AI based colorectal disease detection using real-time screening colonoscopy.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Gastroenterology

Background:

• Colonoscopy is crucial for early colorectal disease detection but faces challenges in diagnostic efficiency and accuracy.
• Distinguishing between various intestinal conditions like polyps, colitis, and colorectal cancer (CRC) requires high precision.

Purpose of the Study:

• To develop and evaluate a deep convolution neural network (CNN) model for accurate identification of colorectal diseases from colonoscopy data.
• To assess the model's performance in differentiating between normal cases and patients with polyps, colitis, or CRC.

Main Methods:

• A CNN model was constructed and trained using a large dataset of 117,055 colonoscopy images from 16,004 individuals.
• The model's diagnostic performance was validated on multi-center real-time colonoscopy videos and images, and compared against skilled endoscopists and novices.
• The model's capability in diagnosing specific polyp types, such as adenomatous and hyperplastic polyps, was also evaluated.

Main Results:

• The CNN model achieved a high accuracy of 0.933 in identifying patients with polyps, colitis, and CRC from normal cases in the validation dataset.
• The model demonstrated accurate diagnostic performance and generalization capabilities across external validation datasets.
• An area under the receiver operating characteristic curve of 0.975 was achieved for diagnosing adenomatous and hyperplastic polyps.

Conclusions:

• The developed CNN model shows significant potential in assisting clinicians with efficient and accurate diagnosis of colorectal diseases.
• The model's ability to generalize across different datasets and its high accuracy in polyp differentiation suggest its utility in clinical decision support.