Beyond the Limits of Visual Learning: Prediction of Time From Onset From Noncontrast CT of Acute Ischemic Stroke.

HomeStrokeVol. 54, No. 1Beyond the Limits of Visual Learning: Prediction of Time From Onset From Noncontrast CT of Acute Ischemic Stroke Free AccessResearch ArticlePDF/EPUBAboutView PDFView EPUBSections ToolsAdd to favoritesDownload citationsTrack citationsPermissionsDownload Articles + Supplements ShareShare onFacebookTwitterLinked InMendeleyReddit Jump toSupplemental MaterialFree AccessResearch ArticlePDF/EPUBBeyond the Limits of Visual Learning: Prediction of Time From Onset From Noncontrast CT of Acute Ischemic Stroke Yutong Chen, Ernst Mayerhofer, Livia Parodi, Andreas Harloff, Puneet Batra, Jonathan Rosand and Christopher D. Anderson Yutong ChenYutong Chen https://orcid.org/0000-0001-8024-8775 Henry and Allison McCance Center for Brain Health, Massachusetts General Hospital, Boston, MA (Y.C., E.M., L.P., J.R., C.D.A.). The Broad Institute of Harvard and MIT, Cambridge (Y.C., E.M., L.P., P.B., J.R., C.D.A.). *Y. Chen and E. Mayerhofer contributed equally. Search for more papers by this author , Ernst MayerhoferErnst Mayerhofer Correspondence to: Ernst Mayerhofer, MD, Center for Genomic Medicine, Massachusetts General Hospital, 185 Cambridge St, Boston, MA 02114. Email E-mail Address: [email protected] https://orcid.org/0000-0001-8902-4209 Henry and Allison McCance Center for Brain Health, Massachusetts General Hospital, Boston, MA (Y.C., E.M., L.P., J.R., C.D.A.). The Broad Institute of Harvard and MIT, Cambridge (Y.C., E.M., L.P., P.B., J.R., C.D.A.). Department of Neurology and Neurophysiology, Medical Center – University of Freiburg, Faculty of Medicine, University of Freiburg, Germany (E.M., A.H.). *Y. Chen and E. Mayerhofer contributed equally. Search for more papers by this author , Livia ParodiLivia Parodi https://orcid.org/0000-0003-0605-2381 Henry and Allison McCance Center for Brain Health, Massachusetts General Hospital, Boston, MA (Y.C., E.M., L.P., J.R., C.D.A.). The Broad Institute of Harvard and MIT, Cambridge (Y.C., E.M., L.P., P.B., J.R., C.D.A.). Department of Neurology, Brigham and Women’s Hospital, Boston, MA (L.P., C.D.A.). Search for more papers by this author , Andreas HarloffAndreas Harloff https://orcid.org/0000-0002-3252-7910 Department of Neurology and Neurophysiology, Medical Center – University of Freiburg, Faculty of Medicine, University of Freiburg, Germany (E.M., A.H.). Search for more papers by this author , Puneet BatraPuneet Batra https://orcid.org/0000-0001-6822-0593 The Broad Institute of Harvard and MIT, Cambridge (Y.C., E.M., L.P., P.B., J.R., C.D.A.). Search for more papers by this author , Jonathan RosandJonathan Rosand https://orcid.org/0000-0002-1014-9138 Henry and Allison McCance Center for Brain Health, Massachusetts General Hospital, Boston, MA (Y.C., E.M., L.P., J.R., C.D.A.). The Broad Institute of Harvard and MIT, Cambridge (Y.C., E.M., L.P., P.B., J.R., C.D.A.). Search for more papers by this author and Christopher D. AndersonChristopher D. Anderson https://orcid.org/0000-0002-0053-2002 Henry and Allison McCance Center for Brain Health, Massachusetts General Hospital, Boston, MA (Y.C., E.M., L.P., J.R., C.D.A.). The Broad Institute of Harvard and MIT, Cambridge (Y.C., E.M., L.P., P.B., J.R., C.D.A.). Department of Neurology, Brigham and Women’s Hospital, Boston, MA (L.P., C.D.A.). Search for more papers by this author Originally published7 Dec 2022https://doi.org/10.1161/STROKEAHA.122.041370Stroke. 2023;54:e7–e8Other version(s) of this articleYou are viewing the most recent version of this article. Previous versions: December 7, 2022: Ahead of Print Time from onset is unknown in approximately one-third of patients with ischemic stroke, hampering reperfusion therapy.1 Early ischemic changes can be detected as hypoattenuation on noncontrast computed tomography (NCCT), but there is no established CT-based marker for time from onset. As convolutional neural networks have been successful in identifying cryptic associations,2 we sought to predict time from stroke onset from NCCT.NCCT of ischemic stroke patients with known symptom onset time <24 hours from admission were obtained from the Massachusetts General Hospital stroke registry.3 All NCCT were inspected for quality control. The dataset was randomly split into training, validation, and test sets. 2D and 3D convolutional neural networks were trained to predict time from onset both for regression and classification (4.5, 6, 9, and 12 hours). To rule out insufficient information content, technical problems, or power limitations, additional models were trained to predict patient age as a positive control, as this has been successfully achieved previously.4 Best performing models in the validation dataset were evaluated on the test dataset (Supplemental Methods).In total, 1836 scans from 1208 patients (60% men; mean±SD age, 71±14 years) admitted from 2005 to 2020 were obtained and 1425 scans were retained for analysis (Figure [A];Table S1). Median time from symptom onset to NCCT was 3.8 hours (interquartile range, 1.6–7.6) with 57% and 86% taken within 4.5 and 12 hours, respectively. The best model for 4.5 hours yielded an area under the curve of 0.68 (95% CI [0.62, 0.75]) with specificity 0.81 and sensitivity 0.42. The best obtained median absolute error was 2.2 hours with poor correlation between predicted and true time from onset (r=0.31, Figure [B];Supplemental Results). In contrast, the best patient age prediction model yielded a median absolute error of 5.5 years, with a strong correlation between predicted and actual age (r=0.83, Figure [C]).Download figureDownload PowerPointFigure. Study design (A) and predicted values vs ground truth in the test dataset (B, C). CT indicates computed tomography; and MGH, Massachusetts General Hospital.Our results suggest that NCCT lacks extractable information on time from onset in acute ischemic stroke. NCCT may lack specificity, and individual variation in stroke progression reduces the reliability of time as a surrogate for known ischemic findings like hypoattenuation.5 Limitations of our study include stroke subtype differences in infarct evolution, complicating the development of a universal time inference model. Restriction to patients with known onset time could bias toward milder stroke severity or subtler ischemic changes. Similar approaches might still be feasible for certain patient subgroups or imaging modalities. Our sample might have been too small, although prediction of patient age was successful and more accurate than previous work,4 ruling out obvious technical errors or poor image quality.In conclusion, a visual learning approach using NCCT in acute ischemic stroke was not able to predict time from symptom onset. This finding provides further support, excluding CT-derived features as surrogates for estimating time from symptom onset. Future studies should continue to investigate reperfusion therapy in patients with unknown symptom onset.Article InformationData AvailabilityThe analysis source code can be found at https://github.com/Yutong441/ISCT. Individual-level NCCT cannot be shared due to restrictions of the obtained consent forms.Sources of FundingCDA is supported by NIH R01NS103924, U01NS069673, AHA 18SFRN34250007, AHA-Bugher 21SFRN812095, and the MGH McCance Center for Brain Health. JR receives research grants from NIH and the American Heart Association-Bugher Foundation.Supplemental MaterialExpanded Materials and MethodsFigures S1–S2Tables S1–S7Disclosures JR reports compensation from National Football League for expert witness services from Takeda Development Center Americas and Boehringer Ingelheim for consultant services, all unrelated to this work. CDA has received sponsored research support from Bayer AG and has consulted for ApoPharma unrelated to this work.Footnotes*Y. Chen and E. Mayerhofer contributed equally.Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/STROKEAHA.122.041370.For Sources of Funding and Disclosures, see page e8.Correspondence to: Ernst Mayerhofer, MD, Center for Genomic Medicine, Massachusetts General Hospital, 185 Cambridge St, Boston, MA 02114. Email emayerho@broadinstitute.orgReferences1. Søyland M-H, Tveiten A, Eltoft A, Øygarden H, Varmdal T, Indredavik B, Mathiesen EB. Wake-up stroke and unknown-onset stroke; occurrence and characteristics from the nationwide Norwegian Stroke Register.Eur Stroke J. 2022; 7:143–150. doi: 10.1177/23969873221089800CrossrefGoogle Scholar2. Feng R, Badgeley M, Mocco J, Oermann EK. Deep learning guided stroke management: a review of clinical applications.J NeuroInterventional Surg. 2018; 10:358–362. doi: 10.1136/neurintsurg-2017-013355CrossrefGoogle Scholar3. Ali SF, Singhal AB, Viswanathan A, Rost NS, Schwamm LH. Characteristics and outcomes among patients transferred to a regional comprehensive stroke center for tertiary care.Stroke. 2013; 44:3148–3153. doi: 10.1161/STROKEAHA.113.002493LinkGoogle Scholar4. Bermudez C, Plassard AJ, Chaganti S, Huo Y, Aboud KS, Cutting LE, Resnick SM, Landman BA. Anatomical context improves deep learning on the brain age estimation task.Magn Reson Imaging. 2019; 62:70–77. doi: 10.1016/j.mri.2019.06.018CrossrefGoogle Scholar5. Gao J, Parsons MW, Kawano H, Levi CR, Evans T-J, Lin L, Bivard A. Visibility of CT early ischemic change is significantly associated with time from stroke onset to baseline scan beyond the first 3 hours of stroke onset.J Stroke. 2017; 19:340–346. doi: 10.5853/jos.2016.01424CrossrefGoogle Scholar eLetters(0)eLetters should relate to an article recently published in the journal and are not a forum for providing unpublished data. Comments are reviewed for appropriate use of tone and language. Comments are not peer-reviewed. Acceptable comments are posted to the journal website only. Comments are not published in an issue and are not indexed in PubMed. Comments should be no longer than 500 words and will only be posted online. References are limited to 10. Authors of the article cited in the comment will be invited to reply, as appropriate.Comments and feedback on AHA/ASA Scientific Statements and Guidelines should be directed to the AHA/ASA Manuscript Oversight Committee via its Correspondence page.Sign In to Submit a Response to This Article Previous Back to top Next FiguresReferencesRelatedDetails January 2023Vol 54, Issue 1 Advertisement Article InformationMetrics © 2022 American Heart Association, Inc.https://doi.org/10.1161/STROKEAHA.122.041370PMID: 36475469 Originally publishedDecember 7, 2022 PDF download Advertisement SubjectsComputerized Tomography (CT)Ischemic Stroke