Table 2. Quantitative comparison of kidney tumor segmentation performance across different tumor size groups.

Size	Methods	DSC (%)	Precision (%)	Recall (%)	Bal. Acc.(%)
Small(tumor <2.4cm)	nnU-Net	61.71±27.18	70.58±32.34	61.04±29.73	80.52±14.86
	CLIP-Driven-Universal Model (baseline)	56.92±31.60	64.86±34.55	53.97±33.16	76.98±16.58
	Support image-mask pair condition + Size-aware	65.70±25.84	69.25±28.65	65.12±28.19	82.56±14.09
Medium(2.4cm ≤ tumor <4.3cm)	nnU-Net	78.01±18.37	89.60±15.35	75.04±22.48	87.52±11.24
	CLIP-Driven-Universal Model (baseline)	66.51±26.07	70.58±28.50	67.44±27.28	83.71±13.64
	Support image-mask pair condition + Size-aware	72.34±19.01	74.95±22.83	74.51±21.42	87.25±10.71
Large(4.3cm ≤ tumor <7.4cm)	nnU-Net	78.24±24.22	77.03±24.22	86.03±11.42	92.98±5.69
	CLIP-Driven-Universal Model (baseline)	67.63±24.28	84.68±17.01	65.04±27.29	82.51±13.64
	Support image-mask pair condition + Size-aware	70.17±26.33	75.25±30.75	68.89±27.42	84.44±13.70
Extremely large(7.4cm ≤ tumor)	nnU-Net	80.82±24.07	85.87±25.42	85.36±19.88	92.60±9.91
	CLIP-Driven-Universal Model (baseline)	79.45±20.52	82.46±23.56	83.17±16.05	91.51±8.04
	Support image-mask pair condition + Size-aware	81.81±19.91	82.46±24.49	84.81±14.42	92.33±7.21
Overall	nnU-Net	75.20±21.96	82.09±24.21	76.60±23.26	88.27±11.61
	CLIP-Driven-Universal Model (baseline)	67.46±26.18	74.89±27.26	67.41±27.65	83.68±13.82
	Support image-mask pair condition + Size-aware	72.28±22.42	75.68±25.77	73.51±23.47	86.73±11.73