A Real-Time Polyp-Detection System with Clinical Application in Colonoscopy Using Deep Convolutional Neural Networks.

Area of Science:

• Medical Imaging
• Gastroenterology
• Artificial Intelligence

Background:

• Colorectal cancer (CRC) is a significant global health concern, with colonoscopy being the primary prevention method.
• Gastroenterologists may miss polyps during colonoscopies, highlighting the need for assistive technologies.
• Existing automated polyp detection systems are largely confined to research and not clinically implemented.

Purpose of the Study:

• To develop and clinically implement the first fully open-source automated polyp detection system.
• To enhance polyp detection accuracy and assist gastroenterologists during colonoscopies.
• To provide a system that surpasses current literature benchmarks.

Main Methods:

• Developed ENDOMIND-Advanced, an automated polyp detection system.
• Created a comprehensive dataset of over 500,000 annotated images by combining hospital data and open-source datasets.
• Utilized a video detection post-processing technique for real-time image stream analysis.

Main Results:

• ENDOMIND-Advanced achieved a 90.24% F1-score on the CVC-VideoClinicDB benchmark, outperforming existing systems.
• The system is integrated into a prototype ready for clinical intervention.
• Demonstrated superior performance compared to the best-known systems in the literature.

Conclusions:

• ENDOMIND-Advanced represents a significant advancement in automated polyp detection for clinical use.
• The open-source nature of the system promotes wider adoption and further research.
• This technology has the potential to improve colorectal cancer prevention by reducing missed polyps.