Please enable JavaScript.
Coggle requires JavaScript to display documents.
Traumatic Brain Injury, Critique: The paper lacked further explanation in…
Traumatic Brain Injury
Dataset availability
Major requirement for training Deep Neural Networks
Small datasets result to poor model performance
Small datasets cannot perform well on unknown data
Dataset for the experiments is relatively small
Divided into two groups
mTBI induced mice
Sham or controlled mice
EEG data capture for 24 hours
Awake signals
Sleep signals
Machine learning approach
Able to detect mild TBI presence
Able to accurately diagnose mTBI
Contribution of paper
Use of electroencephalogram (EEG) signals
Fluid Percussion Injury (FPI) induced mice
Portability
Use of Raspberry Pi 4
Previous works
Signal availability
Signal analysis
Robotics
Improvement in algorithms
Model and Experiments
System specifications
Raspberry Pi module
Analog to Digital Converter
Captures EEG signal data
Connects to the Raspberry Pi module
Digital to Analog Converter
For generating data
Terminal display
Shows classification performance
Process
EEG data pass through architecture
Epoch queues are created for each data
Signals are pre-processed
Features are extracted
Classification
Evaluated based on accuracy, precision, and recall
XGBoost outperformed CNN
Raspberry Pi Implementation has same results compared to a high performance computer
Diagnosis
Current analysis techniques are subjective
Computer Tomography Scan
Magnetic Resonance Imaging
Takes time to get results
Critique: The paper lacked further explanation in the pre-processing techniques employed and feature extraction used. Both are critical steps in training models.
Note: Multiple CNN architectures could have been used
Overall: Not fully convinced with the paper. Dataset is too low to generalize performance. Additionally, comparing only two models will not yield conclusive results.
Note: Metrics such as AUC and F1 score should have been considered.
Dhillon, N. S., Sutandi, A., Vishwanath, M., Lim, M. M., Cao, H., and Si, D. A raspberry pi-based traumatic brain injury detection system for single-channel electroencephalogram, 2021.