176 hasConcept C41008148 @default.
• W2896397176 hasConcept C502032728 @default.
• W2896397176 hasConcept C54170458 @default.
• W2896397176 hasConcept C58693492 @default.
• W2896397176 hasConcept C71924100 @default.
• W2896397176 hasConcept C81363708 @default.
• W2896397176 hasConceptScore W2896397176C118552586 @default.
• W2896397176 hasConceptScore W2896397176C126838900 @default.
• W2896397176 hasConceptScore W2896397176C142724271 @default.
• W2896397176 hasConceptScore W2896397176C143409427 @default.
• W2896397176 hasConceptScore W2896397176C153180895 @default.
• W2896397176 hasConceptScore W2896397176C154945302 @default.
• W2896397176 hasConceptScore W2896397176C2778373026 @default.
• W2896397176 hasConceptScore W2896397176C2779134260 @default.
• W2896397176 hasConceptScore W2896397176C2781192897 @default.
• W2896397176 hasConceptScore W2896397176C2989005 @default.
• W2896397176 hasConceptScore W2896397176C41008148 @default.
• W2896397176 hasConceptScore W2896397176C502032728 @default.
• W2896397176 hasConceptScore W2896397176C54170458 @default.
• W2896397176 hasConceptScore W2896397176C58693492 @default.
• W2896397176 hasConceptScore W2896397176C71924100 @default.
• W2896397176 hasConceptScore W2896397176C81363708 @default.
• W2896397176 hasIssue "7S_Part_29" @default.
• W2896397176 hasLocation W28963971761 @default.
• W2896397176 hasOpenAccess W2896397176 @default.
• W2896397176 hasPrimaryLocation W28963971761 @default.
• W2896397176 hasRelatedWork W1991514405 @default.
• W2896397176 hasRelatedWork W2106392979 @default.
• W2896397176 hasRelatedWork W2132587081 @default.
• W2896397176 hasRelatedWork W2673946014 @default.
• W2896397176 hasRelatedWork W2767651786 @default.
• W2896397176 hasRelatedWork W2771176174 @default.
• W2896397176 hasRelatedWork W2912288872 @default.
• W2896397176 hasRelatedWork W2963023839 @default.
• W2896397176 hasRelatedWork W4206076898 @default.
• W2896397176 hasRelatedWork W4310334659 @default.
• W2896397176 hasVolume "14" @default.
• W2896397176 isParatext "false" @default.
• W2896397176 isRetracted "false" @default.
• W2896397176 magId "2896397176" @default.
• W2896397176 workType "article" @default.