Table 1. ANN model performance as compared to previous automated sleep scoring approaches to classify rodent behavioral state.
| Source | Approach | Overall accuracy (%) | Manual scoring (% of total epochs) | Freely-available, open source (environment) | Include epochs with visually-scored artifact? |
|---|---|---|---|---|---|
| Current manuscript | Artificial neural network | 91 | 2.6 | Yes (R) | Yes |
| Current manuscript | Artificial neural network | 92 | 2.6 | Yes (R) | No |
| Current manuscript | Artificial neural network + manual rescore |
93 | ~14 | Yes (R) | Yes |
| Current manuscript | Artificial neural network + same rat generalization |
89 | 2.6 (training day) 0 (other test days) |
Yes (R) | Yes |
| Exarchos et al., 2020 | Convolution neural network | 93 | ~15 | Yes (Google colab) | Yes |
| Exarchos et al., 2020 | Dimension reduction + clustering | 89 | 0; unsupervised | Yes (Google colab) | Yes |
| Yamabe et al., 2019 | Convolution neural network + long short-term Memory |
97 | Not reported | Yes (Python) | Yes |
| Miladinović et al., 2019 | Convolution neural network + hidden Markov model |
93–99 | ~9 | Yes (torch) | No |
| Miladinović et al., 2019 | Convolution neural network + hidden Markov model |
89 | ~9 | Yes (torch) | Yes |
| Barger et al., 2019 | Convolution neural network + mixture z-scoring |
97 | ~1–2 | Partially2 (Matlab—GUI) | Yes |
| Allocca et al., 2019 | Support vector machine | 0.941 | <1 | Partially2 (Matlab–GUI) | No |
| Yan et al., 2017 | Threshold decision tree | 91 | 0; Threshold | No (Matlab) | Yes |
| Gao, Turek & Vitaterna, 2016 | Multiple classifier system | ~95 | ~9 | No (Matlab) | No |
| Kreuzer et al., 2015 | Threshold decision tree | 91 | 0; Threshold | Partially2 (LabVIEW—GUI) | No |
| Bastianini et al., 2014 | Threshold decision tree | 89–97 | 0; Threshold | No (Matlab) | Yes |
| Yaghouby, O’Hara & Sunderam, 2016 | Hidden Markov model | 90 | 0; unsupervised | No (Matlab) | ~5% of epochs excluded |
| Rytkönen, Zitting & Porkka-Heiskanen, 2011 | Naïve Bayes classifier | 93 | 5 | No (Matlab) | Only days with <5% artifact |
| Gross et al., 2009 | Threshold decision tree | 80 | 0; Threshold | Partially2 (Matlab—GUI) | Yes |
| Stephenson et al., 2009 | Threshold decision tree | 89 | 0; Threshold | Yes (Spreadsheet) | No |
| Crisler et al., 2008 | Support vector machine | 96 | ~4 | No (Matlab) | No visually-scored artifacts |
Notes:
1
Alternate accuracy metric.
2
Standalone open-source tools enable algorithm implementation, but not editing.