Knowledge Vault 3/84 - G.TEC BCI & Neurotechnology Spring School 2024 - Day 9
Assessing preserved brain functions and outcome
prognostication in comatose patients after cardiac arrest
Marzia De Lucia, CHUV (CH))
<Resume Image >

Concept Graph & Resume using Claude 3 Opus | Chat GPT4 | Llama 3:

graph LR classDef marzia fill:#f9d4d4, font-weight:bold, font-size:14px; classDef coma fill:#d4f9d4, font-weight:bold, font-size:14px; classDef tests fill:#d4d4f9, font-weight:bold, font-size:14px; classDef deeplearning fill:#f9f9d4, font-weight:bold, font-size:14px; classDef future fill:#f9d4f9, font-weight:bold, font-size:14px; A[Marzia De Lucia] --> B[Marzia studies brain function
in post-arrest coma. 1] A --> C[Most post-arrest coma patients
have low survival. 2] C --> D[Early prognosis crucial for
personalized interventions. 2] C --> E[Clinical tests suggest poor
outcomes, uncertainty remains. 3] C --> F[Treatment limits reliability of
tests in first 24h. 4] C --> G[Tests predict poor outcomes,
need to identify recovery. 5] A --> H[Data from coma patients
in two Swiss hospitals. 6] A --> I[Early studies: EEG responses
to auditory mismatch negativity. 7] I --> J[Patients discriminated sounds, improvement
day 1-2 predicted recovery. 8] I --> K[Sound discrimination preserved, didn't
predict outcome alone. 9] I --> L[Tested sensitivity to complex
patterns requiring consciousness. 10] L --> M[Used local-global paradigm eliciting
late EEG responses. 11] L --> N[Global effect in some
controls and patients. 12] N --> O[Global effect possible without
consciousness, debated finding. 13] L --> P[Follow-up: more patients, HD-EEG,
scales to clarify. 14] A --> Q[Predicting outcomes from resting
EEG using deep learning. 15] Q --> R[High accuracy predicting 3-month
outcomes, day 1 EEG. 16] Q --> S[Graded survival probability, links
to known EEG patterns. 17] Q --> T[Captured EEG info beyond
current clinical assessments. 18] Q --> U[Accurate for 'gray zone'
patients, current tests unclear. 19] Q --> V[Recovery linked to stronger
alpha/beta on day 1. 20] Q --> W[Complements clinical tests, encourages
clinical adoption of tools. 21] A --> X[Factors influencing day 1
vs day 2 accuracy. 22] A --> Y[Generalize to coma beyond
cardiac arrest, share data/algorithms. 23] Y --> Z[Results generalize across EEG
systems, simple models better. 24] Y --> AA[Share new datasets after
publication, invite collaboration. 25] Y --> AB[Prognostic EEG may work
best for cardiac arrest. 26] Y --> AC[Extend to other causes
like traumatic brain injury. 27] A --> AD[Related work: EEG mapping
cardiac risk, complementary approaches. 28] A --> AE[Aim: improve management, research
through open data sharing. 29] A --> AF[Optimistic about advancing prognosis
through collaboration, new tech. 30] class A,B marzia; class C,D,E,F,G coma; class H,I,J,K,L,M,N,O,P tests; class Q,R,S,T,U,V,W,X deeplearning; class Y,Z,AA,AB,AC,AD,AE,AF future;


1.- Marzia De Lucia studies preserved brain functions in coma patients after cardiac arrest to predict chances of recovery.

2.- Most coma patients after cardiac arrest have low survival rates. Accurate early prognosis is important for personalized interventions.

3.- Clinical tests in first 3 days suggest poor outcomes for some, but are uncertain for many in the "gray zone".

4.- Most patients are treated with temperature management and sedation in the crucial first 24 hours, limiting reliability of tests.

5.- Current clinical tests are tailored to predict poor outcomes. More is needed to identify patients likely to recover.

6.- Data was collected from coma patients in two main Swiss hospitals. Having data across sites helps generalize findings.

7.- Early studies looked at EEG responses to auditory mismatch negativity paradigms within 24 hours of coma onset.

8.- Decoding analysis showed coma patients could discriminate sounds. Improvement from day 1 to 2 predicted 93% of recoveries.

9.- Test complemented clinical scores and could have benefited 15/90 patients. Sound discrimination was preserved but didn't predict outcome.

10.- Next tested if coma patients are sensitive to more complex sound patterns that normally require consciousness.

11.- Used auditory local-global paradigm that elicits late (>300ms) EEG responses and recruits broad cortical network in healthy people.

12.- 4/11 healthy controls showed global effect. Heterogeneous late EEG decoding of global patterns in 10/24 coma patients.

13.- Suggests global effect doesn't require consciousness and can occur in coma. Provoked debate in the literature.

14.- Follow-up study being done with more patients, HD-EEG, consciousness scales to clarify global effect in coma.

15.- Another approach looked at predicting outcomes from resting state EEG using deep learning in 165 coma patients.

16.- 5-fold cross-validation showed high accuracy predicting 3-month outcomes, especially using EEG from day 1 vs day 2.

17.- Deep learning provides graded probability of survival. Visualizations linked prediction to known EEG patterns.

18.- Deep learning captured EEG info related to behavior in coma beyond current clinical assessments.

19.- Deep learning was particularly accurate for patients in the "gray zone" not clearly classified by current tests.

20.- Higher predicted chance of recovery correlated with stronger EEG amplitudes in alpha/beta bands on day 1.

21.- Results suggest deep learning of acute coma EEG can accurately predict outcomes and complement clinical tests.

22.- Further work needed on factors influencing day 1 vs day 2 accuracy and encouraging clinical adoption of tools.

23.- Hope to generalize to coma with other causes beyond cardiac arrest. Making data/algorithms openly available.

24.- Results appear to generalize across different EEG systems. Simple models generalized better than complex ones.

25.- Plan to share new datasets after publication, in addition to some already public. Invites collaboration.

26.- Prognostic EEG may work best for cardiac arrest coma because the brain is not directly injured initially.

27.- Extending the approaches to other causes of coma like traumatic brain injury which are more heterogeneous.

28.- Mentions related work using EEG to map cardiac risk by another group. Complementary approaches.

29.- Coma prognosis work aims to improve clinical management and research through open data sharing.

30.- Speaker is optimistic about advancing coma prognosis through collaborative research using shared data and new technologies.

Knowledge Vault built byDavid Vivancos 2024