Research on automatic recognition radiomics algorithm for early sacroiliac arthritis based on sacroiliac MRI imaging

Table 2 AUC, accuracy, sensitivity and specificity of T1WI, T2WI and bimodal (T1 + T2) models

Model	Group	AUC	95% CI	Accuracy	Sensitivity	Specificity
LR-T1WI	Training	0.884	0.817–0.951	0.800	0.800	0.800
LR-T1WI	Testing	0.848	0.669–1.000	0.818	0.786	0.875
SVM-T1WI	Training	0.943	0.893–0.993	0.878	0.836	0.943
SVM-T1WI	Testing	0.875	0.678–1.000	0.909	0.929	0.875
LightGBM -T1WI	Training	0.910	0.853–0.967	0.811	0.745	0.914
LightGBM -T1WI	Testing	0.790	0.547–1.000	0.818	0.857	0.750
LR-T2WI	Training	0.874	0.799–0.950	0.820	0.833	0.800
LR-T2WI	Testing	0.902	0.763–1.000	0.864	0.857	0.875
SVM-T2WI	Training	0.975	0.948–1.000	0.933	0.889	0.750
SVM-T2WI	Testing	0.902	0.762–1.000	0.864	0.889	0.800
LightGBM -T2WI	Training	0.929	0.878–0.979	0.888	0.944	0.800
LightGBM -T2WI	Testing	0.911	0.780–1.000	0.864	0.786	1.000
LR-bimodal	Training	0.926	0.876–0.976	0.831	0.759	0.943
LR-bimodal	Testing	0.902	0.750–1.000	0.909	0.929	0.875
SVM-bimodal	Training	0.974	0.946–1.000	0.921	0.889	0.971
SVM-bimodal	Testing	0.964	0.888–1.000	0.955	1.000	0.875
LightGBM—bimodal	Training	0.926	0.870–0.981	0.876	0.870	0.886
LightGBM—bimodal	Testing	0.821	0.624–1.000	0.773	0.714	0.875

ISSN: 1749-799X