Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

White Light Specular Reflection Data Augmentation for Deep Learning Polyp Detection

Jose Angel Nuñez, Fabian Vazquez, Diego Adame, Xiaoyan Fu, Pengfei Gu, Bin Fu

TL;DR

Colorectal polyp detection via deep learning is hampered by white-light specular reflections from endoscopes, causing false positives. The paper introduces White Light Specular Reflection (WLSR) data augmentation, which builds a bank of realistic artificial light spots, excludes prohibited regions, and uses a sliding-window mechanism to place lights in suitable areas, thereby creating harder, more diverse training examples. Evaluated on the Harvard Dataverse DVN/FCBUOR_2021 dataset, WLSR improves detection performance, with a 20% augmentation yielding a 3.2 percentage-point gain in mAP50 over original-data training, and modest gains when combined with other augmentations, while maintaining feasibility for near-real-time use. This approach enhances detector robustness to endoscopic lighting artifacts, offering a practical plug-in augmentation to reduce missed polyps and false positives in clinical settings with open data for reproducibility.

Abstract

Colorectal cancer is one of the deadliest cancers today, but it can be prevented through early detection of malignant polyps in the colon, primarily via colonoscopies. While this method has saved many lives, human error remains a significant challenge, as missing a polyp could have fatal consequences for the patient. Deep learning (DL) polyp detectors offer a promising solution. However, existing DL polyp detectors often mistake white light reflections from the endoscope for polyps, which can lead to false positives.To address this challenge, in this paper, we propose a novel data augmentation approach that artificially adds more white light reflections to create harder training scenarios. Specifically, we first generate a bank of artificial lights using the training dataset. Then we find the regions of the training images that we should not add these artificial lights on. Finally, we propose a sliding window method to add the artificial light to the areas that fit of the training images, resulting in augmented images. By providing the model with more opportunities to make mistakes, we hypothesize that it will also have more chances to learn from those mistakes, ultimately improving its performance in polyp detection. Experimental results demonstrate the effectiveness of our new data augmentation method.

White Light Specular Reflection Data Augmentation for Deep Learning Polyp Detection

TL;DR

Colorectal polyp detection via deep learning is hampered by white-light specular reflections from endoscopes, causing false positives. The paper introduces White Light Specular Reflection (WLSR) data augmentation, which builds a bank of realistic artificial light spots, excludes prohibited regions, and uses a sliding-window mechanism to place lights in suitable areas, thereby creating harder, more diverse training examples. Evaluated on the Harvard Dataverse DVN/FCBUOR_2021 dataset, WLSR improves detection performance, with a 20% augmentation yielding a 3.2 percentage-point gain in mAP50 over original-data training, and modest gains when combined with other augmentations, while maintaining feasibility for near-real-time use. This approach enhances detector robustness to endoscopic lighting artifacts, offering a practical plug-in augmentation to reduce missed polyps and false positives in clinical settings with open data for reproducibility.

Abstract

Colorectal cancer is one of the deadliest cancers today, but it can be prevented through early detection of malignant polyps in the colon, primarily via colonoscopies. While this method has saved many lives, human error remains a significant challenge, as missing a polyp could have fatal consequences for the patient. Deep learning (DL) polyp detectors offer a promising solution. However, existing DL polyp detectors often mistake white light reflections from the endoscope for polyps, which can lead to false positives.To address this challenge, in this paper, we propose a novel data augmentation approach that artificially adds more white light reflections to create harder training scenarios. Specifically, we first generate a bank of artificial lights using the training dataset. Then we find the regions of the training images that we should not add these artificial lights on. Finally, we propose a sliding window method to add the artificial light to the areas that fit of the training images, resulting in augmented images. By providing the model with more opportunities to make mistakes, we hypothesize that it will also have more chances to learn from those mistakes, ultimately improving its performance in polyp detection. Experimental results demonstrate the effectiveness of our new data augmentation method.
Paper Structure (9 sections, 4 figures, 1 table)

This paper contains 9 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Generating Bank of Lights
  4. Finding Prohibited Areas
  5. Finding Areas that Fit
  6. Experiments and Results
  7. Conclusion
  8. Compliance with ethical standards
  9. Acknowledgements

Figures (4)

  • Figure 1: Visual examples highlighting challenges in detecting polyps: (a) small polyps, (b) flat polyps, and (c) white light specular reflections causing confusion with polyps.
  • Figure 2: The overview of our WLSR data augmentation for DL polyp detection.
  • Figure 3: WLSR data augmentation results. Yellow circles indicate the new lights generated by our method, while green boxes highlight the real polyps.
  • Figure 4: Visual results demonstrating the effectiveness of our method.