White Light Specular Reflection Data Augmentation for Deep Learning Polyp Detection
Jose Angel Nuñez Fabian Vazquez Diego Adame Xiaoyan Fu Pengfei Gu Bin Fu · 2025
Read White Light Specular Reflection Data Augmentation for Deep Learning Polyp Detection by Jose Angel Nuñez Fabian Vazquez Diego Adame Xiaoyan Fu Pengfei Gu Bin Fu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-05248.txt.
Introduction
Jose Angel Nuñez and co-authors present a 2025 method to train colon polyp detectors with fake glare. Colon cancer is a top killer, yet early polyp removal can stop it. Camera scopes miss polyps at times, so deep learning aids are in growth. A common error is that bright glare spots on wet tissue look like polyps to the model, raising false alarms.
The team builds a bank of fake glare from real training images. They mark zones where glare must never be added, then slide the fake light over fit zones. Training on the harder set gives the model more chances to err and learn. Tests show the trick improves detector scores. Better training data of this kind costs no new patient scans.
Catalog record
| CREATOR | Jose Angel Nuñez Fabian Vazquez Diego Adame Xiaoyan Fu Pengfei Gu Bin Fu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 18 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-05248
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
White Light Specular Reflection Data Augmentation for Deep Learning Polyp Detection Jose Angel Nuñez Fabian Vazquez Diego Adame Xiaoyan Fu Pengfei Gu Bin Fu 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. † † address: 1 Department of Computer Science, University of Texas Rio Grande Valley, Edinburg, TX, USA 2 The Second Affiliated Hospital of Fujian University of Traditional Chinese Medicine, Fuzhou, Fujian, China 1 Introduction Colorectal cancer (CRC) is a significant global health issue, which is the third most common cancer worldwide in 2021 and the second leading cause of cancer-related deaths [ 1 , 2 , 3 , 4 , 5 ] . The key to preventing CRC is the early detection and removal of adenomatous polyps. Colonoscopies are known as the gold standard for detecting polyps [ 6 , 7 , 8 ] , but the procedure’s effectiveness depends heavily on the experience of the endoscopist. Even experienced endoscopists miss about 12% of large adenoma polyps (at least 1 cm) [ 9 ] , and missed polyp rates can vary between 15% to 35%, depending on the size of the polyp [ 10 ] . Deep learning (DL) object detectors are capable of locating and classifying objects of interest [ 11 ] . There are two types of DL object detection algorithms: Single-stage detectors, such as YOLO (You Only Look Once) [ 12 ] and its variants and two-stage detectors. such as Faster R-CNN [ 13 ] and its variants. Single-stage detectors are useful for real-time applications like polyp detection during colonoscopy due to their speed and efficiency. These detectors perform object localization and classification in one step, allowing them to process images faster. However, this could come at the cost of accuracy, especially when detecting small, flat, or occluded polyps that may not be well-defined. Single-stage detectors can also suffer from higher false positive rates in challenging conditions, such as reflections or poor lighting, which are common in medical imaging [ 8 ] . Two-stage detectors offer higher accuracy by separating region proposal from classification. This makes them known for more complex tasks where fine-grained detection is necessary. However, two-stage detectors are slower due to their sequential processing, making them harder to use for real-time applications. They are also computationally more expensive, which limits their practicality in scenarios that require immediate feedback during procedures. Although advancements like spatial pyramid pooling have improved their speed, they still lag behind single-stage models in real-time performance [ 14 ] . 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. Leveraging DL detectors for polyps detection may aid endoscopists in reducing the missed polyp rate. Many researchers have explored variations of YOLO for polyp detection. ResYOLO [ 15 ] incorporated residual connections and efficient convolution operators for polyp detection, achieving a precision of 88.6% and a recall of 71.6%. [ 16 ] utilized YOLOv3 [ 17 ] and an object tracking algorithm to reduce false positives by filtering out bounding boxes. Although effective, they reported lower predictive performance on flat polyps. [ 8 ] proposed a robust real-time DL polyp detection system, using data augmentations, such as noise invocation, flipping, rotation, brightness, and contrast adjustment. [ 18 ] fine-tuned YOLOv5 variants for improving the detection of smaller polyps. Many problems exist when attempting to detect polyps, such as a polyp being small or flat. Another common challenge is that polyp detectors tend to confuse the white light specular reflections produced by the endoscope with polyps (see Fig. 1 ). However, this challenge is a less explored topic. To address this challenge, in this paper, we propose a new data augmentation called White Light Specular Reflection (WLSR), for DL polyp detection. The purpose of this data augmentation is to generate more specular reflections and add them to the training images to make training harder. 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. Our contributions are as follows: (1) We develop a new data augmentation for DL polyp detection. (2) We demonstrate the effectiveness of our new method on the Harvard Dataverse dataset [ 19 ] . Figure 2: The overview of our WLSR data augmentation for DL polyp detection. 2 Methodology Fig. 2 shows the pipeline of our new data augmentation method: First, we generate a bank of artificial lights using the training dataset (Section 2.1 ). Then for input training images, we identify the regions in which we cannot place these lights (Section 2.2 ). Next, we place these artificial lights to the areas that fit to form the augmented images. Finally, the augmented training images are used to train the DL detector (e.g., YOLOv5 [ 18 ] ) for polyp detection. 2.1 Generating Bank of Lights Simply placing white circles or ovals may not be beneficial for training since we need to mimic the real thing. Therefore, a crucial question is: How can we generate the artificial lights that look like real white light specular reflections that occur during colonoscopies? To address this challenge, we propose to generate a bank of lights, consisting of the following steps: (1) We identify regions of the real white light specular reflections of the training images. (2) We generate various shades of white light on these regions by adjusting the intensity levels of the regions and turn regions that are not real white light specular reflections to black. Different real white light specular reflections regions have different grayscale values (from dark gray to bright white). This can simulate a more realistic