TABLE 3.
The methods of using artificial intelligence for epilepsy-assisted localization in MEG data.
| References | Moda-lity | Problem that was solved in that study | Database | Data acquisition | Source locali-zation | Fea-tures | Classifi-cation | Performance metrics | |||||||||
| Sample size | Age range |
Sex (M:F) |
Source | Total dura-tion | Cha-nnels |
Seg-ment/ epoch length |
Fre-quency samp-ling | Pre-process-ing |
Sensiti-vity (%) |
Specifi-city (%) |
Accu-racy (%) |
||||||
| Khalid et al., 2016b | MEG | spike detection | 20 Eps | / | / | KFMC | 15 min | 306 | / | 1,000 Hz | band-pass filter: 1–50 Hz | / | CSP features | CSP-LDA | 91.03 | 94.21 | / |
| Alotaiby et al., 2017 | MEG | spike detection | 30 Eps | 14–43 | 22:8 | KFMC | 15 min | 306 | 100 ms | 1,000 Hz | band-pass filter: 1–50 Hz | / | statistical features | KNN | 91.75 | 92.99 | / |
| Khalid et al., 2017 | MEG | spike detection | 28 Eps | 14–43 | / | KFMC | 15 min | 306 | 100 ms | 1,000 Hz | band-pass filter: 1–50 Hz | / | amplitude threshold-based features | Dynamic Time Warping (DTW) | 92.45 | 95.81 | / |
| Chahid et al., 2019 | MEG | spike detection | 8 Eps, 8 HCs | / | / | KFMC | 15 min | 306 | sliding window of size 100 sample-points with a step of 2 sample-points | 1,000 Hz | band-pass filter: 1–50 Hz | / | Semi-Classical Signal Analysis (SCSA) method-based features | SVM | 92.52 | 89.1 | 90.88 |
| Chahid et al., 2020 | MEG | spike detection | 8 Eps, 8 HCs | / | / | KFMC | 15 min | 306 | sliding window of size 100 sample-points with a step of 2 sample-points | 1,000 Hz | band-pass filter: 1–50 Hz | / | QuPWM-based features | SVM | 87 | 97 | / |
| Sdoukopoulou et al., 2021 | MEG+ EEG |
spike detection | 1 Eps | 20 | female | / | 8 min | 304 | 400 ms | 2,400 Hz | band-pass filter: 1–100 Hz | / | EMEG features (statistical, spectral, functional connectivity metrics) | SVM | 95.1 | 90.2 | 92.8 |
| Kaur et al., 2022 | MEG | spike detection | 20 EPs | 15–52 | / | Magnetoence-phalography Center of Xuanwu Hospital of Capital Medical University | 60 min | 306 | 10 s | 1,000 Hz | band-pass filter: 0.1–500 Hz | / | Phase locking value (PLV) | SVM | / | / | 93.8 |
| Zheng et al., 2019 | MEG | spike detection | 20 focal Eps | 10–49 | 11:9 | the Sanbo Hospital of Capital Medical University, Beijing, China |
10 min (90 min) | 306 | 300 ms | 1,000 Hz | band-pass: 1–100 Hz | / | / | EMS-Net | 91.61–99.53 | 91.60–99.96 | 91.82–99.89 |
| Hirano et al., 2022 | MEG | spike detection | 375 EPs | 0–79 | 1:1 | Osaka University hospital | 4 or 5 min | 160 | 2,048 ms | 1,000 Hz or 2,000 Hz | band-pass filter: 3–35 Hz; downsampled: 1,000 Hz | / | / | SE-ResNet + DeepUNet | 79.52 | 99.71 | / |
| Guo et al., 2018 | MEG | HFO detection | 20 EPs | 6–60 | 1:1 | / | 60 min | 306 | 2 s | 2,400 Hz | band-pass filter: 1–70 Hz, 80–250 Hz, 250–500 Hz; down-sample factor: 10 | / | SSAE model-based features | SMO | 88.2 | 91.6 | 89.9 |
| Guo et al., 2020 | MEG | HFO detection | 20 EPs | 6–60 | 1:1 | / | 60 min | 306 | 1,000 ms | 4,000 Hz | band-pass filter: 80–250 Hz, 250–500 Hz | / | / | ARF-AttNN | 82.6 | 92.7 | 89.3 |
| Liu et al., 2020 | MEG | HFO detection | 20 EPs | 6–60 | 1:1 | / | 60 min | 306 | 500 ms | 2,400 Hz | band-pass filter: 80–250 Hz, 80–500 Hz | / | / | MEGNet | 94 | / | 94 |
| Tanoue et al., 2021 | MEG | HFO detection | 16 left mTLE, 19 right mTLE | 8–71 | 2:3 | Osaka City University Hospital | 5 min | 160 | 10 s | 1,000 Hz | band-pass filter: 0.3–200 Hz | COH algorithms imple-mented in SPM-12, which is similar to sLORETA | the laterality index (LI) in | SVM | 68–75 | 96 | 91 |
| Guo et al., 2022 | MEG | HFO detection | 20 EPs | 6–60 | 1:1 | / | 60 min | 306 | 1 s | 2,400 Hz | band-pass filter: 1–70 Hz, 80–500 Hz | / | / | TransHFO | 92.86 | 100 | 96.15 |