Deep learning tool improves accuracy of lung nodule malignancy detection

An artificial intelligence (AI) deep learning tool that estimates the malignancy risk of lung nodules achieved high cancer detection rates while significantly reducing false-positive results. Results of the study, which used data from large, multi-site lung cancer screening trials, were published today in Radiology, a journal of the Radiological Society of North America (RSNA).

Lung cancer remains a significant global health issue, causing the most cancer-related deaths worldwide. Screening high-risk individuals with low-dose chest CT has been shown to reduce lung cancer mortality. However, early screening trials have reported high false-positive rates, leading to unnecessary follow-up procedures, increased patient anxiety and health care costs.

Pulmonary nodules-small round or oval growths in the lungs-are common, and identifying which are malignant is challenging in lung cancer screening.

Deep learning offers promising solutions, but robust validation is essential. AI accounts for factors that we might not even see on the CT scan to further assess a nodule as likely to be malignant."

Noa Antonissen, M.D., lead researcher and Ph.D. candidate at Radboud University Medical Center, Nijmegen, the Netherlands

Most current lung cancer screening protocols rely on nodule size, type and growth to estimate malignancy risk. The Pan-Canadian Early Detection of Lung Cancer (PanCan) model, which estimates nodule malignancy risk through a combination of patient and nodule characteristics, illustrates how probability-based tools can refine risk assessment. Such probability-based risk thresholds are increasingly used to guide management protocols. Deep learning offers a promising alternative by enabling fully data-driven predictions, but more evidence is needed before it can be adopted in clinical practice.

In the retrospective study, the researchers trained their in-house developed deep learning algorithm to estimate the risk for malignancy for lung nodules using data from the National Lung Screening Trial which included 16,077 nodules (1,249 malignant).

External testing was conducted using baseline CT scans from the Danish Lung Cancer Screening Trial, the Multicentric Italian Lung Detection trial and the Dutch–Belgian NELSON trial. The pooled cohort included 4,146 participants (median age 58 years, 78% male, median smoking history 38 pack-years) with 7,614 benign and 180 malignant nodules.

The researchers assessed the algorithm's performance for the pooled cohort and two subsets: indeterminate nodules (5-15 mm) and malignant nodules that were size-matched to benign nodules.

"We selected nodules sized 5–15 mm, due to their diagnostic challenges and frequent need for short-term follow-up," Dr. Antonissen said. "Accurate risk classification of these nodules could reduce unnecessary procedures."

For comparison, the algorithm's performance was evaluated against the PanCan model at nodule and participant levels using the area under the receiver operating characteristic curve (AUC), among other parameters. AUC sums up how well a model can produce relative scores to discriminate between positive or negative instances across all classification thresholds.

In the pooled cohort, the deep learning model achieved AUCs of 0.98, 0.96, and 0.94 for cancers diagnosed within one year, two years, and throughout screening, respectively, compared to PanCan at 0.98, 0.94, and 0.93.

For indeterminate nodules (129 malignant, 2,086 benign), the deep learning model significantly outperformed PanCan across all timeframes with AUCs of 0.95, 0.94, 0.90 vs. 0.91, 0.88, 0.86. For the cancers size-matched to benign nodules, (180 malignant, 360 benign), the deep learning model's AUC was 0.79 versus PanCan at 0.60.

At 100% sensitivity for cancers diagnosed within 1 year, the deep learning model classified 68.1% of benign cases as low risk compared to 47.4% using the PanCan model, representing a 39.4% relative reduction in false positives.

"Deep learning algorithms can assist radiologists in deciding whether follow-up imaging is needed, but prospective validation is required to determine the clinical applicability of these tools and to guide their implementation in practice," Dr. Antonissen said. "Reducing false positive results will make lung cancer screening more feasible."