Table 24 Cross-domain performance with different synthetic data ratios.

From: Novel metaheuristic optimized latent diffusion framework for automated oral disease detection in public health screening

Dataset

0% synthetic (95% CI)

25% synthetic (95% CI)

50% synthetic (95% CI)

75% synthetic (95% CI)

100% synthetic (95% CI)

Performance gain vs. real-only

Effect size (η²)

p-value

ExtVal-1 (rural)

73.8% (72.1–75.5)

82.4% (80.9–83.9)

87.9% (86.6–89.2)

89.4% (88.2–90.6)

86.2% (84.8–87.6)

+ 15.6%

0.743

< 0.001

ExtVal-2 (urban)

76.2% (74.8–77.6)

84.7% (83.5–85.9)

89.3% (88.3–90.3)

91.7% (90.8–92.6)

88.1% (87.0-89.2)

+ 15.5%

0.718

< 0.001

ExtVal-3 (mobile)

69.1% (67.2–71.0)

78.3% (76.6–80.0)

83.6% (82.1–85.1)

86.2% (84.8–87.6)

82.7% (81.1–84.3)

+ 17.1%

0.689

< 0.001

ExtVal-4 (international)

71.9% (70.3–73.5)

81.2% (79.8–82.6)

86.4% (85.2–87.6)

88.9% (87.8–90.0)

85.3% (84.0-86.6)

+ 17.0%

0.701

< 0.001

ExtVal-5 (early pathology)

67.4% (65.4–69.4)

76.8% (75.0-78.6)

81.9% (80.3–83.5)

84.6% (83.1–86.1)

80.1% (78.4–81.8)

+ 17.2%

0.672

< 0.001

ExtVal-6 (emergency)

64.7% (62.6–66.8)

74.2% (72.3–76.1)

79.8% (78.1–81.5)

82.3% (80.7–83.9)

78.4% (76.6–80.2)

+ 17.6%

0.658

< 0.001

ExtVal-7 (longitudinal)

75.9% (74.3–77.5)

83.6% (82.2–85.0)

88.7% (87.5–89.9)

90.8% (89.7–91.9)

87.2% (85.9–88.5)

+ 14.9%

0.724

< 0.001

Average external

71.3% (70.7–71.9)

80.2% (79.6–80.8)

85.4% (84.9–85.9)

87.7% (87.2–88.2)

83.9% (83.3–84.5)

+ 16.4%

0.704

< 0.001