Important Note

This dataset has been completely included in the M2CAI 2016 "m2cai-workflow" dataset, available at http://camma.u-strasbg.fr/datasets. This dataset is no longer available separately. If you obtained it through other channels, you should not use it in conjunction with the M2CAI 2016 dataset, as you will have duplicate data, which can bias or even invalidate your experiments!

Data Description

This sub-challenge focuses on the task of detecting the surgical workflow from laparoscopic images.

Naturally this yields a strong connection to the task of detecting instruments, and we welcome submissions based on this similarity, though the evaluation will not distinguish between approaches taking instrument data into account and those that don't.

Data will be provided for instruments in laparoscopic cholecystectomies. The data is split into a training and test set (see below).

Training Data

7 videos of recorded laparoscopic interventions (24-25 images/second, runtime 26-56 minutes, mean runtime 41 minutes). For each video file, you will find the corresponding annotation file "video<N>_annotations.txt". This file has a tabulator-separated structure as follows:

Download

This dataset has been completely included in the M2CAI 2016 "m2cai-workflow" dataset, available at http://camma.u-strasbg.fr/datasets. This dataset is no longer available separately. If you obtained it through other channels, you should not use it in conjunction with the M2CAI 2016 dataset, as you will have duplicate data, which can bias or even invalidate your experiments!

Time

The time given for each line in the annotation files starts at 0 with the first frame of the video and corresponds to the video runtime. Please note, that the number of lines do not necessarily correspond to the frames in the video, as they can have different framerates. The format of the timestamp is <Hours>:<Minutes>:<Seconds>.<Seconds fraction>.

Workflow Phases

We defined seven phases for the Laparoscopic cholecystectomy (CHE):

We tried to define the transition between different workflow phases as comprehensive and consistent as possible. In general, an "undefined state" during transitions always is accounted to the last phase, i.e. the new phase only starts once the characteristic instrument or structure appears. The following list should explain the transition characteristics.

1. All preparations in the OR (instruments, laparoscope, disinfection of the patient, etc.) belongs to Phase 1. This phase ends with all trocars inserted and the first instrument entering the abdomen.
2. The preparation phase begins, with the insertion of the first instrument (usually the liver retractor).
3. The clipping and cutting phase begins with the insertion of the clip applicator for the first time. Further preparations along the artery or duct between different clips are still accounted to the clipping phase. Once all clips are placed and arteries and ducts are cut, this phase comes to an end.
4. The dissection phase starts with the dissection of the gallbladder along the liverbed.
5. The retrieval phase starts with the complete detachment of the gallbladder, i.e. when the gallbladder is freely movable and not attached to the liver at any point. This phase ends with the extraction of the gallbladder within the retrieval bag.
6. The hemostasis is characterized by the cauterization in the liver bed to stop bleedings and the irrigation of fluids. This procedure is sometimes performed while the gallbladder is still lying (completely detached) inside the abdomen, i.e. before the retrieval phase is completed. Therefore the hemostasis phase starts with the insertion of the irrigation and suction rod and can therefore “pause” the retrieval phase. Usually this phase will be again performed after the retrieval of the gallbladder within the retrieval bag was completed.
7. The last phase starts with the first occurring removal of an instrument and its trocar. The phase ends with the end of the video.

Instruments

Please note: The instrument labels provided here are not the focus of this challenge, and will not be available for the test data!

We provide binary indication for the usage of instruments in an attempt to make this dataset as useful as possible beyond the scope of this challenge. A zero in the annotation table indicates that the given instrument is not in use, while a one states active usage of the instrument in the patient. We prepared the ten tools used in the interventions we observed, though some may not be used in any of the datasets provided for this challenge.

Test Data

Reference Standard

For each training and testing sequence, the reference is given through a separate text file with synchronized labels, defining the surgical workflow phase of each frame. In order to support instrument-based workflow detection methods, the training data will include binary information about which surgical instrument is in use at any given time. Only the workflow labels are relevant for evaluation of the method, though. Additionally detected instruments are an interesting capability, but not of relevance for this sub-challenge.