[Paper Review] CholecSeg8k: A Semantic Segmentation Dataset for Laparoscopic Cholecystectomy Based on Cholec80

Wei Hong, C.-L. Kao+4 more|arXiv (Cornell University)|Dec 23, 2020

Colorectal Cancer Screening and Detection3 references41 citations

TL;DR

CholecSeg8K is an open semantic segmentation dataset derived from Cholec80, containing 8,080 annotated frames across 13 classes for laparoscopic cholecystectomy, released under CC BY-NC-SA 4.0. It provides pixel-level masks and multiple mask formats to support training of segmentation models in computer-assisted surgery.

ABSTRACT

Computer-assisted surgery has been developed to enhance surgery correctness and safety. However, researchers and engineers suffer from limited annotated data to develop and train better algorithms. Consequently, the development of fundamental algorithms such as Simultaneous Localization and Mapping (SLAM) is limited. This article elaborates on the efforts of preparing the dataset for semantic segmentation, which is the foundation of many computer-assisted surgery mechanisms. Based on the Cholec80 dataset [3], we extracted 8,080 laparoscopic cholecystectomy image frames from 17 video clips in Cholec80 and annotated the images. The dataset is named CholecSeg8K and its total size is 3GB. Each of these images is annotated at pixel-level for thirteen classes, which are commonly founded in laparoscopic cholecystectomy surgery. CholecSeg8k is released under the license CC BY- NC-SA 4.0.

Motivation & Objective

• Provide a pixel-level semantic segmentation dataset for endoscopic surgery to support algorithm development (e.g., SLAM, segmentation models).
• Extend Cholec80 by annotating frames with13 surgery-specific classes to enable contextual understanding of laparoscopic cholecystectomy imagery.
• Offer ready-to-use data formats (raw images, color masks, annotation masks, watershed masks) to facilitate training across models.

Proposed method

• Extract 8,080 frames from 17 videos in the Cholec80 dataset.
• Annotate pixels with 13 predefined classes relevant to cholecystectomy.
• Produce three mask formats per image (color mask, annotation mask, watershed mask).
• Organize data into a two-level directory structure by video and frame indices.
• Provide data in PNG format under CC BY-NC-SA 4.0 license.

Experimental results

Research questions

• RQ1 How can a high-quality pixel-wise semantic segmentation dataset be created from existing endoscopic video data?
• RQ2 What is the class taxonomy and its coverage for tissues and instruments in laparoscopic cholecystectomy?
• RQ3 What are the practical data formats and organizational schemes that maximize usability for segmentation model development?
• RQ4 How does class distribution look across the annotated frames and what implications does it have for training?

Key findings

• The dataset comprises 8,080 frames extracted from 17 Cholec80 videos.
• Thirteen pixel-level annotation classes are defined, including both organs (e.g., liver, gallbladder) and instruments (e.g., grasper, L-hook electrocautery).
• Image resolution is 854 × 480 pixels and total dataset size is 3 GB.
• Not all frames contain all 13 classes; class presence is imbalanced with some tools under 1% of annotated pixels.
• Each frame has three masks (color mask, annotation mask, watershed mask) in PNG format.
• The dataset is released under CC BY-NC-SA 4.0 and available on Kaggle.