FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3033467841> ?p ?o ?g. }

• W3033467841 endingPage "1417" @default.
• W3033467841 startingPage "1417" @default.
• W3033467841 abstract "1417 Introduction: Collecting abundant PET data from real patients for prototyping algorithm development has historically been prohibitive because it requires delivering a radiation dose. In this work, we demonstrate the prediction of zero-dose PET (zdPET) volumes from abundant and dose-free MRI volumes, without any a priori segmentation or manual annotation. These zdPET volumes can be used for pseudo-realistic synthetic-PET dataset generation, algorithm development and calibration. Methods: We assembled an anonymized dataset of 57 whole-body 18F-FDG-18 PET/MRI exams. After registering these exams to a common grid, we used a deep neural network, based on a fully-convolutional 3D residual U-net architecture, to perform the cross-domain mapping from T2-weighed MRI to PET. For training, MRI/PET volumes from 40 exams were fed respectively as input/groundtruth on a [32 px x 32 px x 16 px] patch-by-patch basis, without a priori location context or annotation. Stochastic gradient descent was used to optimize the deep neural network using an objective function that linearly-combines terms measuring the mean absolute error, the mean absolute error in high activity regions, and the mean relative error between predicted PET volumes and full-dose SUV-normalized PET volumes in the training dataset. The trained algorithm was subsequently applied to the 17 withheld (test) exams and the resulting images were compared both quantitatively in terms of absolute error, as well as qualitatively based on the predicted uptake patterns. Results: zdPET imagery derived from MR-based DNNs provide good patient-geometry-conforming prediction of physiologic activity patterns in tissues and organs that are exhibit consistent uptake across patients and patient exams (e.g. the liver). The developed algorithm, however, is unable to effectively predict physiologic activity in areas with variable uptake (e.g. myocardium). These observations are seen clearly in Figure 1a-c, by comparing differences in the maximum intensity projection (MIP) and absolute error between the predicted zdPET and full-dose PET volumes of a patient. A comparison of the predicted physiologic activity patterns across patient exams is shown in Figure 1d-f, where each row depicts slices from a different patient exam with different geometry and relative activity. Conclusions: Results indicate the viability of generating large realistic PET imagery datasets from abundant dose-free MRI volumes. Some issues related to consistent SUV-normalization, optimizing relative vs absolute error, patch size selection, and mitigating blocking artifacts remain. One option is to train a different network on each octant or for organ group of interest, as to enhance the context required to reproduce the nominal uptake patterns. Future work includes leveraging zdPET volumes to improve artifact-mitigating PET reconstruction algorithms (e.g. scatter, motion, low-SNR)." @default.
• W3033467841 created "2020-06-12" @default.
• W3033467841 creator A5016864113 @default.
• W3033467841 creator A5041537214 @default.
• W3033467841 creator A5048679648 @default.
• W3033467841 creator A5063805004 @default.
• W3033467841 creator A5071577908 @default.
• W3033467841 creator A5072458448 @default.
• W3033467841 date "2020-05-01" @default.
• W3033467841 modified "2023-09-23" @default.
• W3033467841 title "Deep Learning-based MR-derived PET Prediction for Patient-Conforming PET Phantoms" @default.
• W3033467841 hasPublicationYear "2020" @default.
• W3033467841 type Work @default.
• W3033467841 sameAs 3033467841 @default.
• W3033467841 citedByCount "0" @default.
• W3033467841 crossrefType "journal-article" @default.
• W3033467841 hasAuthorship W3033467841A5016864113 @default.
• W3033467841 hasAuthorship W3033467841A5041537214 @default.
• W3033467841 hasAuthorship W3033467841A5048679648 @default.
• W3033467841 hasAuthorship W3033467841A5063805004 @default.
• W3033467841 hasAuthorship W3033467841A5071577908 @default.
• W3033467841 hasAuthorship W3033467841A5072458448 @default.
• W3033467841 hasConcept C108583219 @default.
• W3033467841 hasConcept C11413529 @default.
• W3033467841 hasConcept C127077266 @default.
• W3033467841 hasConcept C151730666 @default.
• W3033467841 hasConcept C153180895 @default.
• W3033467841 hasConcept C154945302 @default.
• W3033467841 hasConcept C155512373 @default.
• W3033467841 hasConcept C2775842073 @default.
• W3033467841 hasConcept C2779343474 @default.
• W3033467841 hasConcept C2989005 @default.
• W3033467841 hasConcept C41008148 @default.
• W3033467841 hasConcept C71924100 @default.
• W3033467841 hasConcept C81363708 @default.
• W3033467841 hasConcept C86803240 @default.
• W3033467841 hasConcept C89600930 @default.
• W3033467841 hasConceptScore W3033467841C108583219 @default.
• W3033467841 hasConceptScore W3033467841C11413529 @default.
• W3033467841 hasConceptScore W3033467841C127077266 @default.
• W3033467841 hasConceptScore W3033467841C151730666 @default.
• W3033467841 hasConceptScore W3033467841C153180895 @default.
• W3033467841 hasConceptScore W3033467841C154945302 @default.
• W3033467841 hasConceptScore W3033467841C155512373 @default.
• W3033467841 hasConceptScore W3033467841C2775842073 @default.
• W3033467841 hasConceptScore W3033467841C2779343474 @default.
• W3033467841 hasConceptScore W3033467841C2989005 @default.
• W3033467841 hasConceptScore W3033467841C41008148 @default.
• W3033467841 hasConceptScore W3033467841C71924100 @default.
• W3033467841 hasConceptScore W3033467841C81363708 @default.
• W3033467841 hasConceptScore W3033467841C86803240 @default.
• W3033467841 hasConceptScore W3033467841C89600930 @default.
• W3033467841 hasOpenAccess W3033467841 @default.
• W3033467841 hasRelatedWork W2046785176 @default.
• W3033467841 hasRelatedWork W2059939049 @default.
• W3033467841 hasRelatedWork W2189395170 @default.
• W3033467841 hasRelatedWork W2464052377 @default.
• W3033467841 hasRelatedWork W2793450646 @default.
• W3033467841 hasRelatedWork W2904950251 @default.
• W3033467841 hasRelatedWork W2910984772 @default.
• W3033467841 hasRelatedWork W2955800894 @default.
• W3033467841 hasRelatedWork W2963982674 @default.
• W3033467841 hasRelatedWork W2995999682 @default.
• W3033467841 hasRelatedWork W3000261590 @default.
• W3033467841 hasRelatedWork W3007045034 @default.
• W3033467841 hasRelatedWork W3012291095 @default.
• W3033467841 hasRelatedWork W3041500444 @default.
• W3033467841 hasRelatedWork W3125693017 @default.
• W3033467841 hasRelatedWork W3157216896 @default.
• W3033467841 hasRelatedWork W3164816110 @default.
• W3033467841 hasRelatedWork W3165085080 @default.
• W3033467841 hasRelatedWork W3194935959 @default.
• W3033467841 hasRelatedWork W97908006 @default.
• W3033467841 hasVolume "61" @default.
• W3033467841 isParatext "false" @default.
• W3033467841 isRetracted "false" @default.
• W3033467841 magId "3033467841" @default.
• W3033467841 workType "article" @default.