Show simple item record

dc.contributor.authorChen, Xiongchao
dc.contributor.authorPretorius, P. Hendrik
dc.contributor.authorZhou, Bo
dc.contributor.authorLiu, Hui
dc.contributor.authorJohnson, Karen L.
dc.contributor.authorLiu, Yi-Hwa
dc.contributor.authorKing, Michael A.
dc.contributor.authorLiu, Chi
dc.date2022-08-11T08:10:50.000
dc.date.accessioned2022-08-23T17:21:59Z
dc.date.available2022-08-23T17:21:59Z
dc.date.issued2022-04-26
dc.date.submitted2022-07-15
dc.identifier.citation<p>Chen X, Hendrik Pretorius P, Zhou B, Liu H, Johnson K, Liu YH, King MA, Liu C. Cross-vender, cross-tracer, and cross-protocol deep transfer learning for attenuation map generation of cardiac SPECT. J Nucl Cardiol. 2022 Apr 26. doi: 10.1007/s12350-022-02978-7. Epub ahead of print. PMID: 35474443. <a href="https://doi.org/10.1007/s12350-022-02978-7">Link to article on publisher's site</a></p>
dc.identifier.issn1071-3581 (Linking)
dc.identifier.doi10.1007/s12350-022-02978-7
dc.identifier.pmid35474443
dc.identifier.urihttp://hdl.handle.net/20.500.14038/48646
dc.description.abstractIt has been proved feasible to generate attenuation maps (mu-maps) from cardiac SPECT using deep learning. However, this assumed that the training and testing datasets were acquired using the same scanner, tracer, and protocol. We investigated a robust generation of CT-derived mu-maps from cardiac SPECT acquired by different scanners, tracers, and protocols from the training data. We first pre-trained a network using 120 studies injected with (99m)Tc-tetrofosmin acquired from a GE 850 SPECT/CT with 360-degree gantry rotation, which was then fine-tuned and tested using 80 studies injected with (99m)Tc-sestamibi acquired from a Philips BrightView SPECT/CT with 180-degree gantry rotation. The error between ground-truth and predicted mu-maps by transfer learning was 5.13 +/- 7.02%, as compared to 8.24 +/- 5.01% by direct transition without fine-tuning and 6.45 +/- 5.75% by limited-sample training. The error between ground-truth and reconstructed images with predicted mu-maps by transfer learning was 1.11 +/- 1.57%, as compared to 1.72 +/- 1.63% by direct transition and 1.68 +/- 1.21% by limited-sample training. It is feasible to apply a network pre-trained by a large amount of data from one scanner to data acquired by another scanner using different tracers and protocols, with proper transfer learning.
dc.language.isoen_US
dc.relation<p><a href="http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=pubmed&cmd=Retrieve&list_uids=35474443&dopt=Abstract">Link to Article in PubMed</a></p>
dc.relation.urlhttps://doi.org/10.1007/s12350-022-02978-7
dc.subjectAttenuation map generation
dc.subjectSPECT/CT
dc.subjectmyocardial perfusion imaging
dc.subjecttransfer learning
dc.subjectBiological and Chemical Physics
dc.subjectCardiology
dc.subjectMedical Biophysics
dc.subjectRadiology
dc.titleCross-vender, cross-tracer, and cross-protocol deep transfer learning for attenuation map generation of cardiac SPECT
dc.typeArticle
dc.source.journaltitleJournal of nuclear cardiology : official publication of the American Society of Nuclear Cardiology
dc.identifier.legacycoverpagehttps://escholarship.umassmed.edu/radiology_pubs/709
dc.identifier.contextkey30234297
html.description.abstract<p>It has been proved feasible to generate attenuation maps (mu-maps) from cardiac SPECT using deep learning. However, this assumed that the training and testing datasets were acquired using the same scanner, tracer, and protocol. We investigated a robust generation of CT-derived mu-maps from cardiac SPECT acquired by different scanners, tracers, and protocols from the training data. We first pre-trained a network using 120 studies injected with (99m)Tc-tetrofosmin acquired from a GE 850 SPECT/CT with 360-degree gantry rotation, which was then fine-tuned and tested using 80 studies injected with (99m)Tc-sestamibi acquired from a Philips BrightView SPECT/CT with 180-degree gantry rotation. The error between ground-truth and predicted mu-maps by transfer learning was 5.13 +/- 7.02%, as compared to 8.24 +/- 5.01% by direct transition without fine-tuning and 6.45 +/- 5.75% by limited-sample training. The error between ground-truth and reconstructed images with predicted mu-maps by transfer learning was 1.11 +/- 1.57%, as compared to 1.72 +/- 1.63% by direct transition and 1.68 +/- 1.21% by limited-sample training. It is feasible to apply a network pre-trained by a large amount of data from one scanner to data acquired by another scanner using different tracers and protocols, with proper transfer learning.</p>
dc.identifier.submissionpathradiology_pubs/709
dc.contributor.departmentDepartment of Radiology


This item appears in the following Collection(s)

Show simple item record