A cautionary tale for radiologists who use AI to rule out suspicious findings
A new case report should serve as a cautionary tale for radiologists and other providers who utilize artificial intelligence to rule out suspicious findings on imaging.
Published in Emergency Radiology, the report details the case of an AI program that incorrectly flagged a chest radiograph for a pneumothorax. Though the care team’s interpretation differed from AI's ruling, the patient was sent for additional imaging and assessment as a precaution.
Experts involved in the case suggested their experience should be considered by others who have integrated AI into their clinical workflows.
“Artificial intelligence has become increasingly integrated into chest radiograph interpretation, demonstrating excellent diagnostic performance for several thoracic abnormalities, including pneumothorax,” Fulvio Cacciapuoti, MD, with the division of cardiology at Antonio Cardarelli Hospital in Italy, and colleagues noted. “Nevertheless, current AI systems remain predominantly image-based and do not inherently incorporate procedural or clinical information into diagnostic interpretation.”
The case involved an 81-year-old male who had just undergone an uncomplicated, left-sided permanent pacemaker implantation. The standard protocol required a post-procedural chest radiograph, which was interpreted by both the care team and an AI program that specializes in chest X-ray reads. The tool flagged the images for a right apical pneumothorax and highlighted the suspected abnormality with automated visual annotation.
The treating team determined that the finding was clinically discordant with the patient’s presentation. The procedure involved left-sided venous access, while the patient had normal oxygen saturation, stable vital signs and no respiratory symptoms. Because of the discrepancy between the AI finding and the clinical circumstances, providers obtained a chest CT, which definitively excluded a pneumothorax and confirmed the AI alert was a false positive. The patient required no additional treatment and recovered without complications.
AI used for post-procedural imaging may be particularly susceptible to limitations because expected findings and complications can vary depending on the procedure performed, access site and clinical circumstances. In this case, knowledge of the pacemaker implantation and the patient’s stable clinical status helped clinicians recognize that the AI-generated finding warranted further scrutiny rather than immediate treatment.
“This case highlights the importance of integrating AI-generated findings with procedural details, bedside clinical assessment and, when clinically indicated, confirmatory imaging in the presence of clinicoradiological discordance,” the authors noted. “Although AI can accurately identify many radiographic abnormalities, accurate diagnosis ultimately depends on integrating imaging findings with procedural information, clinical presentation and pre-test probability.”
Read more here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.