A comparative study on polyp classification using convolutional neural networks.

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Colorectal cancer is a leading cause of cancer death, often originating from polyps in the colon or rectum.
• Early detection of polyp type is crucial for preventing cancer progression and improving patient outcomes.
• Visual polyp classification is challenging due to endoscopic imaging limitations and subjective human interpretation.

Purpose of the Study:

• To evaluate the efficacy of state-of-the-art deep convolutional neural network (CNN) models for classifying colorectal polyps.
• To compare the diagnostic performance of CNNs against human expert agreement.

Main Methods:

• Trained six state-of-the-art general object classification CNN models end-to-end.
• Utilized a dataset comprising 157 video sequences of hyperplastic and adenomatous polyps.
• Assessed model performance on polyp classification tasks.

Main Results:

• CNN models achieved high accuracy in classifying colorectal polyps.
• The performance of the CNN models was comparable to or exceeded the accuracy reported among gastroenterologists.
• Demonstrated the potential of AI in overcoming challenges of visual polyp classification.

Conclusions:

• State-of-the-art CNNs show significant promise for accurate and objective colorectal polyp classification.
• AI-driven polyp classification can potentially enhance early cancer detection and standardize diagnostic accuracy.
• This research provides a foundation for developing AI tools to assist gastroenterologists in polyp diagnosis.