Creat membership Creat membership
Sign in

Forgot password?

Confirm
  • Forgot password?
    Sign Up
  • Confirm
    Sign In
Creat membership Creat membership
Sign in

Forgot password?

Confirm
  • Forgot password?
    Sign Up
  • Confirm
    Sign In
Collection
For ¥0.57 per day, unlimited downloads CREATE MEMBERSHIP Download

toTop

If you have any feedback, Please follow the official account to submit feedback.

Turn on your phone and scan

home > search >

T114. Predicting coma outcome using resting-state fMRI and machine learning

Author:
Deborah Pugin  Jeremy Hofmeister  Yvan Gasche  Dimitri Van De Ville  Serge Vulliémoz  Sven Haller  


Journal:
Clinical Neurophysiology


Issue Date:
2018


Abstract(summary):

Introduction Early prediction of neurological outcome of post-anoxic comatose patients after cardiac arrest(CA) is challenging. Prognosis of comatose patient relies on multimodal testing: clinical examination, electrophysiological testing and structural neuroimaging (mainly diffusion MRI DWI). This multimodal prognostication is accurate for predicting poor outcome(i.e.death) but not sensitive for identifying patients with good outcome (i.e. consciousness recovery). Resting state functional MRI (rs-fMRI) is a powerful tool for mapping functional connectivity, especially in patients with low collaboration. Several studies showed that rs-fMRI can differentiate states of consciousness in chronically brain damaged patients. A recent study also showed that fMRI can detect early signs of consciousness in patient with acute traumatic brain injury. However, rs-fMRI has not been systematically assessed for the early prognositcation of post-anoxic comatose patient. Methods We assessed whole brain functional connectivity (FC) of 17 post-anoxic comatose patients early after CA using rs-fMRI. Nine patients recovered consciousness(good outcome) while eight died(poor outcome). We estimated FC for each patient following a standard procedure described by Leonardi and Richiardi et al. We statistically compared whole brain FC between good and poor outcome group, to assess which brain connections differed between them. Then, we trained a machine learning classifier(a Support Vector Machine classifier, SVM) using a Leave-One-Out Cross-Validation method, to automatically predict coma outcome (good/poor) based on whole-brain FC of comatose patients. Finally, we compared this outcome-prognostication based on fMRI to those using standard structural DWI. Results Good and poor coma outcome groups were similar in terms of demographics, except for time to ROSC. Good outcome group showed significant increase in whole-brain FC between most cortical brain regions, with the strongest changes occuring within and between occipital and parietal, temporal and frontal regions. Using whole-brain FC and a SVM classifier to predict coma outcome yielded to an overall prediction accuracy of 94.4% (AUC 0.94). Interestingly, automatic outcome prognostication using functional neuroimaging achieved better results that state-of the-art structural neuroimaging methods like DWI (accuracy 70.6%). Conclusion We used rs-fMRI to predict coma outcome in a cohort of post-anoxic comatose patients early after CA. We deliberately chose to include only patients with indeterminate prognosis after standard multimodal testing, in order to assess the contribution of rs-fMRI in the early prognostication of coma outcome. We found that automatic prediction based on fMRI yielded much better results than current diffusion neuroimaging methods, notably for identifying patients who recovered consciousness.


Page:
e46-e46


VIEW PDF

The preview is over

If you wish to continue, please create your membership or download this.

Create Membership

Similar Literature

Submit Feedback

This function is a member function, members do not limit the number of downloads