Multi-View CT and Background-Negative Training Improve Lung Nodule Detection and Reduce False Positives

Combining axial, coronal, and sagittal views with nodule-free background images in a YOLOv8x pipeline boosted the Competition Performance Metric (CPM) from 0.553 to 0.863 and cut average candidates per scan from 32.61 to 9.11 on LUNA16.

• The axial-only YOLOv8x baseline had a CPM score of 0.553 with 32.61 candidates per scan.
• Adding coronal/sagittal views and background-negative training raised CPM to 0.863 and lowered candidates to 9.11.
• The method preserved high sensitivity while substantially reducing false-positive candidates.

Automated summary

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