Table 9.
Mutual information of auto-encoder features. The same trend of Table 8 holds. While the SBP task has low Info-Fraction, the features encode much more information for the HR and RWAT tasks. Ground truth for RWAT is only available for the Aurora-BP dataset60.
Task | MIMIC | Aurora-BP | ||||
---|---|---|---|---|---|---|
MI | Entropy | Info-Fraction | MI | Entropy | Info-Fraction | |
SBP | 0.38 | 2.93 | 12.9% | 0.32 | 3.68 | 8.70% |
HR | 2.60 | 2.82 | 92.2% | 3.40 | 3.65 | 93.1% |
RWAT | – | – | – | 1.78 | 2.54 | 70.1% |