Stroke: Clinical Applications & Predictive Modeling

Wednesday 14 May 2014

Ona Wu¹, Lisa Cloonan², Steven Mocking¹, Mark Bouts¹, Jonathan Rosand², Karen L. Furie², and Natalia S. Rost^2
1Athinoula A Martinos Center, Massachusetts General Hospital, Boston, MA, United States, ²Department of Neurology, Massachusetts General Hospital, Boston, United States

Neuroimaging is often proposed as a potential surrogate for clinical outcome in the evaluation of novel stroke therapies. We analyzed 480 patients from a prospective single center stroke data repository using voxel-based symptom mapping (VLSM) techniques with acute DWI lesions and follow-up modified Rankin Scale Score and NIH Stroke Scale Score. We found acute lesions in the left hemisphere were associated with poor outcome. VLSM methods may provide insight into patients at risk of poor long-term outcome and provide guidance towards which patients will most likely benefit from aggressive acute intervention and follow-up physical therapy.

Venkata Veerendranadh Chebrolu¹, Ashish Rao², Dattesh D Shanbhag¹, Sandeep N Gupta³, Uday Patil^1,4, Patrice Hervo⁵, Catherine Oppenheim^6,7, and Rakesh Mullick^8
1Medical Image Analysis Lab, GE Global Research, Bangalore, Karnataka, India, ²Bio-Medical Signal Analysis Lab, GE Global Research, Bangalore, Karnataka, India, ³Biomedical Image Processing Lab, GE Global Research, Niskayuna, NY, United States, ⁴Manipal Health Enterprises Pvt. Ltd., Bangalore, Karnataka, India, ⁵GE Healthcare, Buc, France, ⁶Departments of Radiology and Neurology, Centre Hospitalier, Sainte-Anne, Paris, France,⁷Université Paris Descartes, Paris, France, ⁸Diagnostics and Biomedical Technologies, GE Global Research, Bangalore, Karnataka, India

Independent component analysis of DSC-MRI data was demonstrated to identify hypo-perfused regions in acute ischemic stroke. Independent component analysis based assessment acute ischemic stroke can be achieved without arterial input function detection and explicit parameterization of the DSC perfusion data.

Mark JRJ Bouts¹, Elissa McIntosh¹, Raquel Bezerra¹, Izzuddin Diwan¹, Steven Mocking¹, Priya Garg¹, William T Kimberly², Ethem M Arsava¹, William A Copen³, Pamela W Schaefer³, Hakan Ay¹, Aneesh B Singhal², Bruce R Rosen¹, Rick M Dijkhuizen⁴, and Ona Wu^1
1Athinoula A. Martinos Center, Dept Radiology, Massachusetts General Hospital, Charlestown, Massachusetts, United States, ²Dept Neurology, Massachusetts General Hospital, Boston, Massachusetts, United States, ³Dept Radiology, Massachusetts General Hospital, Boston, Massachusetts, United States, ⁴Biomedical imaging & Spectroscopy group, Image Sciences Institute, University Medical Center Utrecht, Utrecht, Utrecht, Netherlands

MRI-based prediction algorithms may more accurately assess tissue at risk of infarction following acute ischemic stroke than perfusion-diffusion mismatch. Yet, few studies quantitatively evaluated the predictive performance of several algorithms. We evaluated four algorithms in a cohort of acute ischemic stroke patients not receiving subsequent revascularization intervention nor novel therapeutics. Two regression models (generalized linear model, generalized additive model) were tested against two ensemble methods (adaptive boosting and random forest). All algorithms performed better than perfusion-diffusion mismatch for predicting follow-up infarct. More complex algorithms offered improved accuracy in predicting tissue infarction after stroke.

Jacob Levman¹, George Harston², Yee Kai Tee¹, Thomas W Okell³, Nicholas Blockley³, Michael Chappell¹, Peter Jezzard³, James Kennedy², and Stephen Payne^1
1Engineering Science, Institute of Biomedical Engineering, University of Oxford, Oxford, England, United Kingdom, ²Radcliffe Department of Medicine, John Radcliffe Hospital, University of Oxford, Oxford, England, United Kingdom, ³Department of Clinical Neurosciences, Oxford Centre for Functional MRI of the Brain, John Radcliffe Hospital, University of Oxford, Oxford, England, United Kingdom

This study assessed MRI modalities for predicting tissue outcome in acute stroke subjects. Diffusion, perfusion and pH sensitive MR images were acquired at 4, 6 and 24 hours post stroke. Follow-up imaging was performed at 7 and 28 days. Receiver operating characteristic curve analysis was performed to evaluate each modality’s ability to predict tissue outcome as assessed by follow-up fluid attenuated inversion recovery (FLAIR) imaging. A demonstration of machine learning combining the three acute modalities was able to predict an additional 6% of FLAIR assessed tissue damage that was not suspicious based on diffusion characteristics alone.

Hongyu An¹, Andria L Ford², Yasheng Chen¹, Katie Vo³, William Powers⁴, Jin-Moo Lee², and Weili Lin^1
1Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States, ²Neurology, Washington University in St. Louis, St. Louis, MO, United States, ³Radiology, Washington University in St. Louis, St. Louis, MO, United States, ⁴Neurology, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States

Rates of FLAIR change depends on both time from stroke onset and depth of ischemia. Lesion age prediction is more reliable in moderate to severe ischemia. The median absolute prediction error on ischemia age is ~1hour across all patient using a generalized linear model.

George William John Harston¹, Thomas Okell², Fintan Sheerin³, Martino Cellerini³, Stephen Payne⁴, Peter Jezzard², Michael Chappell⁴, and James Kennedy^1
1Radcliffe Department of Medicine, University of Oxford, Oxford, Oxfordshire, United Kingdom, ²Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, Oxfordshire, United Kingdom, ³Oxford University Hospitals NHS Trust, Oxford, Oxfordshire, United Kingdom, ⁴Department of Engineering Science, University of Oxford, Oxford, Oxfordshire, United Kingdom

Hypoperfusion underlies the pathophysiology of ischemic stroke and as such has been used extensively to identify tissue at risk of infarction. However, presenting perfusion imaging has not been successfully used to select patients for treatment. In this observational cohort study patients with large volume ischemic stroke were scanned serially with quantitative arterial spin labeling perfusion imaging to better understand the natural history of absolute perfusion measurements and the dynamics of cerebral blood flow over the first hours and days following a stroke. A marked individual and temporal heterogeneity of cerebral blood flow was observed.

Jeffry R Alger¹, David S Liebeskind¹, Reza Jahan², Jeffrey L Saver¹, and Chelsea S. Kidwell^3,4
1Neurology, Geffen School of Medicine, UCLA, Los Angeles, CA, United States, ²Radiological Sciences, Geffen School of Medicine, UCLA, Los Angeles, CA, United States, ³Neurology, University of Arizona, Tuscon, Arizona, United States, ⁴Medical Imaging, University of Arizona, Tuscon, Arizona, United States

An analysis of scanner model bias in DSC MRI CBF readings from acute ischemic stroke patients enrolled in the MR RESCUE clinical trial was performed. The data were obtained using 16 unique MRI scanner models at 19 imaging centers. Results indicate a statistically significant scanner model related bias is present in CBF readings from well perfused tissues as well as hypoperfused tissues. Future multicenter studies that use DSC MRI should therefore take scanner model bias into consideration. Between-patient variability that is unrelated to scanner model is also present and this variability is as large as the scanner model-related variability.

Salil Soman^1,2, Gautam Prasad^3,4, Elizabeth Hitchner⁵, Wei Zhou^5,6, Michael Moseley⁷, and Allyson Rosen^8,9
1Radiology, Stanford University, Menlo Park, CA, United States, ²California War Related Illness and Injury Study Center, Palo Alto Veteras Affairs Hospital, Palo Alto, CA, United States, ³LONI, University of Southern California, Los Angeles, CA, United States, ⁴Psychology, Stanford University, CA, United States, ⁵Vascular Surgery, Stanford University, CA, United States, ⁶Vascular Surgery, Veterans Affairs Palo Alto Health Care System, CA, United States, ⁷Radiology, Stanford University, CA, United States, ⁸Pschology, Stanford University, CA, United States, ⁹Pscychology, Veterans Affairs Palo Alto Health Care System, CA, United States

Some patients with carotid stenosis that undergo carotid surgery afterwards experience cognitive decline. Identifying these patients before surgery would allow targeting of therapies to minimize disability. We hypothesized that structural connectivity graphs could identify these patients. We performed T1, DTI, and neuropsychological testing prior to surgery. Repeat neuropsychological testing was then performed 1 month later. FreeSurfer 5.3 whole brain segmentation, whole brain HARDI tractography, and connectivity analysis were then performed. The graph analysis methods “weighted optimal community structure” & “binary connected component sizes metrics” both predicted patients that would experience cognitive decline with 81% sensitivity 83% and specificity (FDR .05).

J. B. De Vis¹, E. T. Petersen¹, N. S. Hartkamp¹, A. Bhogal¹, C. J.M. Klijn², L. J. Kappelle², and J. Hendrikse^1
1Radiology, University Medical Center Utrecht, Utrecht, Utrecht, Netherlands, ²Neurology, University Medical Center Utrecht, Utrecht, Utrecht, Netherlands

Calibrated MRI is an upcoming non-invasive technique to evaluate brain metabolism. So far, it has only been evaluated in healthy volunteers and has not been used in patients yet. In this study we investigate the potential of calibrated MRI to study hemodynamic impairment in patients with cerebrovascular disease. Our patients demonstrate lower BOLD reactivity but equal oxygen extraction fraction suggesting impaired vascular reactivity. In addition, we find that blood flow through collateral pathways introduces artefacts in 27% of our patients limiting the use of calibrated MRI in patients with cerebrovascular disease.

Fatima Nasrallah¹, Zhong Kang Lu², Hong Xin³, Guan Cuntai², Kai Keng Ang², Kok Soon Phua², Irvin The⁴, Wei peng Teo⁵, Zhao Ling Yun⁶, Ning Tang⁶, Effie Chew⁶, and Kai-Hsiang Chuang^3
1Clinical Imaging Research Center, Singapore, Singapore, Singapore, ²Institute for Infocomm Research, A*STAR, Singapore, Singapore, ³Singapore Bioimaging Consortium, Singapore, Singapore, ⁴Department of Cardiovascular Medicine, University of Oxford, Oxford, United Kingdom, ⁵Central Queensland University, Queensland, Australia, ⁶The Division of Neurology and Rehabilitation Medicine, National University Hospital System, Singapore, Singapore

We have combined robot-assisted motor imagery and brain-computer interface (MI-BCI) and transcranial direct current stimulation (tDCS) for the rehabilitation of stroke patients due to the potential of these combined methods to modulate the cortical excitability, and hence the potential to improve motor function recovery. The primary aim of this study was to investigate the structural and functional changes of the brain after rehabilitation training of MI-BCI combined with or without tDCS in stroke patients.