Table 2.

The performances of different models for the prediction of ER of HCC.

	Development cohort					Validation cohort
	AUC (95% CI)	ACC	SEN	SPEC	$P$ value${}^{\ast }$	AUC (95% CI)	ACC	SEN	SPEC	$P$ value${}^{\ast }$
DSFR	0.740 (0.652, 0.816)	0.733	0.750	0.708	Ref	0.717 (0.516, 0.869)	0.750	0.778	0.700	Ref
DSFR-C	0.782 (0.698, 0.853)	0.725	0.667	0.812	0.042	0.744 (0.545, 0.889)	0.750	0.722	0.800	0.028
Model with visual features	0.657 (0.565, 0.742)	0.617	0.486	0.813	0.149	0.583 (0.383, 0.765)	0.572	0.389	0.900	0.287

${}^{\ast }$DeLong’s test. ER: early recurrence; HCC: hepatocellular carcinoma; AUC: area under the curve; ACC: accuracy; SEN: sensitivity; SPEC: specificity; DSFR: deep semantic segmentation feature-based radiomics; Ref: reference; DSFR-C: deep semantic segmentation feature-based radiomics with clinical information.