Table 3.
Top images features discriminating cancer and benign nodules using image features across categories.
| All Categories (129 Reproducible, Non-Redundant Features) | |||||
|---|---|---|---|---|---|
| Features | Training | Testing | |||
| Broad description | Detailed feature | Image | Image + Clinical | Image | Image + Clinical |
| E[AUC], CI | E[AUC], CI | E[AUC], CI | AUC | ||
| Size, Shape and Texture | (F3: ShortAx; F4: Mn-Hu; F26: EllipticFit; F176: 3D-Laws-127) | 0.83 [0.71, 0.94] | 0.82, [0.68, 0.94] | 0.90 [0.89, 0.91] | 0.89 [0.88, 0.9] |
| Location, Texture | (F19: 10a-3D-Relat-Vol-Airspaces;F21: 10c-3D-Av-Vol-AirSpaces;F22: 10d-3D-SD-Vol-AirSpaces; F201:3D-WaveP2-L2-12) | 0.83 [0.70, 0.93] | 0.81, [0.67, 0.92] | 0.87 [0.86, 0.88] | 0.86 [0.85, 0.87] |
| Location, Size, Texture-Runlength | (F19: 10a-3D-Relat-Vol-Airspaces;F34: Volume-pxl; F37: Thickness-Pxl; F48: AvgGLN) | 0.80, [0.65, 0.92] | 0.80, [0.68, 0.89] | 0.89 [0.89, 0.90] | 0.88 [0.869, 0.893 |
| Baseline | Longest Diameter | 0.76, [0.62, 0.93] | 0.85, [0.35, 0.95] | ||
| Volume | 0.78 [0.63, 0.91] | 0.87 [0.37, 0.96] | |||
Statistical estimates were computed using cross validation (hold out) for training and tested, randomized multiple times.