TABLE 4.
Overview of feature selection methods used in CT lung cancer radiomic studies
| Classification | Prognostics | Total | ||
|---|---|---|---|---|
| Single method | LASSO | 8 | 3 | 11 |
| Other a | 3 | 1 | 4 | |
| Random Forest | 3 | 1 | 4 | |
| Logistic Regression | 2 | 2 | 4 | |
| Spearman/Pearson | 2 | 3 | 4 | |
| Component Analysis | 2 | 1 | 3 | |
| Serial methods | ICC/CCC | 28 | 16 | 44 |
| LASSO | 28 | 13 | 41 | |
| Pearson/Spearman | 13 | 7 | 20 | |
| mRMR | 12 | 3 | 15 | |
| (Non)Parametric Stats | 12 | 0 | 12 | |
| Clustering | 7 | 0 | 7 | |
| ICC/CCC + Pearson/Spearman | 4 | 3 | 7 | |
| Parallel methods | Other a | 8 | 13 | 21 |
| RELIEF | 3 | 4 | 7 | |
| (Non‐)Parametric Stats | 2 | 3 | 5 | |
| Pearson/Spearman | 0 | 4 | 4 | |
| mRMR | 1 | 3 | 4 | |
| Fisher Score | 1 | 2 | 3 | |
| Wilcox Rank Sum | 1 | 2 | 3 | |
The breakdown of the most common feature selection methods in “Single Method,” “Serial Method,” and “Parallel Method” studies is displayed. In many cases, the feature selection methods chosen in “Parallel Method” studies were not widely used, resulting in the high frequency of “Other” feature selection methods in these studies.
a
“Other” indicates feature selection methods that were used twice or less.