Table 3 Comparison of all models for Fold 1 and choosing top five models by MAE
Model Architecture | Data Balancing Strategy | Segmentation Type | Precision | Recall | F1 | MAE |
|---|---|---|---|---|---|---|
CheXNet-121 | UNDER | NONE | 65.18 | 64.90 | 64.41 | 0.3953 |
CheXNet-121 | DOUBLE | NONE | 64.88 | 65.02 | 64.62 | 0.4023 |
ResNet-50 | UNDER | NONE | 63.62 | 63.46 | 62.82 | 0.4097 |
ResNet-50 | DOUBLE | SPINE | 63.85 | 64.24 | 63.91 | 0.4144 |
VGG-16 | DOUBLE | SPINE | 62.89 | 62.54 | 62.60 | 0.4151 |
VGG-16 | UNDER | NONE | 62.60 | 62.32 | 62.24 | 0.4161 |
CheXNet-121 | DOUBLE | SPINE | 63.37 | 63.62 | 63.20 | 0.4193 |
VGG-16 | DOUBLE | NONE | 63.45 | 62.76 | 62.36 | 0.4210 |
CheXNet-121 | UNDER | SPINE | 63.55 | 63.62 | 63.25 | 0.4214 |
ResNet-50 | DOUBLE | NONE | 63.02 | 63.75 | 62.77 | 0.4238 |
ResNet-50 | UNDER | SPINE | 63.79 | 62.72 | 62.60 | 0.4279 |
VGG-16 | OVER | LUNG | 60.43 | 61.37 | 61.06 | 0.4320 |
VGG-16 | OVER | NONE | 60.81 | 60.98 | 60.11 | 0.4371 |
CheXNet-121 | OVER | NONE | 61.56 | 61.76 | 60.85 | 0.4392 |
ResNet-50 | DOUBLE | LUNG | 62.54 | 62.22 | 62.14 | 0.4412 |
VGG-16 | UNDER | SPINE | 62.32 | 62.43 | 61.56 | 0.4439 |
ResNet-50 | UNDER | LUNG | 61.31 | 61.38 | 60.93 | 0.4454 |
CheXNet-121 | OVER | SPINE | 61.13 | 61.24 | 60.80 | 0.4489 |
ResNet-50 | OVER | LUNG | 59.63 | 60.58 | 59.64 | 0.4600 |
CheXNet-121 | OVER | LUNG | 58.94 | 59.23 | 58.39 | 0.4613 |
VGG-16 | UNDER | LUNG | 60.95 | 60.17 | 60.02 | 0.4612 |
VGG-16 | DOUBLE | LUNG | 61.01 | 61.07 | 60.73 | 0.4635 |
ResNet-50 | OVER | SPINE | 60.39 | 61.20 | 60.37 | 0.4665 |
CheXNet-121 | UNDER | LUNG | 60.55 | 60.72 | 60.40 | 0.4669 |
ResNet-50 | OVER | NONE | 59.54 | 57.31 | 56.07 | 0.4864 |
CheXNet-121 | DOUBLE | LUNG | 59.58 | 60.60 | 59.68 | 0.4973 |
VGG-16 | OVER | SPINE | 35.79 | 46.46 | 38.04 | 0.9625 |