Machine learning-based CT radiomics approach for predicting WHO/ISUP nuclear grade of clear cell renal cell carcinoma: an exploratory and comparative study

Insights into Imaging

Table 5 Predictive performance of three classifiers: SVM-LASSO, SVM-RFE, and SVM-ReliefF

Classifier	SVM-LASSO			SVM-RFE			SVM-ReliefF
Classifier	Training cohort	Validation cohort	Testing cohort	Training cohort	Validation cohort	Testing cohort	Training cohort	Validation cohort	Testing cohort
AUC	0.838 [0.721–0.934]	0.795 [0.674–0.899]	0.754 [0.652–0.889]	0.842 [0.747–0.945]	0.803 [0.687–0.904]	0.761 [0.648–0.893]	0.860 [0.759–0.963]	0.824 [0.736–0.915]	0.787 [0.710–0.892]
Accuracy (%)	76.70 [68.05–85.26]	73.94 [73.49–89.75]	68.78 [55.67–82.89]	77.36 [65.35–86.25]	74.11 [62.98–87.19]	71.58 [59.87–83.74]	83.61 [75.85–92.65]	77.14 [69.38–86.05]	73.42 [61.63–82.66]
Sensitivity (%)	77.57 [64.45–87.31]	73.48 [60.67–86.48]	70.07 [61.71–82.99]	80.33 [69.35–89.79]	75.47 [63.78–89.36]	71.25 [61.36–80.74]	84.78 [74.36–93.74]	78.72 [64.35–91.48]	82.15 [73.74–91.92]
Specificity (%)	81.22 [72.19–89.24]	79.63 [61.34–70.49]	80.30 [71.58–89.43]	84.31 [73.74–92.18]	83.04 [71.39–95.46]	79.22 [66.57–88.34]	82.67 [74.35–93.14]	80.33 [69.74–92.26]	76.48 [63.36–89.17]
PPV (%)	76.88 [62.38–85.69]	73.86 [59.67–88.74]	69.71 [60.98–78.35]	79.46 [64.38–91.59]	74.78 [62.88–86.74]	72.98 [67.64–83.35]	77.79 [69.16–85.97]	75.98 [68.30–83.57]	74.82 [65.87–83.64]
NPV (%)	84.12 [76.41–95.37]	80.48 [71.70–92.26]	79.25 [66.28–91.33]	85.34 [75.65–96.43]	83.92 [74.54–91.35]	80.25 [72.36–92.17]	83.90 [72.74–95.46]	80.43 [71.32–89.91]	82.62 [71.64–92.35]