Discovering Discriminative and Interpretable Patterns for Surgical Motion Analysis
This is the companion web page for our paper titled "Discovering Discriminative and Interpretable Patterns for Surgical Motion Analysis".
This paper has been accepted for publication at the Conference on Artificial Intelligence in Medicine AIME 2017.
If you have any comments/questions regarding the research work or software, please feel free to contact us.
The source code
The software is developed in Java, it uses on JMotif SAX library for the SAX convertion of kinematic data. This package is not included in the source file and should be downloaded separately. You will also need the JIGSAWS dataset to re-run the experiments of the paper. The source code of our method can be downloaded here : donwload the soure code.
The archive contains multiple Java classes (as an Eclipse project) that can be run to obtain the results presented in the paper. The Aime2017.java file contains a simple main() method to launch the experiments. The header of the files have to be modified to point to a folder with JIGSAWS data and to choose the SAX parameters. The best parameters presented in the paper are available in the source code.
Video illustrating the method
Video showing side-by-side the trial recording and the on-going trajectory with heatmap coloring.
The video is encoded using h.264 codec and mp4 containner : Suturing_B005.mp4.
Interpretable visualization of discriminative patterns
The proposed method can be used to highlight the subsequences which are the most distinctive of the skill level.
In the experiments section of the paper, we present this visualization for two Suturing trials, one from an Expert (E) and one from a Novice (B).
The colors are obtained using the tf*idf weight vector of a given class which is projected on the (x,y,z) coordinates of the master right hand.
You can find the other visualizations for all the subjects and all the trials for the Suturing task here : Visualisation for Suturing Task.
The visualizations were obtained using R with the rgl package.
Trial 5 of Suturing task of subject B (novice) using Novice class tf*idf vector weights of 5th fold.
Trial 5 of Suturing task of subject E (expert) using Expert class tf*idf vector weights of 5th fold.
Trial 5 of Suturing task of subject C (intermediate) using Intermediate class tf*idf vector weights of 5th fold.
Last update : Januray 2017