Real-Time AI-Assisted Detection of Colonic Lesions Using the iIDEAS Intelligent-C Module: A Prospective Single-Center Validation Study

Abstract

Aims

Colorectal cancer (CRC) remains a major cause of cancer mortality, with up to 26% of adenomas missed on conventional colonoscopy. Artificial intelligence (AI)–based computer-aided detection (CADe) systems may reduce these miss rates. We aimed to validate the iIDEAS-C CADe module in a prospective cohort, with co-primary endpoints of adenomas per colonoscopy (APC) and lesion-level miss rate.

Methods and Results

In this prospective, nonrandomized, single-center study (NCT05784935), consecutive adults (18–75 years) undergoing screening, surveillance, or diagnostic colonoscopy were enrolled between February and May 2023. Each colon segment was inspected twice: an initial blinded withdrawal without CADe, followed by a second inspection with iIDEAS-C. Missed lesions were those identified by CADe during the second inspection and confirmed endoscopically. Secondary outcomes included PDR, ADR, PPC, lesion characteristics, and procedural times.

Two hundred patients (mean age 44.1 ± 12.7 years; 64% male) were included. Eighty-two polyps were detected across 42 patients (PDR = 21%). Endoscopists detected 61/82 polyps (74.4%), yielding a per-polyp miss rate of 25.6%. The iIDEAS-C module detected all 82 lesions, achieving 0% per-polyp and per-adenoma miss rate. Missed lesions were mostly diminutive and proximal. PPC increased from 0.305 to 0.410 (p < 0.0001), PDR from 17.5% to 21.0% (p = 0.016), and APC from 0.115 to 0.140 (p = 0.029). No adverse events occurred.

Conclusion

The iIDEAS-C module demonstrated a 0% lesion-level miss rate and significantly improved key detection metrics, especially for small proximal lesions. Larger multicenter studies are warranted to confirm these findings and assess long-term outcomes. ClinicalTrials.gov: NCT05784935.

Trial Registration: ClinicalTrials.gov identifier: NCT05784935.