Table 6 Colorectal cancer and polyp dataset comparison (selected datasets)

From: Deep multimodal fusion of patho-radiomic and clinical data for enhanced survival prediction for colorectal cancer patients

Dataset Name

Modality

Image Type

Number of Images/Frames

Labels/Classes

Image Specifications

Annotation Type

EBHI-SEG

H&E Histopathology (Hematoxylin & Eosin)

Microscopic Histopathology

4456 image + 4456 ground truth segmentation masks

6 classes: Normal, Polyp, Low-grade Intraepithelial Neoplasia, High-grade Intraepithelial Neoplasia, Serrated Adenoma, Adenocarcinoma

224 × 224 pixels, 400× magnification (10× eyepiece, 40× objective)

Pixel-level segmentation masks

REAL-Colon

Endoscopy Video

Full-resolution Colonoscopy Video Frames

2,757,723 video frames from 60 full-procedure colonoscopy videos

132 polyps with bounding-box annotations, histology, size, anatomical location

Native full-resolution, multi-center diverse endoscopes, variable fps

Bounding boxes (350,264 annotations)

Kvasir-SEG

Endoscopy Image

Colonoscopy Video Frames

1000 polyp images + 1000 segmentation masks

Binary: Polyp (foreground) vs. Background, bounding boxes available

Variable resolution, extracted from endoscopy videos

Pixel-level segmentation masks

Back to article page