Table 1 Quantitative evaluation metrics of the AI framework on the unseen dataset.
From: A transparent artificial intelligence framework to assess lung disease in pulmonary hypertension
Metric | ResNet-50 | DenseNet121 | DenRes-131 |
|---|---|---|---|
The AI framework for the \(64\times 64\times 3\) patch size | |||
Jaccard score (%) | 74.63 ± 17.95 | 70.12 ± 18.32 | 69.21 ± 17.89 |
Hamming distances (mm) | 16.21 ± 6.21 | 18.01 ± 8.01 | 18.07 ± 9.03 |
Root mean square error | 1.246 ± 0.510 | 1.282 ± 0.634 | 1.281 ± 0.700 |
f1 score (%) | 77.33 ± 10.02 | 73.21 ± 13.54 | 73.01 ± 13.77 |
Recall score (%) | 78.21 ± 9.87 | 74.01 ± 14.01 | 73.99 ± 14.04 |
Precision score (%) | 78.33 ± 9.02 | 74.32 ± 14.21 | 74.35 ± 14.98 |
Accuracy (%) | 78.23 ± 10.00 | 74.00 ± 15.01 | 73.12 ± 16.00 |
MCC (%) | 67.23 ± 17.21 | 64.12 ± 18.12 | 63.89 ± 19.00 |
The AI framework for the \(32\times 32\times 3\) patch size | |||
Jaccard score (%) | 69.41 ± 21.94 | 90.53 ± 4.38 | *91.83 ± *3.48 |
Hamming distances (mm) | 17.29 ± 7.29 | 6.34 ± 3.82 | *5.96 ± *3.17 |
Root mean square error | 1.171 ± 0.419 | 0.904 ± 0.54 | *0.855 ± 0.40 |
f1 score (%) | 71.23 ± 10.11 | 92.06 ± 5.40 | *93.87 ± 4.20 |
Recall score (%) | 70.12 ± 10.33 | 93.21 ± 4.11 | *93.42 ± *2.84 |
Precision score (%) | 71.22 ± 10.43 | 94.53 ± 2.52 | *96.54 ± 2.82 |
Accuracy (%) | 70.15 ± 10.32 | 93.02 ± 3.93 | *93.69 ± 3.90 |
MCC (%) | 65.34 ± 20.32 | 77.74 ± 8.46 | *80.21 ± 7.83 |
The AI framework for the \(16\times 16\times 3\) patch size | |||
Jaccard score (%) | 70.05 ± 20.67 | 87.70 ± 8.90 | 89.01 ± 5.81 |
Hamming distances (mm) | 17.13 ± 7.11 | 8.08 ± 5.26 | 7.99 ± 4.48 |
Root mean square error | 1.146 ± 0.409 | 1.035 ± 0.42 | 1.015 ± *0.32 |
f1 score (%) | 72.01 ± 10.00 | 92.06 ± 5.10 | 92.27 ± *4.87 |
Recall score (%) | 71.00 ± 10.01 | 90.67 ± 4.53 | 91.64 ± 4.33 |
Precision score (%) | 71.87 ± 10.67 | 95.71 ± 2.08 | 95.65 ± *2.20 |
Accuracy (%) | 70.78 ± 10.78 | 90.88 ± 4.74 | 91.51 ± *3.27 |
MCC (%) | 66.01 ± 20.01 | 73.10 ± 11.71 | 74.80 ± *7.21 |
The AI framework for the \(8\times 8\times 3\) patch size | |||
Jaccard score (%) | 68.56 ± 21.09 | 69.20 ± 18.52 | 69.10 ± 18.20 |
Hamming distances (mm) | 16.61 ± 6.23 | 18.27 ± 9.91 | 18.31 ± 9.83 |
Root mean square error | 1.587 ± 0.355 | 1.361 ± 0.839 | 1.432 ± 0.840 |
f1 score (%) | 74.13 ± 13.12 | 72.51 ± 15.54 | 72.41 ± 15.74 |
Recall score (%) | 72.41 ± 10.82 | 73.41 ± 16.01 | 73.49 ± 16.04 |
Precision score (%) | 72.63 ± 12.21 | 74.32 ± 17.22 | 73.35 ± 16.98 |
Accuracy (%) | 73.53 ± 11.90 | 73.60 ± 17.11 | 73.10 ± 16.89 |
MCC (%) | 61.43 ± 18.20 | 64.02 ± 19.18 | 63.59 ± 19.20 |
