Autonomous Surgical Lamps
Autonomous Surgical Lamps
As part of the Creative Unit - Intra-Operative Information, we are developing algorithms for the autonomous positioning of surgical lamps in open surgery. These algorithms work solely on the input of one depth camera, which is positioned above the patient during the surgery. The algorithms identify the operation site (aka the situs) and all possible occlusions. They then move the lamps to avoid occlusions and collisions while optimizing for the least amount of movement feasible over time.
The basic idea is to take the point cloud given by the depth camera and render it from the perspective of the situs towards the working space of the lamps above the operating table. Out of this rendering, we directly get the information, which parts of the lamps workspace are occluded and which not. To be able to minimize the movement over time, we also use information about past occlusions and movements to position the lamps in areas, that are most likely to not be occluded in the future. We arranged the algorithms in a pipeline, which takes the depth image of the depth camera as input, analyzes it to find the situs, and at last outputs the current optimal positions for a given set of lamps.
Our pipeline with illustrations of the outputs of some stages (left). A Screenshot of our testing and visualization environment with input data from a real surgery (right).
- Autonomous Surgical Lamps, Jahrestagung der Deutschen Gesellschaft für Computer- und Roboterassistierte Chirurgie (CURAC), 2015. [BibTex]
- Optimized Positioning of Autonomous Surgical Lamps, SPIE Medical Imaging, Orlando, FL, USA, February 11 - 16, 2017. [BibTex]
Videos on Youtube
Video on local TV station
The following downloadable data is a recording of a complete, open, abdominal surgery. It was recorded using a Microsoft Kinect v2, which was mounted directly above the patient, using a costom recording program. The recordings are stored in HDF5-files which can be read using the appropriate HDF5 library and the following C++ class: header, source. They are also compressed using gzip to preserve server space. Each file contains at most 27000 frames, which corresponds to roughly 16-17 minutes of recording.
This data may only be used for scientific purposes. If you publish research which is using this material, cite the above paper. Also, please contact us, we are always interested to hear what others are doing with this data.
- output_2016-02-04_09-34.h5.gz Compressed file size: 17GB
- output_2016-02-04_09-34.h5.gz Compressed file size: 5.4GB
- output_2016-02-04_09-55.h5.gz Compressed file size: 25GB
- output_2016-02-04_10-10.h5.gz Compressed file size: 19GB
- output_2016-02-04_10-26.h5.gz Compressed file size: 2.9GB
- output_2016-02-04_10-29.h5.gz Compressed file size: 9.2GB
- output_2016-02-04_10-44.h5.gz Compressed file size: 380MB
- output_2016-02-04_11-21.h5.gz Compressed file size: 10GB
- output_2016-02-04_11-37.h5.gz Compressed file size: 13GB
- output_2016-02-04_11-52.h5.gz Compressed file size: 16GB
- output_2016-02-04_12-08.h5.gz Compressed file size: 8.7GB
- output_2016-02-04_12-24.h5.gz Compressed file size: 6.5GB
- output_2016-02-04_12-40.h5.gz Compressed file size: 7.8GB
- output_2016-02-04_12-57.h5.gz Compressed file size: 3.6GB
- output_2016-02-04_13-23.h5.gz Compressed file size: 2.5GB
- output_2016-02-04_13-40.h5.gz Compressed file size: 8.9GB
- output_2016-02-04_13-57.h5.gz Compressed file size: 1.6GB
- output_2016-02-04_14-14.h5.gz Compressed file size: 85MB
- output_2016-02-04_14-31.h5.gz Compressed file size: 15GB
- output_2016-02-04_14-48.h5.gz Compressed file size: 12GB
- output_2016-02-04_15-03.h5.gz Compressed file size: 13GB
- output_2016-02-04_15-19.h5.gz Compressed file size: 4.4GB
- output_2016-02-04_15-34.h5.gz Compressed file size: 2.4GB
As this is real-life data, there are long stretches of the recording, where a surgical lamp obstructs most of the kinects field of view. As a rule of thumb, the smaller the compressed file size, the bigger and longer the obstruction of the field of view.
This work was partially supported by the grant Creative Unit - Intra-Operative Information.