Page 125 - 2024-Vol20-Issue2

121 | Assim & Mahmood

TABLE III.
THE ADVANTAGES AND DISADVANTAGES OF THE RECHERCHE LITERATURE

Ref Advantages Disadvantages

1. Ignores the correlation between decision probability

1. The kernel size in the first layer controls and crucial frequency components in the internal

retrieved features interpretability states of the network.

and the trained models sensitivity. 2. Mishandling the categorization of distinct seizure

[40] 2. Amplitude is the most significant feature in ictal sub-populations.

prediction. 3. Signals from improperly classified segments

3. Learning more complex frequency patterns are often totally ictal or entirely interictal,

would require a larger patient population. with no clear transition between the immediately preceding

pre-itcal segments and seizure onsets.

1. The smallest variance and the best classification

accuracy are produced by hybrid input. The classification accuracy performance deteriorated as the
[41] 2. Depth-wise separable convolution to reduce training sample count decreased.

the parameters in the network.

3. Utilize regularization.

1. The supervised deep convolutional autoencoder

(SDCAE) training is faster than typical

semi-supervised systems. 1. Deep learning requires a large dataset;
2. The number of parameters is reduced because selecting 16 out of the 23 pediatric patients
[43] it uses convolutional layers instead of will increase the detection ratio.
fully connected layers to learn features. 2. Do Not use any de-noising method to clean EEG signals.
3. Auto Encoder (AE) supervised

training is more effective in learning.

4. Plotting with s(1s, 2s, 4s) time segments.

1. Increased phase synchronization and the sample 1. Detecting epilepsy by traditional methods.

[44] entropy improve detection. 2. select 23 from 24 pediatric patients using a small dataset.

2. Using ICA and correlation p-value. 3. The noise type is not mentioned.

1. The results demonstrate that the single channel

algorithm has an accuracy=86%. 1. Training Parameters: affect the training process’s speed.

[45] 2. The multichannel combination technique 2. The Time-size frequency. It’s unclear whether

improves accuracy by about 4% and the scaling process impacts how the model was trained.

raises the TPR to 96.5%.

1. Using 2-dimensional visual data based on 1. The requirement for a GPU computer with significant
memory and processing capability and accompanying
graphical monitoring ensures high accuracy rates. computational expense
2. Image noise suppression was skipped.
[46] 2. The extraction of features is done automatically. 3. A small quantity of training and testing data was used.
3. With 14 million data points in

1,000 categories, the Alex Net CNN model

was successfully trained.

1. Provide patients with epilepsy with a trustworthy

[47] diagnosis of their seizures. 1. Performance declines in the absence of augmentation.
2. The Epileptic-Net model performs 2. The difficulty in labeling EEG samples and their rarity.

well when data are added.

1. Low detection delay (1.0431s) is a feature of the 1. Among other pertinent aspects, statistical and nonlinear

suggested technique. patterns in EEG data can be employed to

[48] 2. The detection success rate is 99.95%, meaning all enhance detection.

epileptic events may be 2. A more effective multi-view learning

identified in less than 10 seconds. process is required.