Training data enhancements for improving colonic polyp detection using deep convolutional neural networks.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Deep learning techniques have advanced polyp detection.
• Small datasets and lack of data variability are key limitations.
• Publicly available colonoscopic datasets often suffer from these issues.

Purpose of the Study:

• To address data limitations in polyp detection.
• To improve polyp detection accuracy and recall.
• To enhance the diversity and number of samples in colonoscopic datasets.

Main Methods:

• Augmenting existing colonoscopic images by inserting polyps into realistic regions.
• Generating novel and varied polyps using generative adversarial networks (GANs).
• Utilizing a Faster R-CNN model to assess the effectiveness of the data enhancement strategy.

Main Results:

• The proposed data enhancement method improved polyp detection performance.
• A reduction in the false-negative rate was observed.
• Enhanced recall metrics were achieved compared to the original dataset and existing studies.

Conclusions:

• The method effectively increases data variability and sample size for polyp detection.
• The approach shows potential for improving computer-assisted medical image analysis.
• Results indicate a significant improvement in polyp detection rates and recall values.