Radiomics-based fertility-sparing treatment in endometrial carcinoma: a review

Insights into Imaging

Table 3 The performance of MRI-based radiomics in predicting LNM, CSI and histological grades

Factors and reference	Model and dataset type	Sensitivity	Specificity	Accuracy	AUC
LNM
Ytre-Hauge et al
	Model^R
	Only one set	0.68 (13/19)	0.73 (99/135)	0.73 (112/154)	0.73
Yan et al
	Model^R
	Test set 1	NA	NA	0.80 (291/351)	0.91
	Test set 2	NA	NA	0.89 (240/271)	0.89
Xu et al
	Model^R
	Training set	NA	NA	NA	0.79
	Test set	NA	NA	NA	0.75
	Model^C
	Training set	NA	NA	NA	0.87
	Test set	NA	NA	NA	0.83
	Model^CR1
	Training set	NA	NA	NA	0.89
	Test set	NA	NA	NA	0.88
	Model^CR2
	Training set	NA	NA	NA	0.84
	Test set	NA	NA	NA	0.82
Kristine et al
	Model^WT
	Training set	NA	NA	NA	0.73
	Test set	NA	NA	NA	0.72
	Model^SS
	Training set	NA	NA	NA	0.83
	Test set	NA	NA	NA	0.56
CSI
Ytre-Hauge et al
	Model^R
	Only one set	0.53 (17/32)	0.78 (114/146)	0.74 (131/178)	0.64
Histological grades
Ytre-Hauge et al
	Model^R
	Only one set	NA	NA	NA	0.66
Zheng et al
	Model^R
	Training set	0.72 (NA)	0.86 (NA)	0.79 (NA)	0.87
	Test set	0.77 (NA)	0.90 (NA)	0.83 (NA)	0.89
	Model^ADC
	Training set	0.61 (NA)	0.74 (NA)	0.68 (NA)	0.72
	Test set	0.43 (NA)	0.79 (NA)	0.62 (NA)	0.62
	Model^M
	Training set	0.89 (NA)	0.82 (NA)	0.85 (NA)	0.93
	Test set	0.92 (NA)	0.79 (NA)	0.85 (NA)	0.92

LNM, lymph node metastasis; CSI, cervical stromal invasion; AUC, area under curve
Model^R: Model constructed by radiomics features. Model^C: Model constructed by clinical features. Model^CR1, Model^CR2, Model^M: Models constructed by clinical and radiomics features. Model^ADC: Model constructed by apparent diffusion coefficient value. Model^WT: Model constructed by whole-tumor radiomics features. Model^SS: Model constructed by single-slice radiomics features