AI-assisted nodule detection could boost early lung cancer diagnosis

The combination of artificial intelligence (AI)-based nodule detection software and a structured, navigator-tracked follow-up clinic helped identify early-stage lung cancer in a pilot program at six European hospitals, according to a study published in Clinical Lung Cancer.

Lung cancer, the leading cause of cancer death worldwide, is often diagnosed in advanced stages, the study authors noted. While low-dose CT screening had been shown to reduce lung cancer mortality, narrow eligibility criteria and uneven access limit its reach. The new study explores an alternate route to early detection: incidental pulmonary nodules (IPNs) in chest CT scans ordered for unrelated reasons.

“Pulmonary nodules are commonly present on chest CT but are easily overlooked,” the authors explained. “Furthermore, studies conducted in the United States of America found that two out of three patients with detected IPN do not receive appropriate follow-up care. In Europe, real-world data on IPN incidence and management are limited, highlighting a significant knowledge gap, though comparable challenges are expected.”

In the PINPOINT trial, three academic and three non-academic hospitals in the Netherlands, Italy, Spain and Portugal added two new tools to their radiology workflows to test the feasibility of the process. The first was a computer-assisted detection (CAD) software to flag and measure pulmonary nodules on every chest CT scan, and the second was a web-based virtual nodule clinic (VNC) that lets a designated navigator track referrals and follow-ups. Radiologists used the CAD software as a concurrent reader and then decided which nodules were actionable under British Thoracic Society or Fleischner Society guidelines.

“Incorporation of CAD within the hospital [Picture Archiving and Communication System] and employing a stand-alone web-based application as VNC allowed for implementation on top of existing workflows, which enhanced the transferability of the program to different clinical settings and information technology structures,” the authors explained.

The mean duration was 14 months, and the program analyzed a total of 114,644 chest CT scans from 65,344 patients. CAD software provided results for 99.5% of scans, flagging abnormalities in 62.7% of scanned patients, or 40,954 people. Radiologists identified 619 patients with IPN to be managed in the VNC, and 10 of those patients, or 1.6%, were diagnosed with lung cancer. All of the lung cancers diagnosed were early stage.

All sites in the study maintained CAD use during the 2-year pilot program, and most reported willingness to adopt the tool following the pilot phase.

“The remarkably high success rate of CAD analysis for all routine chest CT scans across participating centers demonstrated reliable performance in routine practice,” the authors wrote. “Reported improvements in diagnostic confidence as well as a high nodule detection rate highlight the potential of CAD to enhance diagnostic accuracy and certainty, consistent with the positive results of previous validation studies.”

The VNC proved more challenging to maintain, as uptake was less consistent. Therefore, the proportion of patients who were followed up in the study likely do not capture the true prevalence of actionable IPNs in individuals undergoing chest CT, the authors noted.

The PINPOINT trial is still ongoing to evaluate whether implementing CAD and VNC leads to a meaningful shift in early lung cancer detection compared with a historical baseline. The insights gleaned from the pilot program will inform strategies to ensure consistent uptake of the program and comprehensive outcome assessment.

“This pilot provides important insights for the design of a future clinical efficacy study and future implementation of the proposed IPN program,” the authors concluded. “As CAD demonstrated high technical reliability and detection performance, and sustained use across centers suggests clinical acceptability, it is ready for evaluation of clinical effectiveness.”