Feasibility of Using High-Resolution Computed Tomography Features for Invasiveness Differentiation of Malignant Nodules Manifesting as Ground-Glass Nodules

Table 2

Detailed diagnostic performance of the three models in the training and testing cohorts.

	Semantic model		Texture model		Combined model
	Training cohort	Testing cohort	Training cohort	Testing cohort	Training cohort	Testing cohort
Sensitivity	84.6% (95%CI: 78.1%–91.2%)	65.5% (95%CI: 48.2%–82.3%)	78.6% (95%CI: 71.2%–86.0%)	82.8% (95%CI: 69.0%–96.5%)	79.5% (95%CI: 72.2–86.8%)	79.3% (95%CI: 64.6%–94.1%)
Specificity	71.2% (95%CI:62.4%–79.9%)	88.4% (95%CI: 76.2%-1)	75.0% (95%CI: 66.7%–83.3%)	92.3% (95%CI: 82.1%-1)	78.8% (95%CI: 71.0%–86.7%)	92.3% (95%CI: 82.1%-1)
PPV (%)	76.7% (95%CI:69.5%–84.0%)	86.4% (95%CI: 72.0%-1)	78.0% (95%CI: 70.5%–85.4%)	92.3% (95%CI: 82.1%-1)	80.9% (95%CI: 73.7%–88.1%)	92.0% (95%CI: 81.3%-1)
NPV (%)	80.4% (95%CI:72.3%–88.5%)	69.7% (95%CI: 54.0%–85.4%)	75.7% (95%CI: 67.4%–84.0%)	82.8% (95%CI: 69.0%–96.5%)	77.4% (95%CI: 69.4%–85.3%)	80.0% (65.7%–94.3%)