AI May Catch High-Impact Incidental Findings on Routine Oncology CT

AI applications designed to automatically screen routine oncology CT scans may help radiologists identify life-threatening or other important incidental findings earlier, according to research presented Monday.

In a pair of retrospective studies from Gustave Roussy Cancer Center in France, two FDA-cleared algorithms—CINA-iPE and CINA-VCF Quantix (Avicenna.AI)—demonstrated strong performance in detecting incidental pulmonary embolism (iPE) and vertebral compression fractures (VCFs) during standard oncologic imaging.

Cancer patients often undergo frequent CT imaging to monitor disease progression, but Sarah Quenet, director of product at Avicenna.AI, who presented the study said some findings may be missed.

“Oncology patients undergo many CT scans, but radiologists are under intense time pressure, so important incidental findings can be overlooked,” Quenet said. “We wanted to see whether AI could ‘re-read’ these routine scans and flag clinically significant but unexpected problems early enough to prevent bad patient outcomes.”

The team selected iPE and VCFs because both conditions are common in cancer patients, affect quality of life and survival and are highly treatable when detected early.

AI Detects Rare but Urgent Pulmonary Emboli

In a cohort of 3,047 patients undergoing follow-up contrast-enhanced CT, the iPE algorithm generated 104 alerts, with 36 confirmed pulmonary emboli. The tool delivered high sensitivity and specificity, with most false positives attributed to motion artifacts or enlarged lymph nodes.

Quenet noted that this combination is “particularly meaningful for a condition like incidental PE, which has a very low prevalence in routine oncology CT scans.” Maintaining specificity was key to limiting unnecessary interruptions, while high sensitivity ensured that rare but critical events were captured.

Seven patients in the study experienced delayed diagnoses—five requiring hospitalization—but all of these PEs were promptly detected by the AI system.

Silent Vertebral Compression Fractures Brought to Attention

Cancer and its treatments can accelerate bone loss—often via profound estrogen depletion—leading to silent vertebral fractures. The VCF algorithm was tested in 1,501 stage IV cancer patients and identified 501 suspected fractures, with 436 confirmed.

“We were struck by the fact that 81% of confirmed vertebral compression fractures were not mentioned in the original radiology reports,” Quenet said.

Among unreported fractures, 10 were classified as severe, and nine were considered eligible for cementoplasty. “That means AI is not just finding ‘incidental trivia,’ but uncovering lesions that could benefit from minimally invasive treatment,” Quenet said.

Both AI tools run automatically in PACS without special acquisition protocols—which study authors said is essential for busy oncology practices where dozens of surveillance CTs may be interpreted each day. The system reviews every eligible scan in the background and pushes only the most concerning cases to the radiologist.

Quenet said that the goal is not to change how radiologists read CT exams, but to provide a dependable layer of backup that supports earlier intervention and more informed patient management.

“AI is not replacing the radiologist; it is giving you a second chance to see what matters most for your patient,” Quenet said. She continued that subtle but clinically meaningful findings “can occasionally be overlooked in the context of complex oncology imaging and AI offers a consistent mechanism to highlight abnormalities that deserve a second look,” Quenet concluded.

Access the presentation, “Artificial Intelligence in Oncology: Opportunistic Screening for Incidental Pulmonary Embolism And Vertebral Compression Fractures On Oncologic Assessment CT Scans,” (M2-STCE1-1) on demand at RSNA.org/MeetingCentral.