Alzheimer's Disease Detection via Machine Learning

By Ethan Zhu
Slide 1: Title slide for Alzheimer's Disease Detection via Machine Learning presentation

Slide-1

Scholar: Ethan Zhu

Mentor: Dr. Behnaz Ghoraani & Marjan Nassajpour Esfahani

Program: FAU Summer I-SENSE 2024

Health and Behavior: Next-Gen Health Monitoring Empowered by Python Programming and Deep Learning Applications

Slide 2: Table of contents showing four main sections of the presentation

Slide-2

Table of Contents

Introduction and Objectives

Data Preprocessing and Visualization

Feature Extraction and ML Classification

Findings / Conclusion

Slide 3: Existing challenges in EEG signal analysis including non-linearity, limited datasets, and feature extraction difficulties

Slide-3

Existing Challenges

EEG signals are highly non-linear and non-stationary, making them noisy and challenging to analyze.

Limited availability of public data sets restricts the ability to develop and validate models

lack of standardized international protocols, complicating consistent data collection and analysis

Extracting significant features from EEG data is difficult

Slide 4: Four main objectives including developing new algorithms, classification methods, validation, and visualization

Slide-4

Objectives

  1. Develop new algorithms to break down data for analysis
  2. Use existing algorithms to classify patients using data
  3. Validate existing methods for classifying patients
  4. Visualization of decomposed data signals and features

Introduce new data processing/analyzing techniques while using existing machine learning methods for sorting and classification

Slide 5: Introduction to Alzheimer's disease characteristics, symptoms, and detection methods

Slide-5

Introduction to Alzheimer's

Characterized by permanent degradation in brain neurons

100% Fatality Rate - 7th leading mortality rate in US

Symptoms: Memory loss, disorientation, behavior change, personality change

Active Methods rely on early detection to curb symptoms

Slide 6: Table of contents repeated showing focus on data preprocessing and visualization section

Slide-6

Table of Contents

Introduction and Objectives

Data Preprocessing and Visualization

Feature Extraction and ML Classification

Findings / Conclusion

Slide 7: Introduction to Electroencephalography (EEG) with brain diagram showing electrode placement and neural activity measurement

Slide-7

Data Preprocessing 1

Electroencephalography (EEG):

Measures electrical activity generated by neurons in the brain using electrodes placed on the scalp

Postsynaptic potentials of pyramidal neurons

High temporal resolution and non-invasive

Slide 8: Five-step data preprocessing pipeline including filtering, standardization, and artifact removal

Slide-8

Data Preprocessing 2

Step 1: Butterworth Band-Pass Filter (0.5-45 Hz)

Step 2: Standardizes signal by using A1 and A2 as references

Step 3: Automatic artifact rejection technique (ASR)

Step 4: ICA Method - RunICA Method

Step 5: Signal Processing

Slide 9: Signal processing methods including Empirical Mode Decomposition and Power Spectral Density analysis

Slide-9

Signal Processing

Empirical Mode Decomposition

Decompose non-linear and non-stationary signals into finite number of components

Sub-categories (EMD, EEMD, MEMD, NA-MEMD, etc)

Significant features capture without distorting time domain

Power Spectral Density

Measures the signal's power over the frequency domain.

Used with RBP for bandpower extraction

Used for understanding energy and power distribution

Slide 10: Visual representations of EEG signals for AD, HC, and FTD patient groups

Slide-10

Visuals

The slide displays EEG signal visualizations for three groups: AD (Alzheimer's Disease), HC (Healthy Controls), and FTD (Frontotemporal Dementia). The visualizations show different patterns of brain activity across these conditions, demonstrating the distinguishable characteristics that can be used for classification purposes.

Slide 11: Table of contents highlighting feature extraction and ML classification section

Slide-11

Table of Contents

Introduction and Objectives

Data Preprocessing and Visualization

Feature Extraction and ML Classification

Findings / Conclusion

Slide 12: Data overview table showing demographics and recording details for AD, FTD, and HC groups

Slide-12

Data Overview

19 scalp electrodes and 2 reference electrodes

AD FTD HC
Recorded Time
Total (min) 485.5 276.5 402
Range (min) [5.1, 21.3] [7.9, 16.9] [12.5,16.5]
Age
Average (yrs) 66.4 63.6 67.9
SD (yrs) 7.9 8.2 5.4
MMSE
Average (yrs) 17.75 22.17 30
SD (yrs) 4.5 8.22 0
Slide 13: Five-step data and feature abstraction process including PSD, MEMD, feature extraction, machine learning, and cross validation

Slide-13

Data and Feature Abstraction

Step 1: PSD - bandpower - extraction (alpha, beta)

Step 2: Multivariate Empirical Mode Decomposition

Step 3: Feature Extraction (see next slide)

Step 4: Machine Learning Models (SVM, XGBoost, R-forest)

Step 5: Cross Validation (LOSO, K-fold)

Slide 14: List of eight extracted features including LBP, Norm, Energy, HFD, KFD, LZC, MF, and HP

Slide-14

Features Extracted

  1. LBP - normalize bandpower
  2. Norm - overall power
  3. Energy - intensity
  4. HFD - complexity
  5. KFD - irregularity and self-similarity
  6. LZC - randomness via distinct patterns
  7. MF - dominant frequency
  8. HP (AMC) - temporal properties
Slide 15: K-fold validation methods and machine learning algorithms used including SVM, XGBoost, and Random Forest

Slide-15

Cross Validation

K-fold Validation

K = 10 (applied ML)

K = n (LOSO)

Machine Learning

Support vector machine (SVM)

XGBoost (XGB)

Random Forest (RF)

Slide 16: Table of contents highlighting findings and conclusion section

Slide-16

Table of Contents

Introduction and Objectives

Data Preprocessing and Visualization

Feature Extraction and ML Classification

Findings / Conclusion

Slide 17: Results table showing accuracy and F1 scores for different machine learning models across HC vs AD, HC vs FTD, and AD vs FTD comparisons

Slide-17

Findings / Conclusion

HC v. AD HC v. FTD AD v. FTD
Acc. F1 Acc. F1 Acc. F1
SVM 56% 71% 57% 71% 62% 75%
RF 65% 65% 68% 72% 63% 77%
XGBoost 65% 66% 66% 72% 62% 77%
Slide 18: Performance comparison table showing speedup improvements for different channel-sample configurations

Slide-18

Findings / Conclusion

Channel-Sample Original (s) New (s) SpeedUp
5-500 2.50 2.82 -11%
10-1000 5.27 3.43 54%
15-5000 22.64 12.56 80%
20-10000 86.83 31.69 174%
Slide 19: Summary of conclusions for both machine learning results and NA-MEMD algorithm performance

Slide-19

Summary of Conclusion

Machine Learning

Testing accuracy of 68% on small data

XGBoost provides good accuracy with faster execution.

RF delivers the best accuracy but with higher time overhead

NA-MEMD

New MEMD method is exponentially faster than alternative

Prototype for NA_MEMD with expansive noise options

Slide 20: Future improvements section outlining enhancements for preprocessing, bandpower, features, and ML classifiers

Slide-20

Future Improvements

Preprocessing

Alternative algorithms and artifact removal

Bandpower

Further bandpower decomposition [0-4Hz], [13-45Hz]

Features

More features focusing on multivariate relationships

ML + Classifier

Implement neural networks (CNN, RNN, etc)

Slide 21: Thank you slide with questions prompt

Slide-21

Thank you!

Questions?

Slide 22: Credits and references listing presentation template source and academic citations

Slide-22

Credit and References

  1. Li Z, Zhang L, Zhang F, Gu R, Peng W, Hu L. Demystifying signal processing techniques to extract resting-state EEG features for psychologists. Brain Science Advances. 2020;6(3):189-209. doi:10.26599/BSA.2020.9050019
  2. AlSharabi, K., Salamah, Y. B., Aljallal, M., Abdurraqeeb, A. M., & Alturki, F. A. (2023). EEG-based clinical decision support system for Alzheimer's disorders diagnosis using EMD and deep learning techniques. Frontiers in Human Neuroscience, 17. https://doi.org/10.3389/fnhum.2023.1190203
  3. Miltiadous, A., Tzimourta, K. D., Afrantou, T., Ioannidis, P., Grigoriadis, N., Tsalikakis, D. G., Angelidis, P., Tsipouras, M. G., Glavas, E., Giannakeas, N., & Tzallas, A. T. (2023). A Dataset of Scalp EEG Recordings of Alzheimer's Disease, Frontotemporal Dementia and Healthy Subjects from Routine EEG. Data, 8(6), 95. https://doi.org/10.3390/data8060095
  4. Zhang Y, Wang G, Li Z, et al. Matlab Open Source Code: Noise-Assisted Multivariate Empirical Mode Decomposition Based Causal Decomposition for Causality Inference of Bivariate Time Series. Front Neuroinform. 2022;16:851645. Published 2022 Jun 16. doi:10.3389/fninf.2022.851645
  5. Miltiadous A, Tzimourta KD, Giannakeas N, Tsipouras MG, Afrantou T, Ioannidis P, Tzallas AT. Alzheimer's Disease and Frontotemporal Dementia: A Robust Classification Method of EEG Signals and a Comparison of Validation Methods. Diagnostics (Basel). 2021 Aug 9;11(8):1437. doi: 10.3390/diagnostics11081437. PMID: 34441371; PMCID: PMC8391578.
  6. Zhu E, Health and Behavior: Next-Gen Health Monitoring Empowered by Python Programming and Deep Learning Applications.(2024). GitHub Repository https://github.com/PiethonProgram/NA-MEMD
Last slide: Contains plain text stating 'End of presentation. Click the right arrow to return to beginning of slide show.'

End of Presentation

Click the right arrow to return to the beginning of the slide show.

For a downloadable version of this presentation, email: I-SENSE@FAU.

Additional Information
The Institute for Sensing and Embedded Network Systems Engineering (I-SENSE) was established in early 2015 to coordinate university-wide activities in the Sensing and Smart Systems pillar of FAU’s Strategic Plan for the Race to Excellence.
Address
Florida Atlantic University
777 Glades Road
Boca Raton, FL 33431
i-sense@fau.edu