Table 4.
Mean and standard deviation values of the 5-fold cross validation in terms of F1 score, precision, recall and accuracy. The first three rows reported the implementation results of other competitive methods using each of our dataset. Type of machine learning model was recorded if the model of the row was ensembled. The number of trainable parameters is displayed in thousands in the rightmost column. The models that shows significant difference with the TN stage (baseline) for all evaluation metrics are marked with an asterisk (*).
| F1 Score | Precision | Recall | Accuracy | Model | Trainable Parameters |
|
|---|---|---|---|---|---|---|
| Clinical (David Cox) [38] | 68.11 (±2.4) | 56.51 (±1.4) | 85.84 (±5.5) | 52.90 (±2.5) | Cox PH | - |
| HCR (Wen Yu et al.) [40] | 72.19 (±3.4) | 63.78 (±2.1) | 83.23 (±5.5) | 62.44 (±3.8) | RSF | - |
| DLR (André Diamant et al.) [41] | 74.55 (±5.4) | 69.12 (±6.3) | 81.12 (±5.2) | 67.35 (±7.1) | CNN | 916 K |
| TN stage (baseline) | 65.67 (±5.1) | 70.09 (±8.0) | 63.81 (±10.6) | 61.54 (±3.8) | NN | 1 K |
| Clinical | 73.54 (±2.4) | 68.99 (±5.1) | 79.07 (±2.2) | 66.46 (±3.8) | NN1 | 1 K |
| HCR | 76.61 (±4.7) | 73.07 (±5.2) | 80.61 (±4.7) | 71.08 (±5.7) | NN2 | 10 K |
| DLR | 76.28 (±4.9) | 70.11 (±3.1) | 84.80 (±11.2) | 69.54 (±4.2) | CNN | 1046 K |
| Clinical & HCR * | 77.58 (±5.0) | 75.29 (±4.1) | 80.08 (±6.4) | 72.92 (±5.6) | QDA | 11 K |
| Clinical & DLR * | 76.86 (±4.7) | 71.03 (±3.7) | 83.75 (±6.1) | 70.46 (±5.6) | LR | 1047 K |
| HCR & DLR * | 77.65 (±5.0) | 73.70 (±4.5) | 82.19 (±6.6) | 72.31 (±5.9) | LR | 1056 K |
| Clinical & HCR & DLR * | 77.79 (±5.3) | 75.71 (±4.8) | 80.08 (±6.4) | 73.23 (±6.0) | LDA | 1057 K |