ERCPMP-Gx: Endoscopic Image and Video Dataset for Morphological, Histopathological, and Genomic Characterization of Colorectal Polyposis
ERCPMP-Gx dataset integrates endoscopic images, histopathology, and genomics for colorectal polyposis AI research.
Key Findings
Methodology
The study developed the ERCPMP-Gx dataset, integrating endoscopic images, histopathology, and genomic data. Using the Olympus EVIS X1 system, 160 images and videos were collected, covering white-light endoscopy (WLE), narrow-band imaging (NBI), and more. Approximately 80% of cases are hereditary polyposis, providing an AI-ready patient-level annotation framework.
Key Results
- Result 1: The dataset includes 160 images and videos, 80% are hereditary polyposis, providing an AI-ready annotation framework.
- Result 2: First to achieve joint modeling of endoscopic phenotype and genotype compared to existing datasets.
- Result 3: The multimodal dataset supports various AI tasks like polyp detection and classification.
Significance
This dataset opens new possibilities for AI research in colorectal polyposis, especially in the identification and classification of hereditary polyposis. By integrating multimodal data, the study aims to improve diagnostic accuracy and support personalized patient management.
Technical Contribution
ERCPMP-Gx's technical contribution lies in the first-time integration of endoscopic images, histopathology, and genomic data into an AI-ready framework. This multimodal integration provides more comprehensive training data for AI models, potentially improving model accuracy and robustness.
Novelty
ERCPMP-Gx is the first publicly available dataset to jointly model endoscopic phenotype and genotype, overcoming the limitations of traditional datasets focused solely on polyp detection.
Limitations
- Limitation 1: The dataset's small size may affect model generalizability.
- Limitation 2: Imbalance among different polyposis syndromes may introduce training bias.
Future Work
Future research should expand the dataset size and increase multicenter collaborations to improve model generalizability and clinical applicability.
AI Executive Summary
The ERCPMP-Gx dataset provides a new tool for AI research in colorectal polyposis. Traditional endoscopic datasets often focus on individual polyps, while ERCPMP-Gx uniquely integrates endoscopic images, histopathology, and genomic information, supporting the development of multimodal AI models.
The dataset includes 160 images and videos collected using the Olympus EVIS X1 system, covering various modes such as white-light endoscopy and narrow-band imaging. Approximately 80% of cases are hereditary polyposis, offering an AI-ready patient-level annotation framework. This multimodal integration provides more comprehensive training data for AI models, potentially improving model accuracy and robustness.
Despite the dataset's small size, which may affect model generalizability, its potential in the identification and classification of hereditary polyposis is significant. Future research should focus on expanding the dataset size and increasing multicenter collaborations to enhance model generalizability and clinical applicability.
Deep Analysis
Background
Colorectal cancer is a leading cause of cancer-related deaths worldwide, with hereditary polyposis syndromes accounting for about 1% of all cases. Early diagnosis and precise classification are crucial for initiating targeted surveillance programs. AI has shown increasing promise in colorectal cancer diagnosis and management, but its application in hereditary polyposis remains limited.
Core Problem
Existing endoscopic datasets focus primarily on individual polyp detection, lacking resources that combine polyp phenotype with histopathological and genomic information. This limits AI's application in identifying hereditary polyposis.
Innovation
ERCPMP-Gx's innovation lies in the first-time integration of endoscopic images, histopathology, and genomic data into an AI-ready framework. This multimodal integration allows for a more comprehensive analysis and classification of hereditary polyposis.
Methodology
- �� Collect endoscopic images and videos using the Olympus EVIS X1 system.
- �� Integrate histopathology and genomic information.
- �� Provide an AI-ready patient-level annotation framework.
Experiments
The experimental design includes 160 images and videos collected using the Olympus EVIS X1 system, covering modes like white-light endoscopy and narrow-band imaging. Approximately 80% of cases are hereditary polyposis, offering an AI-ready annotation framework.
Results
The dataset includes 160 images and videos, 80% are hereditary polyposis, providing an AI-ready annotation framework. It is the first to achieve joint modeling of endoscopic phenotype and genotype compared to existing datasets.
Applications
The dataset can be used to develop various AI tasks, such as polyp detection and classification, supporting personalized patient management.
Limitations & Outlook
The dataset's small size may affect model generalizability. Imbalance among different polyposis syndromes may introduce training bias.
Plain Language Accessible to non-experts
Imagine a factory where the raw materials are endoscopic images, histopathology, and genomic information. ERCPMP-Gx is like a smart factory that integrates these materials to produce more accurate diagnostic tools. Traditional factories handle only one type of material, but this smart factory can process multiple materials simultaneously, producing higher-quality products.
ELI14 Explained like you're 14
Imagine you're playing a super complex puzzle game. Each piece of the puzzle represents different information about a patient, like endoscopic images, histopathology, and genomic data. ERCPMP-Gx is like a super helper that helps you quickly and accurately put these puzzle pieces together to find the cause of the disease. Isn't that cool?
Glossary
Endoscopic Image
Images obtained through endoscopic equipment to diagnose and treat internal conditions.
Used to identify and classify colorectal polyps.
Histopathology
The study of tissue structure and function through microscopic examination.
Used to confirm the pathological features of polyps.
Genomic Information
An individual's complete genetic information, including all genes and non-coding sequences.
Used to identify hereditary polyposis.
Multimodal Data
A dataset combining multiple data types like images, text, and audio for comprehensive analysis.
ERCPMP-Gx combines image, pathology, and genomic data.
AI-ready Framework
A data structure prepared for training and testing AI models.
ERCPMP-Gx provides an AI-ready patient-level annotation framework.
Open Questions Unanswered questions from this research
- 1 How to validate model generalizability on larger datasets?
- 2 How to address sample imbalance among different polyposis syndromes?
Applications
Immediate Applications
Clinical Diagnosis
Doctors can use the multimodal information in this dataset to improve the diagnostic accuracy of colorectal polyposis.
Long-term Vision
Personalized Medicine
Integrating multimodal data to provide personalized patient management and treatment plans.
Abstract
Hereditary polyposis syndromes can be precursor lesions to colorectal cancer and are associated with a broad spectrum of extracolonic tumors. Early identification and accurate classification of these syndromes are essential for timely diagnosis, individualized patient management, and targeted surveillance strategies for affected families. However, public endoscopic datasets are largely organized around the individual sporadic polyp, and none links the polyposis phenotype to histopathology and germline findings at the patient level. Here, we present ERCPMP-Gx, an endoscopic, histopathological, and genomic dataset developed to support the application of artificial intelligence (AI) in the recognition, characterization, and classification of colorectal polyposis. Most procedures were performed using the Olympus EVIS X1 system with white-light endoscopy (WLE), narrow-band imaging (NBI), magnifying NBI (M-NBI), and NBI with near focus modes, yielding 160 images and accompanying video clips. Approximately eighty percent of cases represent clinically and/or genetically confirmed hereditary polyposis syndromes (PG), including familial adenomatous polyposis (FAP), Peutz-Jeghers syndrome (PJS), juvenile polyposis syndrome (JPS), and ganglioneuroma syndrome (GNS), while the remaining twenty percent comprise non-hereditary polyps and polyp-mimicking lesions with overlapping morphological features (Non-PG), included to support differential classification. Each released record is linked, where available, to standardized endoscopic annotations, representative histopathology, and clinically reported germline findings, forming an AI-ready, patient-level annotation framework. The dataset is publicly accessible at Mendeley (https://doi.org/10.17632/nzyfc544bx.2). For the latest updates and further information, readers are referred to the DataBioX website: https://databiox.com.