Colorectal cancer (CRC) is the third most common cancer and the second most deadly cancer worldwide. Polyps are the main CRC biomarkers mainly categorized according to their microvascular patterns among others. The gold standard for polyp classification is carried out from histological inspection. Despite that an early neoplastic polyps polypectomy increases the survival rate by 90%, the characterization from standard colonoscopies remains challenging and expert-dependent (kappa value = 0.56). Recently, the NICE protocol that includes colonoscopy devices with NBI (narrow band imaging) filter allowed real-time optical diagnosis from a vascular patterns characterization enhancing blood vessel patterns, outperforming the polyp detection rate by 13% w.r.t regarding the optical colonoscopy (OC) analysis.

This work introduces an end-to-end computational strategy that allows the enhancement of standard colonoscopy observations, by including vascular patterns, typically observed from NBI mechanisms. These retrieved synthetic images are achieved under a non-aligned translation task from optical colonoscopy (OC) to NBI including an architecture to achieve an automatic classification from enhanced colonoscopy observations.