RRTS Dataset: A Benchmark Colonoscopy Dataset from Resource-Limited Settings for Computer-Aided Diagnosis Research Journal: Computer Methods and Programs in Biomedicine
Ridoy Chandra Shil, Ragib Abid, Tasnia Binte Mamun, Samiul Based Shuvo, Dr. Masfique Ahmed Bhuiyan, Jahid Ferdous · 2025
Read RRTS Dataset: A Benchmark Colonoscopy Dataset from Resource-Limited Settings for Computer-Aided Diagnosis Research Journal: Computer Methods and Programs in Biomedicine by Ridoy Chandra Shil, Ragib Abid, Tasnia Binte Mamun, Samiul Based Shuvo, Dr. Masfique Ahmed Bhuiyan, Jahid Ferdous (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-2511-06769.txt.
Introduction
Ridoy Chandra Shil and colleagues release a colonoscopy image dataset gathered in a resource-limited clinical setting. Finding polyps early prevents colorectal cancer, and computer-aided diagnosis needs training data that looks like real practice. Popular public sets are small, polished, or free of the artifacts that crowd everyday endoscopy suites.
The dataset holds nearly thirteen hundred polyp images from one hundred sixty-four patients, each with an expert-checked mask, plus more polyp-free images. Motion blur, glare, stool, blood, and dim frames all appear as they do in routine work. Baseline classifiers reached about ninety-one percent accuracy and the best segmentation scored moderately, trailing curated sets precisely because the images are harder.
Catalog record
| CREATOR | Ridoy Chandra Shil, Ragib Abid, Tasnia Binte Mamun, Samiul Based Shuvo, Dr. Masfique Ahmed Bhuiyan, Jahid Ferdous |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-06769
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
RRTS Dataset: A Benchmark Colonoscopy Dataset from Resource-Limited Settings for Computer-Aided Diagnosis Research Journal: Computer Methods and Programs in Biomedicine Ridoy Chandra Shil, Ragib Abid, Tasnia Binte Mamun, Samiul Based Shuvo, Dr. Masfique Ahmed Bhuiyan, Jahid Ferdous Abstract Background and Objective: Colorectal cancer prevention relies on early detection of polyps during colonoscopy. Existing public datasets, such as CVC-ClinicDB and Kvasir-SEG, provide valuable benchmarks but are limited by small sample sizes, curated image selection, or lack of real-world artifacts. There remains a need for datasets that capture the complexity of clinical practice, particularly in resource-constrained settings. Methods: We introduce a dataset, BUET Polyp Dataset (BPD), of colonoscopy images collected using Olympus 170 and Pentax i-Scan series endoscopes under routine clinical conditions. The dataset contains images with corresponding expert-annotated binary masks, reflecting diverse challenges such as motion blur, specular highlights, stool artifacts, blood, and low-light frames. Annotations were manually reviewed by clinical experts to ensure quality. To demonstrate baseline performance, we provide benchmark results for classification using VGG16, ResNet50, and InceptionV3, and for segmentation using UNet variants with VGG16, ResNet34, and InceptionV4 backbones. Results: The dataset comprises 1,288 images with polyps from 164 patients with corresponding ground-truth masks and 1,657 polyp-free images from 31 patients. Benchmarking experiments achieved up to 90.8% accuracy for binary classification (VGG16) and a maximum Dice score of 0.64 with InceptionV4-UNet for segmentation. Performance was lower compared to curated datasets, reflecting the real-world difficulty of images with artifacts and variable quality. Conclusions: This dataset provides a representative and challenging resource for developing robust computer-aided diagnosis systems in colonoscopy. By including diverse real-world imaging artifacts, it complements existing curated benchmarks and enables the development of models that generalize better to clinical practice. Baseline results are intended to guide future research and facilitate fair comparisons. Please mail us to get the access of the dataset. \keepXColumns 1 Introduction Colorectal cancer (CRC) is the third most common cancer globally, with approximately 1.9 million cases and over 900,000 deaths reported in 2020. Alarmingly, these numbers are expected to escalate, reaching 3.2 million new cases and 1.6 million deaths by 2040 [ 1 ] . Early diagnosis plays a crucial role in survival outcomes, with a 5-year survival rate of 91.5% for Stage I CRC, dropping drastically to 16.2% for late-stage diagnoses [ 2 ] . Most CRC cases develop from precancerous polyps through the adenoma–carcinoma sequence. These statistics highlight the importance of early and accurate polyp detection. It is a key step in the prevention and timely treatment of CRC. Colonoscopy is widely regarded as the gold standard for CRC screening and polyp removal. It enables direct visualization of the entire colon and allows simultaneous biopsy and resection of suspicious lesions. However, despite its advantages, colonoscopy is not without limitations. Studies report that polyps can be missed in 9% to 34% of cases, with flat or sessile morphologies being particularly prone to oversight [ 3 ] . The situation is even more critical in low-resource settings, where healthcare facilities often rely on older-generation equipment, limited technical expertise, and suboptimal imaging conditions. Hence, polyp detection accuracy is compromised. Figure 1 : Examples of colorectal polyp morphologies observed during colonoscopy. (a) Sessile polyp lying flat against the colonic mucosa, making detection more challenging.(b, c) Pedunculated polyps with stalk-like structures protruding from the mucosal surface. Recent advancements in artificial intelligence (AI), particularly deep learning–based computer-aided detection (CADe) systems, have shown promise in improving polyp detection rates and reducing inter-observer variability [ 4 ] . These systems rely heavily on the availability of high-quality, annotated datasets that capture the visual diversity and complexity of real-world colonoscopy images. While several public datasets such as Kvasir-SEG [ 5 ] , CVC-ClinicDB [ 6 ] , ETIS-Larib [ 7 ] , and PolypGen [ 8 ] have contributed significantly to this domain, they suffer from several limitations. Many are curated from high-resource clinical environments, often using high-definition scopes. Images are pre-selected, typically excluding low-quality, blurry, or transitional frames. Dataset annotations may lack pixel-level precision, contain limited polyp morphology, or exclude non-polyp images. Several datasets are not publicly available, lack recommended data splits, or are highly imbalanced. To address these limitations, we present the BUET Polyp Dataset (BPD), a benchmark dataset comprising 2,945 colonoscopy images collected at a resource-constrained public hospital, Dhaka Medical College and Hospital in Bangladesh. Unlike existing datasets, BPD reflects the real-world clinical constraints encountered in under-resourced settings. It includes both polyp and non-polyp frames, and retains imperfections such as motion blur, glare, and variable illumination, which are typically excluded from curated datasets. Key contributions of BUET Polyp Dataset (BPD) dataset include: 1. Low-resource, real-world context: Images are collected in a resource-limited public hospital, capturing challenges typical of such environments. 2. Segmentation-ready masks: Each polyp image includes expert-verified, pixel-level binary masks, making the dataset suitable for training and evaluation of segmentation models. 3. Polyp and non-polyp cases: Inclusion of negative samples enables robust training for binary classification tasks (polyp vs. non-polyp). 4. Generalization benchmark: BPD serves as a stress test for AI models, highlighting performance gaps between curated and real-world environments. The dataset enables research in preprocessing, contrast enhancement, and artifact suppression algorithms designed for noisy, real-world inputs. The remainder of this paper is structured as follows: Section 2 provides the necessary background on colorectal cancer and polyps, highlighting their clinical significance. Section 3 reviews publicly available polyp datasets, emphasizing their characteristics and limitations. Section 4 describes the study design and data acquisition process, including subject selection, instrumentation, annotation strategy, and ethical considerations. Section 5 presents demographic and categorical analyses of the dataset. Section 6 details the overall data distribution and provides feature-level visualizations. Section 7 discusses challenges encountered during data preparation. Section 8 highlights the dataset’s impact and usability in real-world computer-aided diagnosis. Section 9 presents benchmarking and evaluation results for both segmentation and classification models. Section 10 provides a broader discussion of findings, followed by Section 11 on limitations, and finally, Section 12 concludes the paper with future directions. 2 Background 2.1 Colorectal Cancer and Polyps CRC typically originates from the benign overgrowth of mucosal epithelial cells. These lesions, termed polyps, may persist and enlarge slowly over a period of 10–20 years before undergoing malignant transformation [ 9 ] . The most frequent type is the adenomatous polyp, which arises from glandular cells responsible for secreting mucus within the large intestine [ 10 ] . Although only around 10% of adenomas ultimately develop into invasive carcinoma, the probability of malignant change increases with polyp size as shown in Fig. 2 . When such polyps progress to invasive disease, the resulting malignancy is termed adenocarcinoma, which accounts for approximately 96% of all CRC cases [ 11 ] . Figure 2 : Progression of colorectal cancer from benign adenomatous polyps to malignant invasive cancer [ 12 ] . Table 1 gives an overview of the five most common types of polyps, their cancer risks, and treatment. Table 1 : Comparison of Common Polyp Types Based on Prevalence, Cancer Risk, and Treatment Type of polyp Prevalence Cancer Risk Treatment Inflammatory Frequently observed in patients with chronic inflammatory conditions like Crohn’s disease or ulcerative colitis Generally non-cancerous; low malignant potential Typically removed during colonoscopy Hyperplastic Commonly appear near the