Table 1.
Multi-disease clustering performances of ConvAE configurations and baselines.
| Entropya | Puritya | Disease numberb | |
|---|---|---|---|
| ConvAE 1-layer CNN | 2.61 (0.04, [2.58, 2.67])*** | 0.31 (0.02, [0.31, 0.35])*** | 6.50 (0.62)*** |
| ConvAE 2-layer CNN | 2.75 (0.02, [2.74, 2.78]) | 0.26 (0.01, [0.26, 0.29]) | 5.93 (0.50) |
| ConvAE multikernel CNN | 2.66 (0.03, [2.64, 2.70]) | 0.30 (0.02, [0.29, 0.33]) | 5.94 (0.47) |
| RawCount | 2.90 (0.02, [2.88, 2.92]) | 0.18 (0.01, [0.18, 0.20]) | 4.76 (0.70) |
| SVD-RawCount | 2.90 (0.01, [2.90, 2.92]) | 0.19 (0.01, [0.18, 0.20]) | 5.13 (0.79) |
| SVD-TFIDF | 2.85 (0.02, [2.84, 2.87]) | 0.21 (0.01, [0.21, 0.23]) | 5.83 (0.76) |
| Deep Patient | 2.81 (0.02, [2.80, 2.84]) | 0.24 (0.01, [0.23, 0.25]) | 5.96 (0.74) |
The scores reported are averaged over a 2-fold cross-validation experiment. ConvAE 1-layer CNN significantly outperforms all other configurations and baselines on all measures. Multiple pairwise t tests with Bonferroni correction are used to compare performances.
CNN convolutional neural network, SVD singular value decomposition, TFIDF term frequency-inverse document frequency.
***p < 0.001.
aMean (s.d., CI).
bMean (standard deviation).