Table 4 Diagnostic performance comparison of mucosal thickening and polypoid lesions detection between optimized techniques and baseline in the internal test set (n = 100, Mann–Whitney U test)
From: Expert-guided StyleGAN2 image generation elevates AI diagnostic accuracy for maxillary sinus lesions
Techniques | Mucosal thickening | Polypoid lesion | ||||
|---|---|---|---|---|---|---|
AUPRC | AUROC | F1 score | AUPRC | AUROC | F1 score | |
Baseline | 0.83 ± 0.05 | 0.84 ± 0.04 | 0.81 ± 0.03 | 0.69 ± 0.08 | 0.66 ± 0.08 | 0.69 ± 0.02 |
StyleGAN2-AUG | 0.91 ± 0.03 (P < 0.001) | 0.90 ± 0.03 (P = 0.001) | 0.85 ± 0.03 (P = 0.011) | 0.83 ± 0.02 (P < 0.001) | 0.80 ± 0.02 (P < 0.001) | 0.74 ± 0.02 (P < 0.001) |
ReACGAN-AUG | 0.88 ± 0.06 (P = 0.043) | 0.88 ± 0.06 (P = 0.063) | 0.84 ± 0.05 (P = 0.123) | 0.68 ± 0.14 (P = 0.912) | 0.65 ± 0.12 (P = 0.739) | 0.70 ± 0.03 (P = 0.796) |
Flipping | 0.73 ± 0.10 (P = 0.002) | 0.71 ± 0.11 (P < 0.001) | 0.72 ± 0.04 (P < 0.001) | 0.51 ± 0.06 (P < 0.001) | 0.47 ± 0.10 (P < 0.001) | 0.68 ± 0.01 (P = 0.005) |
Rotation | 0.79 ± 0.04 (P = 0.005) | 0.75 ± 0.06 (P = 0.003) | 0.72 ± 0.04 (P < 0.001) | 0.53 ± 0.08 (P < 0.001) | 0.48 ± 0.09 (P = 0.001) | 0.67 ± 0.01 (P < 0.001) |
StyleGAN2-AUG+Flipping | 0.74 ± 0.11 (P = 0.029) | 0.73 ± 0.12 (P = 0.007) | 0.73 ± 0.06 (P = 0.015) | 0.65 ± 0.04 (P = 0.280) | 0.61 ± 0.04 (P = 0.043) | 0.68 ± 0.01 (P = 0.035) |
StyleGAN2-AUG+Rotation | 0.79 ± 0.05 (P = 0.063) | 0.75 ± 0.07 (P = 0.001) | 0.73 ± 0.04 (P < 0.001) | 0.74 ± 0.05 (P = 0.190) | 0.67 ± 0.06 (P = 1.000) | 0.69 ± 0.02 (P = 0.481) |
