Two-stage Modified Mask R-CNN reduced false positives per scan by 31% and increased patient-level positive predictive value by 10.5% compared to RPN-only model while maintaining high sensitivity (0.892 vs 0.92) in adults undergoing CTPA for pulmonary embolism.
Does a Modified Mask R-CNN with a False-Positive Reduction module improve diagnostic specificity and reduce false positives compared to an RPN-only model in detecting pulmonary embolism on CTPA scans?
Integrating a False-Positive Reduction module into a Region Proposal Network significantly reduces false positives and improves positive predictive value in automated pulmonary embolism detection on CTPA.
Effect estimate: Patient-level sensitivity difference −0.028 (p=0.5); Patient-level PPV improvement from 0.65 to 0.718 (10.5% relative increase, p<0.001); False positives per scan reduced from 0.331 to 0.228 (31% reduction, p<0.001); Patient-level specificity increased from 0.8 to 0.859 (7.4% increase, p<0.001) (95% CI Sensitivity 95% CI: Modified Mask R-CNN (0.821–0.950), RPN-only (0.851–0.966))
Absolute Event Rate: 0.892% vs 0.92%
p-value: p=Sensitivity p=0.5; PPV p<0.001; False positives per scan p<0.001; Specificity p<0.001
Background: High false-positive rates remain a significant challenge in the automated detection of pulmonary embolism (PE) using Computed Tomography Pulmonary Angiography (CTPA). This study evaluated the additional value of a False-Positive Reduction (FPR) module integrated into a Region Proposal Network (RPN). Methods: A retrospective analysis of 303 CTPA scans (163 PE-positive and 140 PE-negative) was conducted from a single tertiary institution. Both models were additionally validated on an independent external cohort of 100 CTPA scans (50 PE-positive and 50 PE-negative) from the RSNA PE Challenge dataset. The diagnostic performance of the one-stage RPN-only model was compared with that of a two-stage Modified Mask R-CNN (Region-based Convolutional Neural Network) incorporating the FPR module. Results: The Modified Mask R-CNN exhibited significant improvement in terms of specificity. The false-positive rate per scan decreased by 31% in comparison to the RPN-only model. Although there was a slight reduction in patient-level sensitivity, the Positive Predictive Value significantly increased by 10.5%. Additionally, patient-level specificity for emboli with a volume ≥ 1000 mm3 increased, reflecting a 7.4% relative improvement in detecting clinically significant emboli. Conclusions: The Modified Mask R-CNN significantly reduced false positives while maintaining high sensitivity over a region proposal network.
Lee et al. (Mon,) conducted a other in Adults (mean age 63.2 years) undergoing computed tomography pulmonary angiography suspected of or diagnosed with pulmonary embolism (n=303). Two-stage Modified Mask R-CNN with False-Positive Reduction (FPR) module vs. One-stage Region Proposal Network (RPN) only model was evaluated on Patient-level diagnostic performance of pulmonary embolism detection including sensitivity, specificity, positive predictive value (PPV), and false positive rate per scan (Patient-level sensitivity difference −0.028 (p=0.5); Patient-level PPV improvement from 0.65 to 0.718 (10.5% relative increase, p<0.001); False positives per scan reduced from 0.331 to 0.228 (31% reduction, p<0.001); Patient-level specificity increased from 0.8 to 0.859 (7.4% increase, p<0.001), 95% CI Sensitivity 95% CI: Modified Mask R-CNN (0.821–0.950), RPN-only (0.851–0.966), p=Sensitivity p=0.5; PPV p<0.001; False positives per scan p<0.001; Specificity p<0.001). Two-stage Modified Mask R-CNN reduced false positives per scan by 31% and increased patient-level positive predictive value by 10.5% compared to RPN-only model while maintaining high sensitivity (0.892 vs 0.92) in adults undergoing CTPA for pulmonary embolism.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: