Table 4.

Performance of models using features other than only lesion features

Metric	NLP	NLP + Lesion	Ring	Clinical	Clinical + Lesion
AUC	0.65 [0.51, 0.78]	0.63 [0.52, 0.75]	0.67 [0.56, 0.78]	0.56 [0.43, 0.70]	0.65 [0.53, 0.77]
Accuracy	0.59 [0.49, 0.70]	0.60 [0.50, 0.71]	0.63 [0.54, 0.73]	0.53 [0.41, 0.64]	0.62 [0.51, 0.72]
Sensitivity	0.52 [0.33, 0.70]	0.60 [0.43, 0.76]	0.67 [0.51, 0.83]	0.56 [0.37, 0.75]	0.62 [0.45, 0.79]
Specificity	0.67 [0.50, 0.85]	0.61 [0.46, 0.75]	0.59 [0.45, 0.74]	0.49 [0.31, 0.67]	0.61 [0.45, 0.77]

These features were extracted from a segmentation of the normal liver parenchyma (NLP); the NLP and the lesion (NLP + Lesion); a ring at the border of the segmentation (Ring); using the clinical characteristics (Clinical); and the clinical characteristics combined with lesion features (Clinical + Lesion)

*Abbreviations: AUC area under the receiver operating characteristic curve; NLP normal liver parenchyma; CNN convolutional neural network