![Stage Parameter Description
Data
At least 24 hr Duration of interictal record
At least 150 min Duration of preictal record
Feature extraction
5 sec EEG epoch length
6 Number of EEG channels
Training 5 fold Cross validation
SVM classification
log2γ ~ [-10, 10] SVM radial basis function kernel parameter
log2C ~ [-10, 10] Cost parameter
20
Classification](https://image.slidesharecdn.com/nwangaerospace13-240618163225-06f63ed9/85/nwang_aerospace13_machine_learning_framework-ppt-20-320.jpg)
nwang_aerospace13_machine_learning_framework.ppt
- 1. A Machine Learning Framework for Space Medicine Predictive Diagnostics with Physiological Signals Ning Wang, Michael R. Lyu Dept. of Computer Science & Engineering, Chinese University of Hong Kong, Hong Kong Chenguang Yang School of Computing and Mathematics, Plymouth University, United Kingdom
- 2. Outline Introduction Electroencephalogram (EEG) in Aerospace Medicine Amplitude and Frequency Properties in EEG Predictive Diagnostics Framework Case study: Epileptic Seizure Prediction with EEG Discussion & Conclusion 2
- 3. Prognostics and health management (PHM) 3 For space missions Focuses on fundamental issues of system failures To predict when failures may occur For healthcare in space Preventive, occupational To predict and prevent health problems timely Subjects are pilots, astronauts, or persons involved in spaceflight Critical to aviation safety
- 4. Aerospace medicine 4 Predictive diagnostics Autonomously predict, prevent and manage potential health problems Identify negative health trends with concerned premonitory symptoms. Predict future health condition. Raise alarms in case of emergency. Disease prediction & health monitoring Computer-based, self-diagnosis, and self-directed treatment programs Forecast acute disease onset. Monitor health condition. Patient-specific.
- 5. EEG in aerospace medicine 5 Long been employed in crew selection and training. Considered as an essential health metric of people involved in space missions. Diagnosis for neurologic events. Help in determining an acute cardiovascular disease, etc.
- 6. How to acquire EEG data? Data recording Noninvasive electrodes uniformly arrayed on the scalp. Channel signal = difference between potentials measured at two electrodes. Annotated to be clinical events or not by medical experts. Scalp EEG 6
- 7. EEG signal’s rhythmic pattern 7
- 8. Amplitude and frequency properties in EEG An EEG signal is typically described in terms of rhythmic activities. Contains multiple frequency components. Differs in structure among subjects. A band-limited signal that describes the kth EEG rhythm is characterized by two sequences: -- amplitude of rhythm; -- phase of rhythm. 8 Extract dominant amplitude and phase components as signal descriptors, i.e., physiological cues!
- 9. Observations 9 Inclusive EEG rhythms Estimated frequency components
- 10. Predictive Diagnostics Framework 10 Physiological signal analysis algorithm Identify primary components Disease prediction and health monitoring architecture Machine learning based, subject-specific
- 11. Machine learning 11 “… a computer program that can learn from experience with respect to some class of tasks and performance measure …” (Mitchell, 1997) “Machine learning, a branch of artificial intelligence, is about the construction and study of systems that can learn from data. For example, a machine learning system could be trained on email messages to learn to distinguish between spam and non-spam messages. After learning, it can then be used to classify new email messages into spam and non-spam folders. …” (from Wikipedia)
- 12. About support vector machine (SVM) 12 Linear discriminant function Maximal margin best hyperplane. Support vectors: data points closest to the hyperplane.
- 13. Case study: epileptic seizure prediction 13 Epilepsy diagnosis EEG with epileptic seizure Prediction system specification Performance
- 14. Epilepsy 14 Neurological disorder characterized by sudden recurring seizures. Affecting 1% of world’s population. Second only to stroke. Frequently encountered in-flight medical events Unpredictable time and occasions. Second only to dizziness.
- 15. What happens today? 15 Diagnosis using electroencephalogram (EEG) Recording electrical activity of brain using multiple electrodes Machine learning techniques applied to classify EEG data Restricted to clinical environment
- 16. EEG with epileptic seizure 16 Preictal – the period before seizure onset occurs. Ictal – the period during which seizure takes place. Postictal – the period after the seizure ends. Interictal – the time between seizures.
- 17. Seizure diagnosis tasks 17 Task Requirements Application scenarios Seizure event detection greatest possible accuracy, not necessarily shortest delay. Apps. requiring an accurate account of seizure activity over a period of time. Seizure onset detection shortest possible delay, not necessarily highest accuracy. Apps. requiring a rapid response to a seizure. e.g., initiating functional neuro-imaging studies to localize cerebral origin of a seizure. Seizure prediction highest possible sensitivity, lowest possible false alarms, actionable warning time. Apps. requiring quick reaction to a seizure by delivering therapy or notifying a caregiver, before seizure onset.
- 18. Current approaches Pattern recognition issue Two-step processing strategy Feature extraction front-end Usually computationally expensive. Standard machine learning techniques Artificial neural networks; Decision trees; Mixture Gaussian models; Support vector machine (SVM). 18 Efficient signal analysis method that can produce physically meaningful and effective features is highly desirable!
- 19. Freiburg EEG database 19 Epilepsy Center, the University Hospital of Freiburg, Germany. Intracranial EEG data: recorded during invasive presurgical epilepsy monitoring. 21 patients: 8 males, 13 females. For each patient: at least 100 min preictal data + approximately 24 hr interictal data.
- 20. Stage Parameter Description Data At least 24 hr Duration of interictal record At least 150 min Duration of preictal record Feature extraction 5 sec EEG epoch length 6 Number of EEG channels Training 5 fold Cross validation SVM classification log2γ ~ [-10, 10] SVM radial basis function kernel parameter log2C ~ [-10, 10] Cost parameter 20 Classification
- 21. Sensitivity 95.2%: 79 out of 83 seizures predicted successfully; Perfect results for 16 out of 19 patients. Specificity 0.144 FAs per hour; Two-in-a-row post- processing: filtering out single positive detection. 21 Performance
- 23. EMG with proposed framework 23 Neuromuscular abnormality detection and muscular fatigue prediction. Long-duration spaceflight and absence of gravity greatly impacts astronauts’ neural-muscular system. Diagnosis using electromyogram (EMG). Indicate human’s physical status. Reflect electrical activity produced by skeletal muscles. Amplitude is closely related to muscle force.
- 24. Conclusion 24 Physiological cues as physical indicators in aerospace medicine predictive diagnostics has been investigated. Primary amplitude and frequency components. A new framework for improved medical operation autonomy during space missions has been developed. With state-of-the-art machine learning techniques. For disease prediction and health monitoring proposes. On a subject-by-subject basis. Promising epileptic seizure prediction performance in case study has been achieved.
- 25. 25
Editor's Notes
- #7: When the EEG is measured using non-invasive electrodes arrayed on an individual’s scalp it is referred to as scalp EEG; and when it is measured using electrodes placed on the surface of the brain or within its depths it is referred to as intracranial EEG. The EEG is a multichannel recording of the electrical activity generated by collections of neurons within the brain. Different channels reflect the activity within different brain regions. The scalp EEG is an average of the multifarious activities of many small zones of the cortical surface beneath the electrode.
- #13: Support vectors: data points closest to the hyperplane
- #15: Among the incidence of neurologic symptoms, seizures are considered to be the most frequently encountered events, second only to dizziness
- #24: muscle force is closely related to the amplitude of EMG signals