Artificial intelligence (AI) software shows efficacy in detecting unsuspected lung cancer on chest radiographs from patients with asymptomatic disease, according to a recent study published in Radiology Advances.
A team of researchers from South Korea conducted the research, with the goal of examining the performance of “commercially available AI software” on lung cancer detection in patients who otherwise appear healthy and have no symptoms of lung cancer.
Chest radiographs from the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial, which was conducted in the US from November 1993 to July 2001, were retrospectively reviewed. The study also evaluated “pathological cancer diagnosis follow-up to 2009.”
The investigators compared the AI predictions to the interpretations of the PLCO radiologists. They also conducted an additional reader study using a subset of data that compared the AI software results to the readings of three experienced radiologists.
The study analyzed 24,370 patients, with a median age of 62 years. Two hundred and thirteen of those patients were also analyzed as part of the reader study, with a median age of 63 years.
According to the findings of the main study, the AI program (0.910) resulted in higher specificity than the radiologists (0.803, P<.001). The positive predictive value for the AI program was also 0.054 compared to 0.032 for the radiologists. However, the AI program (0.326) showed lower sensitivity than the radiologists (0.412, P=.001).
The researchers calibrated the AI software to align with the sensitivity levels of the PLCO radiologists to conduct further investigation. After calibration, the AI program was shown to achieve higher specificity at 0.815 versus 0.803 for the radiologists.
The reader study also resulted in higher sensitivity for the AI program (0.608) when compared to the interpretation of the first (0.588, P=.789) and third (0.588, P=.803) readers. However, the first reader had a higher specificity than the AI software in the sub-study analysis (0.888 for AI vs 0.905 for the first reader, P=.814).
According to the results shown by the second reader in the sub-study analysis, the AI program demonstrated higher specificity (0.888 for AI vs 0.819 for the second reader, P=.153), but lower sensitivity (0.888 for AI vs 0.905 for the first reader, P=.814).
In conclusion, the researchers found that “AI detects lung cancer on chest radiographs among asymptomatic individuals with comparable performance to experienced radiologists.”
Source: Radiology Advances
© 2025 Mashup Media, LLC, a Formedics Property. All Rights Reserved.