Classification of Informative Frames in Colonoscopy Videos Using Convolutional Neural Networks with Binarized Weights

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Colorectal cancer is a prevalent United States cancer.
• Colon polyps are a primary cause of colorectal cancer.
• Early polyp detection significantly improves cancer treatment outcomes.

Purpose of the Study:

• To propose a novel classification method for informative colonoscopy frames.
• To develop a lightweight convolutional neural network (CNN) for polyp detection.
• To enhance the feasibility of implementing AI in medical hardware for colonoscopy.

Main Methods:

• A novel classification of informative frames using a CNN with binarized weights.
• Training the CNN with labeled colonoscopy frames.
• Utilizing binarized weights and kernels to reduce CNN size for medical hardware.

Main Results:

• The proposed CNN was evaluated on the Asu Mayo Test clinic database.
• The method achieved a dice score of 71.20%.
• The classification accuracy exceeded 90% on the test dataset.

Conclusions:

• The binarized CNN is effective for classifying informative frames in colonoscopy videos.
• The reduced model size makes it suitable for deployment on medical hardware.
• This approach shows promise for improving early detection of colorectal cancer through AI.