Can we measure depression
summary
Using 19 EEG channels and 10–15 minute steady‑state recordings, EEGs are cleaned with filtering and ICA plus Independent Component Label to remove blinks and muscle artifacts, then five feature types (wavelet energy, absolute/relative power, spectral centroid, wavelet entropy) produce 475 features per patient. With 15 major depressive and 15 healthy controls SVM and k‑NN reach 76% accuracy (7/30 wrong); sequential feature selection cuts to 4 features and 4 misclassifications; a CNN on raw left/right hemisphere signals tops 80% with CNN. A classifier cannot separate MDD from MDD+BPD.
AI-written summary · about
chapters (12)
- 0:00Start
- 8:29Electroencephalography
- 8:57Why Did We Choose this Method
- 9:13Characteristic Brain Rhythms
- 12:00Acquisition
- 13:06Independent Component Analysis
- 14:28Feature Extraction
- 15:29Results
- 16:09Sequential Feature Selection
- 21:18Training Size and Testing Size
- 24:04Single Biomarker for Depression
- 28:06Certificates of Appreciation
related talks
32mIrina Rish, Research Staff, IBM T.J. Watson Research Center
06mLeveraging Collaborative-Filtering for Personalized Behavior Modeling: A Case Study of ...
26mBrainDrain Using Machine Learning and Brain Waves to Detect Errors in Human
05mMaking Sense of Sleep: Multimodal Sleep Stage Classification in a Large, Diverse Population ...
01h 01mKeynote: Using EEG and Machine Learning to Perform Lie Detection
36mMachine-Brain Intelligence in Brain-Computer Interfaces
34mIrina Rish, Researcher, IBM Watson, NYC 2017
13mEnhanced EEG Data Classification with 3D SNN and On-Line Learning ESN