Table 6.

Top images features discriminating cancer and benign nodules across different size ranges and features with in each image categories.

Feature categories and nodule size ranges.
Range: 4 to 12 mm (129 Reproducible, Non-Redundant Features)
Features	Training		Testing
	Image	Image + Clinical	Image	Image + Clinical
	E [AUC], CI	E [AUC], CI	E [AUC], CI	E[AUC],CI
i. Size and Shape Features (C1)
Radiomics: Size & Shape (F6: Vol-cm; F38: Length-Pxl; F40: Length-by-Width; F41: Border-Leng-Pxl)	0.62, [0.415, 0.811]	0.61, [0.39, 0.82]	0.71 [0.68, 0.72]	0.66, [0.61, 0.69]
ii. Location, Co-occurrence, Run Length & Pixel HU. Histogram (C2)
Radiomics: Location, Cooccurrence, Runlength (F9: 8b-3D-Bord-to-Lung; F17: 9f-3D-Min-Dist-COG-to-Border; F19: 10a-3D-Relat-Vol-Airspaces; F55: AvgRP)	0.83, [0.63, 0.97]	0.81, [0.61, 0.97]	0.764 [0.74, 0.78]	0.665 [0.02, 0.67]
iii. Texture: Laws & Wavelets (C3)
Radiomics: Texture – Laws & Wavelet (F68: 3D-Laws-10; F201: 3D-WaveP2-L2-12; F214: 3D-WaveP1-L2-25; F216: 3D-WaveP1-L2-27)	0.61, [0.27, 0.89]	0.64, [0.40, 0.86]	0.649 [0.62, 0.66]	0.58, [0.53, 0.625]
Range: >12 to 30 mm (129 Reproducible, Non-Redundant Features)
i. Size & Shape Features (C1)
Radiomics: Size & Shape (F37: Thickness-Pxl; F41:Border-Leng-Pxl; F25: Density; F26: EllipticFit)	0.71, [0.19, 0.96]	0.63, [0.25, 0.96]	0.79 [0.72, 0.83]	0.784, [0.70, 0.86]
ii. Location, Co-occurrence, Run Length & Pixel HU. Histogram (C2)
Radiomics: Location, Cooccurrence, Runlength (F187: Hist-Entropy-L1; F188:Hist-Kurt-L1; F42: AvgCoOc-Homo; F48: AvgGLN)	0.67, [0.21, 0.96]	0.57, [0.19, 0.88]	0.68, [0.62, 0.715]	0.659, [0.55, 0.76]
iii. Texture: Laws & Wavelets (C3)
Radiomics: Texture-Laws & Wavelet (F196: 3D-WaveP2-L2-7; F199: 3D-WaveP2-L2-10; F214: 3D-WaveP1-L2-25; F216: 3D-WaveP1-L2-27)	0.76, [0.42, 0.97]	0.68 [0.28, 0.96]	0.591 [0.55, 0.61]	0.634 [0.57, 0.68]

Statistical estimates were computed using cross validation (hold out), randomized multiple times.