Table 1.
Seizure detection characteristics for the 19 channels (S19) and three subsets with a smaller number of electrodes.
| EEG SETUP | OVR [%] | PERR (FP, MT) [%, -, -] | |||
|---|---|---|---|---|---|
| P | C | Y | T | ||
| Synchronization and normalized amplitude | |||||
| S 19 | 82.9 | 0.87 (1437, 832) | 0.12 (44, 32) | 0.07 (7, 6) | 0.36 (1488, 870) |
| S 12 | 78.01 | 0.71 (1147, 735) | 0.23 (77, 58) | 0.13 (13, 11) | 0.36 (1237, 804) |
| S 6 | 79.36 | 0.86 (1282, 775) | 0.05 (18, 14) | 0.04 (4, 4) | 0.32 (1304, 793) |
| S 4 | 69.31 | 0.68 (1196, 743) | 0.03 (11, 9) | 0.02 (2, 2) | 0.25 (1209, 754) |
| Synchronization and normalized amplitude with post-processing | |||||
| S 19 | 80.93 | 0.57 (985, 381) | 0.13 (21, 9) | 0.03 (4, 2) | 0.24 (1010, 392) |
| S 12 | 75.56 | 0.40 (652, 267) | 0.22 (35, 15) | 0.03 (4, 2) | 0.22 (691, 284) |
| S 6 | 76.85 | 0.55 (868, 311) | 0.06 (5, 2) | 0.00 (0, 0) | 0.21 (873, 313) |
| S 4 | 65.10 | 0.38 (746, 278) | 0.03 (4, 2) | 0.00 (0, 0) | 0.13 (750, 280) |
We used a k-nearest neighbor classifier with the synchronization index and normalized amplitude as the features. The overlap (OVR) of segments classified as ictal with absence seizures and relative duration of false positives (PERR) are presented for patients (P), controls (C), young adults (Y), and for segments from all cohorts (T). In parentheses, we give the number of distinct trains of misclassified windows (MT) and false positives (FP). In patients, false detections are predominantly caused by short (<2 s) epileptiform discharges. Therefore, we also tested the possibility of reducing the number of false positives by post-processing the k-nearest neighbors classification. In particular, any isolated ictal segment was labeled non-ictal.