Table 4 Performance comparison when using each of the 3 AEs — Basic AE, Denoising AE and Sparse AE — and for each type of cancer
From: Using autoencoders as a weight initialization method on deep neural networks for disease detection
Top Layers (AEs) | Accuracy (%) | MCC | Precision (%) | Recall (%) | F1 score | |
|---|---|---|---|---|---|---|
Fixing the AE weights (Approach A) | ||||||
Thyroid | AE: Encoding Layers | 91.80 ±3.34 | 0.72 ±0.11 | 76.06 ±11.72 | 78.98 ±12.01 | 76.64 ±8.98 |
AE: Complete Autoencoder | 94.90 ±1.56 | 0.82 ±0.06 | 84.13 ±5.00 | 86.43 ±6.63 | 85.32 ±4.57 | |
DAE: Encoding Layers | 87.63 ±3.49 | 0.57 ±0.09 | 67.41 ±12.33 | 63.48 ±18.10 | 63.49 ±9.42 | |
DAE: Complete Autoencoder | 94.03 ±1.06 | 0.79 ±0.04 | 81.79 ±4.60 | 84.37 ±4.87 | 83.11 ±2.73 | |
SAE: Encoding Layers | 87.93 ±2.66 | 0.57 ±0.14 | 65.43 ±7.70 | 64.25 ±22.35 | 62.41 ±14.32 | |
SAE: Complete Autoencoder | 92.67 ±1.96 | 0.75 ±0.06 | 77.13 ±8.76 | 82.51 ±4.57 | 79.42 ±4.72 | |
Skin | AE: Encoding Layers | 90.77 ±3.52 | 0.63 ±16.17 | 74.71 ±13.54 | 61.40 ±17.15 | 66.83 ±15.08 |
AE: Complete Autoencoder | 92.40 ±2.65 | 0.69 ±0.12 | 80.92 ±9.21 | 67.35 ±13.36 | 73.12 ±11.00 | |
DAE: Encoding Layers | 85.07 ±4.16 | 0.42 ±0.16 | 56.22 ±16.42 | 47.02 ±18.01 | 48.76 ±14.72 | |
DAE: Complete Autoencoder | 89.43 ±1.32 | 0.58 ±0.06 | 69.42 ±5.65 | 60.15 ±8.18 | 63.98 ±5.04 | |
SAE: Encoding Layers | 79.27 ±3.70 | 0.07 ±5.71 | 33.49 ±25.49 | 13.79 ±7.76 | 16.11 ±8.02 | |
SAE: Complete Autoencoder | 85.83 ±2.04 | 0.45 ±0.07 | 56.34 ±7.81 | 49.57 ±6.24 | 52.42 ±5.64 | |
Stomach | AE: Encoding Layers | 91.60 ±1.91 | 0.62 ±0.09 | 76.09 ±9.45 | 58.07 ±10.87 | 65.35 ±8.96 |
AE: Complete Autoencoder | 94.00 ±1.47 | 0.74 ±0.07 | 83.45 ±8.34 | 71.56 ±6.24 | 76.78 ±5.42 | |
DAE: Encoding Layers | 86.50 ±4.33 | 0.33 ±0.16 | 64.35 ±22.75 | 27.46 ±12.98 | 34.93 ±15.68 | |
DAE: Complete Autoencoder | 91.03 ±1.57 | 0.58 ±0.08 | 74.03 ±7.85 | 54.91 ±9.42 | 62.61 ±7.69 | |
SAE: Encoding Layers | 85.93 ±1.34 | 0.16 ±0.11 | 48.24 ±25.25 | 11.06 ±6.73 | 17.49 ±9.35 | |
SAE: Complete Autoencoder | 89.87 ±2.06 | 0.56 ±8.23 | 66.10 ±9.38 | 57.09 ±6.33 | 61.02 ±6.82 | |
Breast | AE: Encoding Layers | 88.40 ±5.52 | 0.59 ±0.17 | 68.39 ±19.13 | 64.80 ±10.84 | 65.91 ±13.72 |
AE: Complete Autoencoder | 91.77 ±3.13 | 0.69 ±0.12 | 80.57 ±11.79 | 67.00 ±11.24 | 72.91 ±10.86 | |
DAE: Encoding Layers | 83.53 ±1.74 | 0.25 ±0.14 | 51.39 ±25.04 | 25.60 ±15.57 | 31.23 ±17.51 | |
DAE: Complete Autoencoder | 87.30 ±1.90 | 0.53 ±0.05 | 63.43 ±7.13 | 58.60 ±5.17 | 60.67 ±4.58 | |
SAE: Encoding Layers | 79.73 ±3.86 | 0.02 ±0.05 | 9.80 ±12.48 | 3.00 ±3.16 | 4.11 ±4.09 | |
SAE: Complete Autoencoder | 84.07 ±2.40 | 0.41 ±0.07 | 53.13 ±8.05 | 47.80 ±4.85 | 50.13 ±5.62 |