Table 4 Stratified 5-fold cross-validation results for NGT positioning classification on the testing and external datasets. Results are reported as mean ± standard deviation (SD).
From: Deep learning based dual stage model for accurate nasogastric tube positioning in chest radiographs
Methods | Macro F1-score | B-ACC | AUC-ROC | AUC-PR |
|---|---|---|---|---|
ResNet-18 (\(x + m^l\)) | 94.36±7.20 | 96.27±4.64 | 97.98±3.85 | 93.61±12.09 |
ResNet-34 (\(x + m^l\)) | 93.95±7.51 | 95.44±5.09 | 97.78±3.62 | 93.38±11.82 |
ResNet-50 (\(x + m^l\)) | 95.72±3.98 | 97.04±1.61 | 99.57±0.35 | 99.04±0.85 |
ResNet-50 (pre-trained w/ MedCLIP) (\(x + m^l\)) | 97.83±1.09 | 98.83±0.66 | 99.96±0.03 | 99.89±0.11 |
ResNet-50 (pre-trained w/ MedCLIP) (\(x + m^l\)) (external) | 78.99±2.37 | 83.94±3.23 | 92.27±2.59 | 86.78±3.62 |
