![Institutional Repository [SANDBOX]](images/dias.logo.png "Institutional Repository [SANDBOX]"){kind=logo}
Institutional Repository [SANDBOX]
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/B2155AD8-0CBB-4EC1-A4CC-AFB62926A585 | ||
| Year | 2022 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Christina Paschalidou, "High frequency oscillations in EEG for the assessment of Epilepsy", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2022 https://doi.org/10.26233/heallink.tuc.91911 | ||
| Appears in Collections |
The visual detection of High Frequency Oscillations (HFOs) by experts is a time demanding process, making automatic detection methods a necessity of major importance. Most of the methods that have been developed over the years, filter the signal and then detect whether special features of interest, such as energy, are above a given threshold. Recently, there has been proposed an HFO detector that is based on semi-supervised clustering. The idea is that the centers of HFOs and non-HFOs classes are determined by a small set of visually inspected HFO and non-HFO segments, and the rest of the segments are assigned to the closest center. Depending on the clustering algorithm, the centroids of clusters may be updated by the non-labeled segments. The main aim of this research project is to compare a semi-supervised HFO detector that uses the seeding k-means, with three threshold detectors, namely: the Root mean square detector, the Line-length detector, and the Hilbert detector. Encouraged by the success of the semi-supervised clustering to detect HFOs, there is a further investigation on whether a totally unsupervised clustering algorithm can be used to discriminate between HFO and non-HFO segments. In this regard there is an examination of the k-means algorithm at first, followed by the DEC algorithm, where an encoder, which reduces the dimensionality of the feature space, and the centers of the clusters are learned simultaneously. As far as we know, this is the first work that uses unsupervised clustering for HFO detection. Although some wrongly classified cases have been identified, the results with DEC are very encouraging. This study suggests that it is worth investigating whether more advanced and deep learning clustering algorithms in HFO detection could reach the performance of human experts.
Copyright © DIAS 2013
