1. NeuroPhone: Brain-Mobile Phone Interface using a Wireless EEG Headset Source: MobiHeld 2010 Presented By: Corey Campbell

2. INTRODUCTION A new way to use the mobile phone Design and Evaluation of NeuroPhone. – EEG headset – iPhone Two different EEG signals to trigger action Challenges involved

3. BRAIN-MOBILE PHONE INTERFACE Mobile apps can be reinvented – Driving example – Many-to-One apps Teacher – Student example Possibility of Group Emotional State – Meeting example Happy Sad Bored Hostile

4. BRAIN-MOBILE PHONE INTERFACE (cont.) Challenges regarding EEG headsets – Research-grade, hard-wired headsets Offer more robust signal Very expensive Not mobile – Gaming headsets Cost is cheaper Encrypted wireless interface More noise in signal

5. BRAIN-MOBILE PHONE INTERFACE (cont.) More challenges – Mobile phones not designed for continuous neural sensing applications Streaming neural info wirelessly and phone processing Where do we use mobile phones, noisy? Filtering out external noises

6. NEUROPHONE DESIGN App titled “Dial Tim” – Think & Wink modes Contacts from iPhone address book User concentrates on a person to call P300 neural signal is the trigger Wink mode uses a left or right wink to trigger The P300 is subtle compared to a wink

7. WHAT IS THE P300? Focus on a person to call When highlighted by app causes brain to produce particular EEG signal – Positive peak – 300ms latency from onset of stimulus Neuroscience uses this as P300 Other neural signals have potential

8. WIRELESS EEG HEADSET Emotiv EPOC headset – 14 data-collecting electrodes – 2 reference electrodes – International 10-20 system config. Transmits encrypted data Windows-based 2.4Ghz frequency range

9. WIRELESS EEG HEADSET (cont.) Can detect facial expressions Training then detection of activities – Push, pull, rotate, lift Gyroscope Headset not totally reliable – Challenge to extract finer P300 signals Still, it is very useful and cost is cheap to deploy on large scale

10. DESIGN CONSIDERATIONS Signal to Noise Ratio (SNR) – Lots of noise on every electrode – Bandpass filtering – Average multiple trials of data Signal Processing – Bandpass filtering

11. DESIGN CONSIDERATIONS (cont.) Phone Classifiers – Classification algorithms designed for powerful machines – Algorithms not practical to run on mobile phones Power efficiency Resource issues – Resolving issues Provide relevant subset of EEG channels Use lightweight classifiers

12. EVALUATION Tested think and wink modes in various scenarios – Sitting, walking, etc Wink mode performance – Declines with really noisy data – Handles reasonably noisy data well

13. EVALUATION (cont.) Think mode performance – Accuracy is higher as more data is averaged – P300 signals susceptible to external noise – Sitting still provides best results – Accuracy declines more when person stands up More data accumulation and averaging provides better detection accuracies

14. EVALUATION (cont.) Ongoing work – Usable P300 data from a single trial – Find new algorithms to handle extra noise iPhone app usage stats – CPU = 3.3% – Total memory = 9.40MB 9.14MB for GUI – Battery drain