Surgical Procedure Understanding, Evaluation, and Interpretation: A Dictionary Factorization Approach

  • Author(s) / Creator(s)
  • In this study, we present a novel machine learning-based
    technique to help surgical mentors assess surgical motion trajectories
    and corresponding surgical skills levels in surgical training programs.
    The proposed method is a variation of sparse coding and dictionary
    learning that is straightforward to optimize and produces approximate
    trajectory decomposition for structured tasks. Our approach is superior
    to existing stochastic or deep learning-based methods in terms
    of transparency of the model and interpretability of the results. We
    introduce a dual-sparse coding algorithm which encourages the elimination
    of redundant and unnecessary atoms and targets to reach the
    most informative dictionary, representing themost important temporal
    variations within a given surgical trajectory. Since surgical tool trajectories
    are time series signals, we further incorporate the idea of floating
    atoms along the temporal axis in trajectory analysis, which improves
    the model’s accuracy and prevents information loss in downstream
    tasks.Using JIGSAWS data set,we present preliminary results showing
    the feasibility of the proposed method for clustering and interpreting
    surgical trajectories in terms of user’s skills-related behaviors.

  • Date created
  • Subjects / Keywords
  • Type of Item
    Article (Published)
  • DOI
  • License
    Attribution-NonCommercial 4.0 International