An AI deep learning algorithm for detecting pulmonary nodules on ultra-low-dose CT in an emergency setting

The OPTIMACT trial investigated the added value of ultra-low-dose CT (ULDCT) compared to chest x-ray in patients suspected of pulmonary disease at the emergency department (ED). As expected, more incidental pulmonary nodules required follow-up in the ULDCT-arm (4.5%) compared to the CXR-arm (0.6%) of the trial. The number of these nodules in the ULDCT-arm was lower than reported in other studies on chest CT at the ED (8.3% to 9.9%).

In the hectic work environment of the ED, the primary focus lies on the acute problem of presentation; the focus on relevant incidental findings may be lower compared to the outpatient setting. We hypothesized that in this situation an artificial intelligence (AI) algorithm for detecting pulmonary nodules might be of additional value, aiding early pulmonary cancer detection.

We found that by using AI, 5.8 times more true positive nodules are found; additionally, 43 times more false positives are found, a majority of which are found in patients with major abnormalities. The high number of false positive results needs to be put into perspective. For example, in patients with bronchopneumonia, the pattern is instantly recognized by the radiologist; scrutiny of all false positive results is not necessary.

Still, a reduction of false positive results is desired for clinical practice. This can be done by optimizing an AI algorithm when detecting pulmonary nodules, requiring follow-up specifically for ULDCT. We believe that further development of AI systems in nodule characterization will improve cancer risk prediction of incidental pulmonary nodules, contributing to a further reduction in false positive results.

Key points:

• An AI deep learning algorithm was tested on 870 ULDCT examinations acquired in the ED.
• AI detected 5.8 times more pulmonary nodules requiring follow-up (true positives).
• AI resulted in the detection of 42.9 times more false positive results, clustered in patients with major abnormalities.
• AI in the ED setting may aid in early pulmonary cancer detection with a high trade-off in terms of false positives.

Article: An AI deep learning algorithm for detecting pulmonary nodules on ultra-low-dose CT in an emergency setting: a reader study

Authors: Inge A. H. van den Berk, Colin Jacobs, Maadrika M. N. P. Kanglie, Onno M. Mets, Miranda Snoeren, Alexander D. Montauban van Swijndregt, Elisabeth M. Taal, Tjitske S. R. van Engelen, Jan M. Prins, Shandra Bipat, Patrick M. M. Bossuyt, Jaap Stoker & The OPTIMACT study group