A Real-Time Seizure Classification System Using Computer Vision Techniques

Computer vision for automated seizure detection and classification: A systematic review

“…23,32,51,52,53,55,56,57 Tree-based methods Figure 3C Tree-based alogrithms are simple iterative methods often used in seizure analysis. 14,76 At each node of the tree, a feature is used to split the dataset into two groups based on whether they are above or below a threshold. Depending on the result, the algorithm then proceeds to another node, where the dataset is once again split into two groups based on another feature and another threshold, repeating until a classification can be made.…”

“…32,42,43,76 ANN Figure 3D FFNNs FFNNs are the simplest type of artificial neural network and were the first CV method used for automated seizure detection. 22,31,76,82 These networks pass information from node to node through activation and bias functions, culminating in a final layer that provides a weighted sum for classification. In seizure video analysis, FFNNs can be used to classify precomputed features (such as those from OF), but the curse of dimensionality prevents them from performing well on raw video data without prohibitively large datasets.…”

“…Pothula (2022) 76 Comparison of CV methods in distinguishing between gelastic and dacrystic seizures and no seizure using facial emotion recognition (FER) data.…”

“…41 Random forest and boosted tree models modify this paradigm, training an ensemble of models that interact with one another to avoid overfitting but often lose interpretability in the process. 32,42,43,76 ANN Figure 3D FFNNs FFNNs are the simplest type of artificial neural network and were the first CV method used for automated seizure detection. 22,31,76,82 These networks pass information from node to node through activation and bias functions, culminating in a final layer that provides a weighted sum for classification.…”

“…New studies incorporating larger datasets of real patients using more robust CV methods are lacking. CV methods have been used to detect gelastic (laughing) and dacrystic (crying) seizures with an accuracy of >98%, although with a very small dataset ( n = 4) consistent with their rare presentation 76 . Apnea detection is described later with neonatal epilepsy, 29 and has a richer literature in sleep medicine 77 …”