Figure 1.
Sample template waveforms collected from a training dataset.
To find inter-ictal epileptiform activity in new unseen EEGs, each template searches for matching waveforms with high correlations to themselves. Thereafter, individual template detections are combined to form “IED nominations” which are then presented to the reviewer in decreasing order of likelihood.
Figure 2.
Outline of the IED detection, grouping, and presentation steps.
Multiple detections of inter-ictal activity are made using a database of matching template waveforms. Individual template detections are merged and grouped (see Fig. 3) to form IED nominations, and the nominations are presented for review in an iterative manner, ordered by nomination certainty.
Figure 3.
Grouping and merging of template detections.
Overlapping templates are merged together and grouped to form single IED nominations, which are presented to the reviewer as a single event in time. Channels containing template detections are highlighted to point out where inter-ictal activity was found.
Table 1.
Subject information of the EEGs used to test the detection and reviewing method.
Figure 4.
A screenshot showing how the review process works.
For each iteration, the reviewer is presented with 10 nominations, one at a time, and is given the opportunity to either confirm, reject, or ignore each event as inter-ictal epileptiform activity (see top right panel). After the ten nominations have been graded, the user can choose to either stop with the review or to review another ten events. Nomination certainties are updated between iterations.
Table 2.
Adaptive iterative reviewing: IED nominations are iteratively presented to the reviewer one-by-one using 10 events per iteration.
Figure 5.
Template contributions per recording: The first bar shows the number of templates that have contributed to all the nominations presented during a review of fifteen iterations.
The second bar shows the number of templates that contributed to confirmed IEDs only.