This thesis presents a semi-automatic detector of high frequency oscillations (HFO) on scalp EEG of epileptic patients. This detector aims at helping the expert in finding all the HFOs on an EEG recording, a difficult task considering the recent interest in this field of study, and the subjectivity required in the visual detection. The detector will have the function of limiting this subjectivity, in addition of providing a time gain compared to the currently used visual detection method.
A quantitative performance computation will be possible because of a database of eight patients marked by an expert. The detector will be evaluated on its ability to detect expert’s markers (sensitivity) while limiting false detections (FPR). A base detector is also provided as a starting point, and will serve as a second reference for evaluation of the detector’s performance.
The principal contributions presented in this thesis are the following: An analysis of the expert’s markers that aims to find common characteristics that would allow for a more sensitive and specific detection of HFOs, a learning method of thresholds that will adapt the detector’s parameter for a specific patients based on a subsample of his markers, and the implementation of a scoring system that helps separate true positives from false positives.
The results of the semi-automatic detector show a net performance gain in comparison to the initial detector. The expert’s markers are detected at 77% with a false positive rate of 10 per minute. This performance is an improvement in both sensitivity and specificity compared to what has been previously obtained with other approaches.
| Date | 8 Feb 2016 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Jean-Marc Lina (Supervisor) |
|---|
Beaudry, J. (Author),
Lina (Supervisor),
8 Feb 2016Student thesis: Master's thesis › Master in Engineering: Engineering