Table 3.
A list of studies that have used LIDC-IDRI to predict lung nodule malignancy based on the ratings provided by radiologists. Note that some of the studies (refs.5,32) included hand-crafted features, requiring expert annotations.
| Method | Area Under the Curve (AUC) | Accuracy | Specificity | Sensitivity |
|---|---|---|---|---|
| Proposed model | 0.964 | 93.12% | 90% | 94.94% |
| CNN5 | 0.938 | 87.9% | 87.9% | 87.9% |
| CNN in combination with hand-crafted features5 | 0.971 | 93.2% | 98.5% | 87.9% |
| Deep residual network30 | 0.9459 | 89.90% | 88.64% | 91.07% |
| Deep belief network31 | — | 81.19% | — | — |
| CNN in combination with hand-crafted features32 | — | 86.79% | 95.42% | 60.26% |
| Multi-crop CNN33 | 0.93 | 87.14% | 93% | 77% |