Skip to main content
. 2024 Dec 28;17(1):58. doi: 10.3390/cancers17010058

Table 2.

Performance comparison between feature selection algorithms and the features selected. Baseline characteristics includes age, sex, smoking status, pack years, and the number of metastatic sites (full list can be found in Supplementary Materials).

Feature Selection
Algorithm		 Including Baseline
Characteristics		 ROC AUC
(Discovery Training Set)		 Features Selected
(Up to 5)
1. mRMR * Yes 0.77 ± 0.05

  1. No. of metastatic sites

  2. GrayLevel6_RING

  3. FFT2_RING

  4. ECOG Score
2. mRMR * No 0.85 ± 0.10

  1. MinAxis_RING

  2. RadVarAngle_RING
1. ReliefF Yes 0.67 ± 0.09

  1. Liver Lesion

  2. V-Corre1ation_CORE
2. ReliefF No 0.70 ± 0.08

  1. RadSD-Angle_CORE

  2. RadKurtosis-Angle_CORE
3. SFS † Yes 0.85 ± 0.02

  1. Pack Years

  2. No. of metastatic sites

  3. OD-CentroidDifference_RING

  4. RadCentre_RING
4. SFS † No 0.87 ± 0.09

  1. TotalVariance_CORE

  2. MeanRadius_RING

  3. ODEccentricity_RING

  4. RadVarianceAngle_RING

  5. RadCentre_RING

* mRMR = minimum Redundancy Maximum Relevance, † SFS = Sequential Forward Feature Selection.