AI detects pneumoperitoneum on abdominal CT with high accuracy, though trace volumes remain challenging. In a multicenter study, an AI model achieved an AUC of 0.97 on internal testing. External validation showed 84.3% sensitivity and 89.6% specificity, improving to 96% sensitivity when excluding free air under 1 mL. Volume quantification showed excellent agreement with reference standards.

An AI model for detecting pneumoperitoneum on abdominal CT achieved 84.3% sensitivity and 89.6% specificity in an external validation set. Sensitivity increased to 96% when cases with <1 mL free gas were excluded, showing strong performance for clinically relevant volumes.

• The AI algorithm was developed and tested on 2072 patients across multiple centers and externally validated on 214 emergency CTs.
• Primary performance in the external validation set included an accuracy of 86.9%, a PPV of 89.2%, and an NPV of 84.8%.
• A study limitation is the reliance on diagnostic reports as the reference standard, and prospective studies are needed to confirm clinical impact.

Automated summary

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