Insights into Imaging

Table 2 Prediction accuracies of models for different testing datasets using various training sample sizes

From: Generalizable transfer learning of automated tumor segmentation from cervical cancers toward a universal model for uterine malignancies in diffusion-weighted MRI

Model	Training data	Test dataset
Model	Training data	CX (n = 25)	UT (n = 64)	All (n = 89)
UT-only model	UT (n = 13)	0.24 (0.19, 0.28)	0.49 (0.44, 0.54)	0.43 (0.39, 0.48)
	UT (n = 51)	0.35 (0.32, 0.39)	0.65 (0.61, 0.69)	0.62 (0.53, 0.67)
	UT (n = 256)	0.54 (0.49, 0.58)	0.79 (0.75, 0.83)	0.72 (0.67, 0.76)
TL model	UT (n = 13)	0.69 (0.65, 0.73)	0.61 (0.57, 0.65)	0.62 (0.58, 0.66)
	UT (n = 51)	0.71 (0.66, 0.75)	0.70 (0.65, 0.74)	0.70 (0.66, 0.74)
	UT (n = 256)	0.72 (0.68, 0.77)	0.71 (0.67, 0.76)	0.72 (0.68, 0.76)
Aggregated model	CX (n = 144) + UT (n = 13)	0.69 (0.62, 0.74)	0.48 (0.43, 0.53)	0.55 (0.46, 0.65)
	CX (n = 144) + UT (n = 51)	0.71 (0.66, 0.75)	0.66 (0.62, 0.71)	0.68 (0.64, 0.73)
	CX (n = 144) + UT (n = 256)	0.91 (0.87, 0.94)	0.74 (0.71, 0.78)	0.81 (0.76, 0.85)
Pretrained CX model	CX (n = 144)	0.77 (0.73, 0.81)	0.31 (0.25, 0.34)	0.43 (0.35, 0.51)

Data are presented in mean with parentheses for 95% CI
CX Cervical dataset, UT Uterine dataset

Back to article page