An improved feature extraction algorithm of EEG signals

🎓 EEG Motor Imagery Signal Classification using ICA, Wavelet Denoising, CSP & LDA
📅 M.Tech Final Year Mid-Semester Project Evaluation

This presentation covers the complete EEG motor imagery signal processing pipeline — from raw signal acquisition to classification.

🧠 Key Highlights:

Introduction to EEG and Brain-Computer Interface (BCI)

Dataset: PhysioNet EEG Motor Movement/Imagery Dataset (64 channels, 160 Hz)

Preprocessing using ICA and Wavelet Denoising (Coiflet-5)

Feature Extraction using CSP (Common Spatial Patterns)

Classification using LDA (Linear Discriminant Analysis)

Performance evaluation with precision, recall, F1-score, confusion matrix, and SNR improvement of 18.04 dB after artifact removal.

🚀 Outcome:

Significant noise reduction through ICA + Wavelet Transform.

Feature dimensionality reduced from 64 channels to 4 discriminative CSP components.

Achieved ~64% classification accuracy on full dataset using LDA.

📊 This work demonstrates a complete EEG-based BCI pipeline suitable for motor imagery classification—a foundation for applications like neuroprosthetics, rehabilitation, and assistive technologies.

#EEG #BCI #MTech #CSP #LDA #Wavelet #MachineLearning #SignalProcessing #MidSemProject #Neuroscience #Research #BrainComputerInterface

1. NATIONAL INSTITUTE OF TECHNOLOGY,SILCHAR An improved feature extraction algorithm of EEG signals DEPARTMENT OF ELECTORNICS & COMMUNICATION BY Anirban Nath (2424223) M.Tech (3rd Sem) A Presentation on Guided By – Dr. Koushik Guha

2. Contents Introduction to EEG Signal Introduction to BCI Methodology Result and Discussion Future Work Conclusion References Objectives Dataset Description Literature Review

3. EEG = Electroencephalogram Records electrical activity of the brain using scalp electrodes. Reflects real-time brain function and neural communication. Introduction to EEG Signals

4. Low amplitude: 10–100 μV Frequency range: 0.5 – 100 Hz Non-stationary and time-varying High temporal resolution (captures millisecond-level activity) Key Characteristics: Fig-Raw EEG Signal (01)

5. Introduction to BCI BCI = Brain-Computer Interface Connects the human brain directly to external devices Converts brain signals into commands Helps users control devices using thoughts only Useful for patients with motor disabilities

6. S.No. Authors Name Method Adopted 1. Ying Geng, Zhenhao Zhang, Zhe Chen, Xiaoxuan Ma, Peiying Zhang, and Lei Wang Appl ied ICA for artifact removal, used wavelet packet decomposition for denoising, extracted spatial features using CSP, and classified using machine learning algorithms. 2. Pooja Pawar, Snehal Dhamangaonkar, and Kalyani Bansod EEG S ignal Processing and Feature Extraction Desc ribed preprocessing techniques, extracted spectral, statistical, entropy, and spatial features using CSP, and applied both machine learning and deep learning methods for classification. 3. Hoss am Mohammad, Raghad Kh. Ismaeel, and Ahmed Al-Araji F eature Extraction from EEG Signals: A Deep Learning Perspective Review ed traditional signal decomposition methods such as DWT and EMD, applied feature selection techniques, and utilized deep learning models like CNNs, RNNs, and EEGNet for feature learning. An improved feature extraction algorithms of EEG signals based on motor imagery Literature Review

7. Obje ctive Project Goals: Replicate EEG signal pr ocessing pipeline using Python (MNE, PyWavelets, sklearn). Implement ICA + Wavelet + CSP + LDA model for motor imagery classification. Evaluate classification accuracy for Left vs Right Hand mental imagery.

8. Dataset Description Fig 3: Electrodes arrangement position.[1] Dataset: PhysioNet EEG Motor M ovement/Imagery (EEGMMIDB) Subjects: 109 volunteers Channels: 64 Tasks: Left vs Right Hand Imagery Sampling Rate: 160 Hz File Used Initially: S001R03.edf Final Model: Trained on entire dataset (S001–S109)

9. Methodology Overview 1 ICA: Separate signal components 2 Wavelet Denoising (Noise Suppression) 3 CSP: Feature Extraction 4 LDA (Classification)

10. Independent Component Analysis (ICA) Purpose Separate mixed brain signals Artifact removal Eliminate EOG & ECG interference

11. Fig: The Mathemetical model of ICA .[1] Observed signal: X(t) = A · S(t) Output: Y(t) = W · X(t) Where, X(t): Observed Signal A:Mixing Coefficient Matrix S(t):Source Signal S(t)= A^(-1) . X= WX ICA Mathematical Model

12. ICA Fig 3: Electrodes arrangement position.[1]

13. Wavelet Transform (WT) in EEG Purpose Denoising EEG signals Technique Decompose signals in frequency domain Thresholding Remove noise components effectively

14. Procedure 1.Wavelet Transform EEG signals converted to Wavelet Coefficient . High Frq. Coefficient ~ Noise Low Frequency Coeff. ~Real Signal 2.Thresholding the Coefficient Remove or shrink the High wavelet coefficient (assumed to be noise) 3.Inverse Wavelet Transform Modified wavelet coefficient are transform back to get clean signal

15. Wavelet Transform

16. Wavelet Transform Fig 3: Electrodes arrangement position.[1]

17. ICA + Wavelet Transform before Avg SNR=9 dB After Avg SNR= 27.40 dB Avg SNR Improvement =18.40dB

18. Common Spatial Pattern (CSP) Feature Extraction algo CSP maximize variance between two classes. Used to distinguish between two mental task. Example -Imagining Left Hand Vs Right Hand Movement

19. Common Spatial Pattern (CSP) C ommon Spatial Pattern (CSP) Applied on Subject 01

20. Common Spatial Pattern (CSP) C ommon Spatial Pattern (CSP) Applied on entire Dataset

21. Classification (LDA) Used Linea r Discriminant Analysis (LDA) as the classifier. LDA finds a projection line that maximally separates classes.. 5-fold cross-validation accuracy ≈ 64.1% . L R LDA on Subject 01

22. LDA on Entire Dataset L R

23. Proposed Workflow

24. Imp rovements from Original Paper Reference Paper (1) Propsed Model Aspects

25. Results Summary ICA-WT-CSP-LDA approach: Mean Accuracy -64.10% Robust to noise Preserves useful motor features

26. Future Work 1 Improve recognition accuracy 2 Tun e classifier (try SVM, CNN) 3 FPGA implementation of CSP + LDA.

27. Conclusion Proposed method is effective Combines strengths of ICA, WT, CSP & LDA. Ac hieved ~64% accuracy

28. References [1]Xiaozhong Geng, Dezhi Li, Hanlin Chen, Ping Yu, Hui Yan, Mengzhe Yue, An improved feature extraction algorithms of EEG signals based on motor imagery brain-computer interface,Alexandria Engineering Journal,Volume 61, Issue 6,2022,Pages 4807-4820,ISSN 1110-0168, https://doi.org/10.1016/j.aej.2021.10.034. [2]International Journal for Modern Trends in Science and Technology Volume 9, Issue 08, pages 45-50 ISSN: 2455-3778 online, DOI: https://doi.org/10.46501/IJMTST0908008 [3] A. Mohammad, F. Siddiqui and M. Afshar Alam, "Feature Extraction from EEG Signals: A deep learning perspective," 2021 11th International Conference on Cloud Computing, Data Science & Engineering (Confluence), Noida, India, 2021, pp. 757-760, doi: 10.1109/Confluence51648.2021.9377108. [4] ] Z. Ling, C. Shuyue, S. Yuqiang, M.a. Zhe, Extraction of Evoked Related Potentials by using the Combination of Independent Component Analysis and Wavelet Analysis, J. Biomed. Eng. 27 (4) (2010) 741–745. [5] International Conference on Computational Intelligence and Networks. IEEE, (2016) 84-89

29. Thank You Any Questions?