Table 2.
Simulated online peak accuracies for sessions and pathologies.
| Peak accuracies for different Adaptive BCI configurations | |||||
|---|---|---|---|---|---|
| Best class-combination | Auto-AdBCI | SMR-AdBCI | Worst class-combination | ||
| Session 1† | Stroke | 73.2 | 71.2 | 65.2 | 57.6 |
| SCI | 80.8 | 74.5 | 62.0 | 60.2 | |
| Mean (SD) | 77.6 (6.1) | 73.1 (8.5) | 63.4 (5.1) | 59.1 (2.5) | |
| Session 2 | Stroke | 73.9 | 69.9 | 60.5 | 59.1 |
| SCI | 85.3 | 80.3 | 70.8 | 60.9 | |
| Mean (SD) | 80.2 (7.7) | 75.7 (8.4) | 66.3 (7.2) | 60.1 (2.8) | |
| Mean (SD) | 78.4 (6.1) | 73.6 (7.7) | 64.5 (3.5) | 59.5 (2.1) | |
The Auto-AdBCI, initially auto-selected one of six class combinations according to a heuristic. Based on seven trials per class, the heuristic tried to select a class-combination that would allow for a highest possible peak control accuracy over the session. To compare, we simulated the overall session accuracy not only with the auto-selected class-combination (Auto-AdBCI), but also with all other class-combinations. The column “Best Class-Combination” is the average when considering for every user only the one class-combination that eventually produces the highest overall accuracy. In analogy, the column “Worst Class-Combination” considers for every user only the one class-combination that eventually produces the lowest overall accuracy.
Notice, the data of Session 1 is “seen data” as it has been previously used to determine the configurations of the BCIs.