RT Journal Article
SR Electronic
T1 Interpretable classification of Alzheimer’s disease pathologies with a convolutional neural network pipeline
JF bioRxiv
FD Cold Spring Harbor Laboratory
SP 454793
DO 10.1101/454793
A1 Ziqi Tang
A1 Kangway V. Chuang
A1 Charles DeCarli
A1 Lee-Way Jin
A1 Laurel Beckett
A1 Michael J. Keiser
A1 Brittany N. Dugger
YR 2018
UL http://biorxiv.org/content/early/2018/10/30/454793.abstract
AB Neuropathologists assess vast brain areas to identify diverse and subtly-differentiated morphologies. Standard semi-quantitative scoring approaches, however, are coarse-grained and lack precise neuroanatomic localization. We report a proof-of-concept deep learning pipeline identifying specific neuropathologies—amyloid plaques and cerebral amyloid angiopathy—in immunohistochemically-stained archival slides. Using automated segmentation of stained objects and a cloud-based interface, we annotated >70,000 plaque candidates from 43 whole slide images (WSIs) to train and evaluate convolutional neural networks. Networks achieved strong plaque classification on a 10-WSI hold-out set (0.993 and 0.744 areas under the receiver operating characteristic and precision recall curve, respectively). Prediction confidence maps visualized morphology distributions for WSIs at high resolution. Resulting plaque-burden scores correlated well with established semi-quantitative scores on a 30-WSI blinded hold-out. Finally, saliency mapping demonstrated that networks learned patterns agreeing with accepted pathologic features. This scalable means to augment a neuropathologist’s ability may suggest a route to neuropathologic deep phenotyping.List of AbbreviationsAβAmyloid-betaADAlzheimer’s diseaseAUPRCarea under the precision recall curveAUROCarea under the receiver operator characteristicCERADConsortium to Establish a Registry for Alzheimer’s DiseaseCNNconvolutional neural networkCAAcerebral amyloid angiopathyFNfalse negativeFPfalse positivegrad-CAMgradient-weighted class activation mappingHSVhue-saturation-valueIHCimmunohistochemicalLHClightness-chroma-hueMPPmicrons per pixelRGBred-green-blueSQLstandardized query languageTPtrue positiveTNtrue negativeWSIwhole slide image