Virtual crystal violet chromoendoscopy: a digital alternative to dye-based staining

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Crystal violet (CV) chromoendoscopy aids colorectal lesion characterization but has safety limitations.
• Pit pattern analysis is crucial for diagnosing colorectal lesions.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-based virtual crystal violet (vCV) chromoendoscopy technique.
• To assess if vCV can reproduce CV-equivalent images without dye application.

Main Methods:

• An unsupervised image-to-image translation framework (UVCGANv2) was used to train a model for transforming endoscopic images into vCV images.
• A retrospective study analyzed 71 colorectal lesions with paired CV and narrow-band imaging (NBI) images.
• Six endoscopists evaluated vCV, CV, and NBI images for pit pattern interpretation.

Main Results:

• Interobserver agreement for pit pattern interpretation was similar between vCV and conventional CV.
• Quantitative analysis confirmed morphological fidelity between vCV and CV images.
• vCV demonstrated diagnostic accuracy for histological prediction comparable to CV and superior to NBI.

Conclusions:

• AI-based vCV chromoendoscopy effectively replicates the diagnostic interpretability of conventional CV staining.
• This AI technique offers a safe, dye-free alternative for colorectal lesion characterization.