AI “Adjunctive” Boosts CT Sensitivity by 40% for Detecting Lung Metastases in Colorectal Cancer Patients

A newly published study in the American Journal of Roentgenology demonstrates that using AI alongside radiologist reads significantly enhances the detection of lung metastases in colorectal cancer (CRC) patients. According to the report, CT interpretation with adjunctive AI improved sensitivity by 40.3%—from 32.1% without AI to 72.4% when supported by AI—as well as increasing treatment-management changes by over 30% (55.2% vs. 25.0%).

The research compared 663 CRC patients (mean age 63) whose CT scans were read using AI-augmented software luCAS Plus v1.00.04 against 647 patients (mean age 64) who underwent radiologist-only interpretation. Notably, adjunctive AI achieved similar specificity (99.7% vs. 98.9%) and negative predictive value (98.8% vs. 97.0%) compared to unassisted reads.

Lead author Dr. Sowon Jang and colleagues from Seoul National University Bundang Hospital observed that nodules missed by AI-assisted radiologists were generally smaller (mean 3.1 mm) than those missed without AI (mean 4.7 mm). “The findings support the application of AI systems for lung nodule detection on chest CT performed for metastasis surveillance. The observed high sensitivity for AI-assisted interpretation is critical in this setting given the substantial impact of delayed diagnosis on clinical outcomes,” the authors wrote.

A separate evaluation of the AI tool used on its own—without radiologist review—found even higher sensitivity (82.8%), but much lower specificity (39.9%) and overall accuracy (41.8%). The researchers attributed many false positives to AI alone: “The radiologists may have used additional CT features of the nodules … to classify AI-reported nodules as benign,” and to knowledge of scheduled follow-up scans.

Three key takeaways from the study:

1. AI-enhanced CT interpretation markedly improved lung-metastasis detection in colorectal cancer patients (72.4% vs. 32.1%).

   ADVERTISEMENT

2. Clinical impact: Use of AI led to over a 30% increase in changes to patient management (55.2% vs. 25.0%).

3. Standalone AI limitation: While it further raised sensitivity, standalone AI had substantially lower specificity and accuracy, underscoring the role of radiologist oversight.

The authors noted some limitations: this was a single-center retrospective study with low metastasis prevalence, no analysis based on radiologist experience, no per-lesion data, and no evaluation of how AI affected reading times.