GynSurg: Gynecology Laparoscopic Surgery Dataset

GynSurg is a comprehensive multi-task dataset for gynecologic laparoscopic surgery, designed to support a wide range of video-based surgical analysis tasks. GynSurg integrates high-resolution video, dense temporal annotations for surgical actions and side effects, and pixel-level segmentation masks for surgical instruments and anatomical structures. This unified dataset provides a foundation for advancing research in surgical action recognition, instrument and anatomical segmentation, intraoperative event detection, and workflow understanding. The structure and contents of each dataset subset are detailed in the following subsections.

Description

Action Recognition Dataset

The action recognition component of GynSurg consists of 152 gynecologic laparoscopic surgery videos, selected from over 600 recorded procedures at the Medical University of Vienna and Medical University of Toronto. All videos were captured at 30 frames per second (fps) with a resolution of 1920*1080 pixels. Each video was meticulously annotated by clinical experts for four key operative actions: coagulation, needle passing, suction/irrigation, and transection. Additionally, a dedicated subset of the data was annotated for two intraoperative side effects: bleeding and smoke, supporting research on complex action and event recognition in laparoscopic surgery.

Dataset Composition

Purposes

Overview

Table 1 summarizes the dataset composition across procedural actions and side effect labels. In addition to the primary laparoscopic actions, a “Rest” category is defined to represent intervals with no instrument activity or irrelevant motion. Side effects are annotated as binary classification tasks (bleeding vs. non-bleeding, smoke vs. non-smoke). The “Segments” column reports the number of extracted video segments per class, while the “Total Duration” column gives the cumulative length of all segments. To standardize the data for model development, video segments are further partitioned into three-second clips (with a one-second overlap), with the resulting clip counts also reported. Figure 1 presents example frames illustrating each surgical action and side effect.

Timeline visualization actions

Semantic Segmentation Dataset

The semantic segmentation component of GynSurg comprises 15 laparoscopic hysterectomy videos obtained from the Medical University of Toronto. Of these, ten videos were annotated for surgical instruments and five for anatomical structures. All annotations were performed by clinical experts following standardized guidelines to ensure high-quality, pixel-level precision.

Instrument Segmentation

The instrument segmentation dataset includes 10 annotated videos, totaling 11,352 frames at a resolution of 750*480 pixels. A total of 21 distinct instrument classes were labeled. Figure 2 (left) shows the distribution of instrument annotations per frame. Multiple instruments often appear within the same frame, with graspers being the most frequently duplicated. For analysis, the instruments are categorized into two functional groups:

Representative annotated frames illustrating both instrument and anatomy segmentation overlays are shown in Figure 3.

Name 1
Grasper
forcep
Bipolar forcep
scissor
Scissor
suturing
Suturing instrument
irigator_grasper
Irrigator
hook
Hook
thread
Thread
cannula
Cannula
trocar-sleeve
Trocar-sleeve
morcellator
Morcellator
uterus and tube
Uterus and Tube
uterus_ovary
Uterus and Ovary

Anatomy segmentation

The anatomical segmentation dataset comprises five laparoscopic hysterectomy videos, yielding 1,010 labeled frames at a resolution of 750*480 pixels. Four anatomical structures were annotated: "uterus", "ovary", "fallopian tube", and a general "organ" category. For quantitative evaluation, only the uterus, ovary, and tube classes are retained due to their frequent occurrence and clinical significance. Figure2 (right) displays the per-frame distribution of anatomical labels, with sample annotated images presented in the rightmost panels of Figure 3.

Unannotated videos

In addition to the annotated datasets, GynSurg includes a collection of 75 unannotated high-definition (1920*1080) laparoscopic hysterectomy videos (LHE), also acquired at the University of Toronto. These recordings are provided to support research in semi-supervised and self-supervised learning for surgical video analysis.

Disclaimer

The dataset is exclusively provided for scientific research purposes and as such cannot be used commercially or for any other purpose. If any other purpose is intended, you may directly contact the originator of the videos, Prof. Dr. Heinrich Husslein or Assoc. Prof. DI Dr. Klaus Schoeffmann.

In addition, reference must be made to the following publication when this dataset is used in any academic and research reports:

BibTeX: 

@inproceedings{
    author    = {Sahar Nasirihaghighi and
                Negin Ghamsarian and
                Leonie Peschek and
                Matteo Munari and 
                Heinrich Husslein and
                Raphael Sznitman and
                Klaus Schoeffmann},
    title     = {{GynSurg: A Comprehensive Gynecology Laparoscopic Surgery Dataset},
    booktitle = {},
    series    = {},
    volume    = {},
    pages     = {},
    publisher = {},
    year      = {2025},
    url       = {},
    doi       = {},
    timestamp = {},
    biburl    = {},
    bibsource = {}
}

ENID is licensed under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0, Creative Commons License) and is created as well as maintained by Distributed Multimedia Systems Group of the Institute of Information Technology (ITEC) at Alpen-Adria Universität in Klagenfurt, Austria.

This license allows users of this dataset to copy, distribute and transmit the work under the following conditions:

For further legal details, please read the complete license terms.

Download

If you agree to above conditions, you are free to download: