Classification accuracies for each combination of input variables
| DMLP development using 477 patients divided into training (80%), validation (10%) and test (10%) sets | |||
| Combination of input variables | Training set (%) | Validation set (%) | Test set (%) |
| Biometrics with spirometry and plethysmography | 97.38±1.03 | 95.53±2.55 | 94.79±2.56 |
| Biometrics with spirometry | 89.58±1.14 | 87.87±2.47 | 89.81±2.06 |
| Spirometry and plethysmography | 62.50±7.58 | 57.41±7.43 | 59.87±9.17 |
| Spirometry | 61.28±4.19 | 57.42±8.76 | 59.39±7.30 |
| DMLP refinement and application to unseen data: first 477 patients used for training, later 271 patients for validation (136) and test (135) sets | |||
| Combination of input variables | Training set (%) | Validation set (%) | Test set (%) |
| Biometrics with spirometry and plethysmography | 98.24±0.85 | 95.42±1.88 | 95.04±1.71 |
| Biometrics with spirometry | 98.39±1.07 | 94.83±2.7 | 95.19±2.35 |
| Spirometry and plethysmography | 67.51±7.29 | 65.00±5.71 | 63.33±6.64 |
| Spirometry | 58.01±7.41 | 56.69±8.24 | 56.37±8.91 |
Accuracies (mean±SD) for each combination of input variables after 10 experimental runs are shown. For each experimental run, the model randomly selected one test per patient. Biometric variables are sex, age, height. Spirometry included FVC, FEV1, FEF25-75 and plethysmography metrics are RV and TLC.
DMLP, deep multilayer perceptron; FEF, forced mid-expiratory flow; FEV1, forced expiratory volume in 1 s; FVC, forced vital capacity; RV, residual volume; TLC, total lung capacity.