Epilepsy Detection and Monitoring - Quantitative EEG Analysis Methods and Clinical Applications - page 161

is not available when using the technique in an online manner so it cannot be
generalized to online use.

6.13.1 Optimality Index
From these four totals (TP, TN, FP, FN) we can calculate two statistics that give a
large amount of information regarding the success of a given technique. The first
statistic is the sensitivity ( S ), which is defined in (6.26). In detection this indicates the
probability of detecting an existent seizure and is defined by the ratio of the number
of detected seizures to the number of total seizures. In prediction this indicates the
probability of predicting an existent seizure and is defined by the ratio of the num-
ber of predicted seizures to the number of total seizures.

TP

S = (6.26)

TP + FN

In addition to the sensitivity, the specificity ( K ) is also used and is defined in
(6.27). This indicates the probability of not incorrectly detecting/predicting a sei-
zure and is defined by the ratio of the number of interictal segments correctly identi-
fied in comparison to the number of interictal segments.

TN
TN

K = (6.27)

+ FP

A third metric used to measure the quality of a given algorithm is the predict-
ability. This indicates how far in advance of a seizure the seizure can be predicted or
how long after the onset of the seizure it can be detected. In other words, the predict-
ability (

T T a T e where T a is the time at which the given algo-
rithm detects the seizure and T e is the time at which the onset of the seizure actually
occurs according to the EEG.
Note that either of these metrics alone is not a sufficient measure of quality for a
seizure detection/prediction technique. Consider a detection/prediction algorithm
that always said the signal was in the ictal or preictal state, respectively. Such a
method would produce a sensitivity of 1 and a specificity of 0. On the other hand,
an algorithm that always said the signal was in the interictal state would produce a
sensitivity of 0 and a specificity of 1. The ideal algorithm would produce a value of 1
for each. To accommodate this, Talathi et al. [48] defined the optimality index ( O ),
a single measure of goodness, which takes all three of these metrics into account. It
is defined in (6.28), where D *

Δ T ) is defined by Δ

is the mean seizure duration of the seizures in the

dataset:

SK + Δ * T

D

O = − (6.28)

2

6.13.2 Specificity Rate
The specificity rate is another metric used to assess the performance of a seizure pre-
diction/detection algorithm [50]. It is calculated by taking the number of false pre-