Fig. 8: Comparison of PPV and detection rate ratios at different scoring thresholds in the training set, internal validation set, and two external validation sets. | npj Digital Medicine

Fig. 8: Comparison of PPV and detection rate ratios at different scoring thresholds in the training set, internal validation set, and two external validation sets.

From: Interpretable machine learning model for digital lung cancer prescreening in Chinese populations with missing data

Fig. 8: Comparison of PPV and detection rate ratios at different scoring thresholds in the training set, internal validation set, and two external validation sets.

a Positive predictive values (PPV) at various high and low-risk score thresholds. Exploring a range of high to low-risk thresholds from 72 to 50 points, where a lower score indicates poorer health and a higher risk of lung cancer. b The detection rate of the BOUND model is compared to the higher detection rate of 2.3% observed in current lung cancer LDCT screening, with the detection rate ratio calculated as the BOUND model’s detection rate divided by 2.3%.

Back to article page