MASTER Assignment
Exploring simulated data for machine learning to predict neurological outcome in patients with a postanoxic encephalopathy
Type : Master CS, together with the Clinical Neurophysiology group
Duration : TBD
Student : unassigned
If you are interested please contact :
Background:
Reliable outcome prediction in patients with a postanoxic coma is an important clinical challenge, and important for decision making and optimal treatment. Visual inspection of continuous EEG recordings is increasingly recognized as a reliable tool for prognostication [1, 3, 4].
In order to support clinicians with the interpretation of the EEG, machine learning approaches have been applied obtaining sensitivities of 40-50% at specificity 100% for poor and sensitivities of 50-55% at specificity of 95% for good outcome [6, 5].
A key challenge is to further increase this diagnostic accuracy with machine learning algorithms. One way to improve predictions is to use more data for training the model. As clinical data is rather scarce, artificial data that well represents EEG data from patients can be used. Such data from simulations is available [2], and may serve for additional or primary training of algorithms.
This thesis should explore the usefulness of artificial data for machine learning. The following research questions should be answered:
R-1 Do the predictions of classifiers trained on real and artificial EEG data differ and to which extent?
R-2 How does the amount of artificial data influence the prediction?
R-3 Can we use artificial data, only, to improve predictions?
To answer the research questions, the following conditions should be experimentally compared:
C-A Training on just real data ; testing on real data
C-B Training on combination of real and artificial data (but with the same total amount of items as in condition C-A); testing on real data
C-C Training on real and artificial data (vary the total amount, to have 10 - 90% artificial data); testing on real data
C-D Training on only artificial data; testing on real data.
About you
You already have or are eager to obtain Python programming skills.
You already have good knowledge in machine learning or data science and basic knowledge of deep learning.
You are interested in interdisciplinary research to broaden your view on science.
You are interested in applying machine learning technology to applications with high impact for society.
References
[1] J Hofmeijer, T.M.J. Beernink, F.H. Bosch, A Beishuizen, M.C. Tjepkema-Cloostermans, and M.J.A.M. van Putten. Early EEG contributes to multimodal outcome prediction of postanoxic coma. Neurology, (85):1–7, 2015.
[2] B.J. Ruijter, J. Hofmeijer, H.G.E. Meijer, and M.J.A.M. van Putten. Synaptic damage underlies EEG abnormalities in postanoxic encephalopathy: A computational study. Clinical Neurophysiology, 128(9):1682–1695, 2017.
[3] Adithya Sivaraju, Emily J. Gilmore, Charles R. Wira, Anna Stevens, Nishi Rampal, Jeremy J. Moeller, David M. Greer, Lawrence J. Hirsch, and Nicolas Gaspard. Prognostication of post-cardiac arrest coma: early clinical and electroencephalographic predictors of outcome. Intensive Care Medicine, 41(7):1264–1272, 2015.
[4] M. Spalletti, R. Carrai, M. Scarpino, C. Cossu, A. Ammannati, M. Ciapetti, L. Tadini Buoninsegni, A. Peris, S. Valente, A. Grippo, and A. Amantini. Single electroencephalographic patterns as specific and time-dependent indicators of good and poor outcome after cardiac arrest. Clinical Neurophysiology, 127(7):2610–2617, 2016.
[5] Marleen C Tjepkema-cloostermans, Jeannette Hofmeijer, Albertus Beishuizen, HaroldW Hom, Michiel J Blans, Frank H Bosch, and Michel J A M Van Putten. Cerebral Recovery Index: Reliable Help for Prediction of Neurologic Outcome After Cardiac Arrest. Critical Care Medicine, pages 1–9, 2017.
[6] M.J.A.M. Van Putten, J. Hofmeijer, B.J. Ruijter, and M.C. Tjepkema-Cloostermans. Deep learning for outcome prediction of postanoxic coma. In IFMBE Proceedings, volume 65, 2018.