Multiphase and multiparameter MRI-based radiomics for prediction of tumor response to neoadjuvant therapy in locally advanced rectal cancer

Table 2 Performance of the Model_NoRes_NAT and Model_pCR in the prediction of tumor response to neoadjuvant therapy in LARC patients

	AUC (95%)	Sensitivity (95%)	Specificity (95%)	PPV (95%)	NPV (95%)	Accuracy
Model_NoRes_NAT
Training cohort	0.84 [0.79–0.89]	77.9 [66.7–86.2]	78.6 [73.5–82.9]	45.7 [36.9–54.8]	93.9 [90.2–96.3]	0.78
Testing cohort	0.81 [0.70–0.91]	70.6 [46.9–86.7]	79.7 [69.2–87.3]	44.4 [27.6–62.7]	92.2 [83.0–96.6]	0.78
External validation cohort	0.79 [0.67–0.91]	78.3 [58.1–90.3]	71.3 [61.0–79.7]	41.9 [28.4–56.7]	92.5 [83.7–96.8]	0.73
Model_pCR
Training cohort	0.86 [0.81–0.91]	71.2 [60.1–80.4]	85.4 [80.1–89.5]	61.9 [51.2–71.6]	89.9 [85.1–93.3]	0.82
Testing cohort	0.87 [0.76–0.98]	85.0 [64.0–94.8]	83.3 [71.3–91.0]	65.4 [46.2–80.6]	93.8 [83.2–97.9]	0.84
External validation cohort	0.87 [0.78–0.95]	84.0 [65.4–93.6]	81.2 [71.6–88.1]	56.8 [40.9–71.3]	94.5 [86.7–97.9]	0.82

Data in parentheses are 95% CIs. Model_NoRes_NAT is for non-response prediction and Model_pCR is for pathological complete response prediction. AUC, area under the curve; PPV, positive predictive value; NPV, negative predictive value

ISSN: 1748-717X