Colorectal cancer detection using non-contrast CT and deep learning: a multicenter and international cohort study
Menée à partir de données portant sur 1 321 patients atteints d'un cancer colorectal et sur 1 357 témoins en bonne santé puis validée sur 29 496 patients supplémentaires, cette étude multicentrique évalue la sensibilité et la spécificité d'une méthode combinant intelligence artificielle et tomographie numérique sans rehaussement de contraste pour détecter la maladie
Background : Colorectal cancer (CRC) is a leading cause of cancer deaths, with early screening vital to reduce mortality. While methods like colonoscopy and CT colonography are available, they face challenges such as bowel prep, invasiveness, and low adherence. We aimed to develop COCA, a novel, non-invasive, cost-effective, and scalable method for CRC screening using non-contrast CT scans.
Patients and methods : This retrospective, multicenter, and international study included 1,321 CRC patients and 1,357 normal controls from two centers to develop COCA. We enhance the CRC detection capabilities of COCA by employing a joint lesion segmentation and classification architecture, optimized with mixed-supervised learning. For validation, we gathered abdominal and pelvic CT data from four external centers and chest CT data from four centers. A reader study involving 10 radiologists with varying levels of experience evaluated diagnostic performance on non-contrast CT first without COCA assistance and then with it. Additionally, we evaluated both the initial and iteratively improved versions of COCA in two real-world, multi-scenario cohorts comprising 27,433 consecutive patients.
Results : In a multicenter and international validation involving 2,053 patients across six centers, COCA demonstrated an area under the curve (AUC) ranging from 0.967 to 0.996 for CRC detection. COCA improved CRC detection sensitivity by 20.4% and specificity by 5.4% compared to radiologists. In the first real-world multi-scenario validation with 9,016 consecutive patients, COCA achieved a sensitivity of 88.2% and specificity of 99.5% for CRC detection. In the second external real-world validation involving 18,427 consecutive patients, COCA maintained a sensitivity of 86.6% and specificity of 99.8%, with a positive predictive value of 63.4%.
Conclusions : COCA demonstrated robust performance across various clinical scenarios, including physical exams, emergency departments, outpatient, and inpatient settings, effectively preventing missed CRC diagnoses. These findings suggest that COCA could serve as a potential tool for large-scale opportunistic CRC screening.
Annals of Oncology , résumé, 2026