Deep learning algorithm detects in-stent restenosis on CT angiography with higher sensitivity than radiologists

For detecting significant in-stent restenosis on coronary CT angiography, a deep learning algorithm achieved 75.4% sensitivity versus 43.9% and 56.1% for two radiologists, while maintaining similar overall accuracy (87.8%).

• In a retrospective internal validation set of 131 patients (222 stents), the algorithm's sensitivity (75.4%) was significantly higher than both readers (43.9% and 56.1%; p<0.05), with no significant difference in overall accuracy (McNemar's p=0.133 and 0.150).
• The algorithm's NPV (91.6%) was significantly higher than Reader1 (p=0.026), while specificity was numerically slightly lower.
• Subgroup analyses confirmed higher sensitivity across scanner platforms (GE systems p=0.016) and at 80 kVp (p=0.006).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.