Table 3. Risk Scores for 3 approaches from baseline NLST for predicting subsequent cancer.
Lung-RADS obtained a total accuracy of 71.4%. The McWilliams approach obtained an accuracy of 78.9%. The Radiomic approach obtained an accuracy of 80.0%. The Volume Only approach obtained an accuracy of 71.8%.
| Model | Category | # Malignant | # Benign | Total | Accuracy | TPR | TNR |
|---|---|---|---|---|---|---|---|
| Overall | Total | 58 | 127 | 185 | |||
| Lung- RADS (71.4%) | 2 | 32 | 99 | 131 | 75.6% | ||
| 3 | 13 | 20 | 33 | 60.6% | |||
| 4A | 10 | 7 | 17 | 58.8% | |||
| 4B | 3 | 1 | 4 | 75.0% | |||
| Total | 58 | 127 | 185 | 71.4% | 22.4 % | 93.7% | |
| McWilliams (78.9%) | Low | 24 | 98 | 122 | 79.7% | ||
| Intermediate | 7 | 21 | 28 | 75.0% | |||
| High | 27 | 8 | 35 | 77.1% | |||
| Total | 58 | 127 | 185 | 78.9% | 46.5 % | 93.7% | |
| Radiomics (80.0%) | Low | 6 | 80 | 86 | 93.0% | ||
| Intermediate-Low | 22 | 38 | 60 | 63.3% | |||
| Intermediate-High | 17 | 8 | 25 | 68.0% | |||
| High | 13 | 1 | 14 | 92.9% | |||
| Total | 58 | 127 | 185 | 80.0% | 51.7% | 92.9% | |
| Volume (71.8%) | Low | 17 | 85 | 102 | 83.3% | ||
| Intermediate-Low | 13 | 20 | 33 | 60.6% | |||
| Intermediate-High | 5 | 15 | 20 | 25.0% | |||
| High | 23 | 7 | 30 | 76.6% | |||
| Total | 58 | 127 | 185 | 71.8% | 48.2% | 82.7% |
Lung RADS categories include: Benign Appearance and Behavior (2), Probably Benign (3), and Suspicious (4A, 4B). McWilliams categories include: Low, Intermediate, and High Radiomics categories include: Low, Intermediate-Low, Intermediate-High, and High Volume Only categories include: Low, Intermediate-Low, Intermediate-High, and High TPR – True Positive Rate (Sensitivity), TNR – True Negative Rate (Specificity) Top accuracies per method are in bold