Opportunistic AI detects colorectal cancer using routine, noncontrast CT (radiologybusiness.com)

Researchers have introduced a promising new AI tool named COCA (Colorectal Cancer detection with AI) that detects colorectal cancer using routine noncontrast CT scans, which are typically performed for other medical reasons. This development is significant as it offers a noninvasive alternative to colonoscopy, which is often underutilized despite being the current standard for colorectal cancer diagnosis. COCA leverages the vast number of abdominal and pelvic CT scans performed annually to identify potential cancerous lesions, transforming these routine procedures into valuable screening opportunities. The COCA model employs a sophisticated joint lesion segmentation and classification architecture, optimized through mixed-supervised learning. In extensive testing involving over 30,000 scans, COCA demonstrated impressive metrics with a sensitivity of 88.2% and a specificity of 99.5%, significantly outperforming radiologists without AI assistance. This advancement could potentially address low patient adherence to existing screening methods by removing the need for bowel preparation while maintaining high diagnostic accuracy. As the team continues to refine the tool, its integration into clinical practice could reshape colorectal cancer screening protocols and improve early detection rates in diverse healthcare settings.