Automatic Detection of Epileptic Waves in Electroencephalograms Using Bag of Visual Words and Machine Learning

Muñoz, Marlen Sofía; Torres, Camilo Ernesto Sarmiento; López, Diego M.; Salazar-Cabrera, Ricardo; Vargas-Cañas, Rubiel

doi:10.1007/978-3-030-59277-6_15

Cited by 1 publication

(5 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To perform this work and others related [5,18,[29][30][31], 100 pediatric EEGs were collected from the same number of children aged between 22 days and 17 years old, suspected of suffering epilepsy. The exams were performed with the patients asleep, on the recommendation of the neurologist.…”

Section: Data Collection: Acquisition Of Encephalogramsmentioning

confidence: 99%

“…In this step, the k-fold method was used with cross-validation to obtain the best model from different performance indices. This methodology guarantees that the results are independent of training and testing datasets [18]. First, a dataset partition was performed.…”

Section: Model Evaluationmentioning

confidence: 99%

“…These, plus the normal class, are the seven final classes considered at this stage. Figure 4 are independent of training and testing datasets [18]. First, a dataset partition was performed.…”

Section: Second Stage: Multi-class Classificationmentioning

confidence: 99%

“…Approaches to detect epileptiform events using ML include biomedical signal processing, analysis of characteristics extracted from the signals, and analysis of images in a lesser proportion [17]. The works of Molina et al [5] and Muñoz et al [18] provide an example of the use of ML for the detection of epileptiform events. Molina et al developed an intelligent component to automatically detect abnormal segments of EEG tests using conventional machine learning algorithms over a dataset generated from EEG signals [5].…”

Section: Introductionmentioning

confidence: 99%

“…Molina et al developed an intelligent component to automatically detect abnormal segments of EEG tests using conventional machine learning algorithms over a dataset generated from EEG signals [5]. Muñoz et al present a machine learning-based methodology using a visual bag of words taken from raw EEG images as input to identify images with abnormal signals [18]. Although ML provides exciting results, it has many drawbacks due to the many channels used, the low amplitude per channel, and the non-stationarity of each channel in the signal [19].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Digital Transformation in Epilepsy Diagnosis Using Raw Images and Transfer Learning in Electroencephalograms

et al. 2022

Self Cite

View full text Add to dashboard Cite

Epilepsy diagnosis is a medical care process that requires considerable transformation, mainly in developed countries, to provide efficient and effective care services taking into consideration the low number of available neurologists, especially in rural areas. EEG remains the most common test used to diagnose epilepsy. In recent years, there has been an increase in deep learning techniques to analyze electroencephalograms (EEG) to detect epileptiform events. These types of techniques support the epilepsy diagnostic processes performed by neurologists. There have been several approaches such as biomedical signal processing, analysis of characteristics extracted from the signals, and image analysis to detect epileptiform events. Most of the works reported in the literature, which use images, transformed the signals into a two-dimensional space interpreted as an image. However, only a few of them use the raw EEG image. This paper presents a computational model for detecting epileptiform events from raw EEG images, using convolutional neural networks and a transfer learning approach. To perform this work, 100 pediatric EEGs were collected, noting six characteristics of epileptiform events in each exam: spikes, poly-spikes, spike-and-wave, sharp waves, periodic, and a combination of them. Then, pre-trained convolutional neural networks were used, which, through transfer learning techniques, were retrained to classify possible events. The model’s performance was evaluated in terms of precision, accuracy, and Mathews’ correlation coefficient. The model offered a performance above 95% accuracy for binary classification and above 87% for multi-class classification. These results demonstrated that identifying epileptiform events from raw EEG images combined with deep learning techniques such as transfer learning is feasible. Significance: The proposed method for the evaluation of EEG tests, as a support tool for the diagnosis of epilepsy, can help to reduce the time of reading EEGs, which is very important, especially in developing countries with a limitation of a specialist in neurology.

show abstract