Progrès de la recherche en sciences appliquées Libre accès

Abstrait

EMTLAB: A Toolbox for the Analysis of Electromagnetic Tracking Data in Brachytherapy

Goetz ThI, Herzberger F, Tome AM, Hensel B and Lang EW

Background: High dose rate brachytherapy (HDR-BT) of female breast cancer patients relies on electromagnetic tracking (EMT) for localizing the prescribed dwell positions of the radiation source. A collection of machine learning techniques like Particle Filtering (PF), Singular Spectrum Analysis (SSA), Ensemble and Multivariate Empirical Mode Decomposition (EEMD/MEMD) represent powerful signal processing techniques and are employed in this study to achieve this goal. Information-theoretic similarity measures allow comparing extracted signal components for artifact identification and elimination.
New toolbox: We present a new toolbox, called EMTLAB, which is designed as an extensible toolbox for electromagnetic tracking data analysis. It contains all machine learning techniques mentioned above and is written in MATLAB®.
Results: EMTLAB offers the practitioner a convenient way to easily and efficiently perform particle filtering, signal decomposition and manual or automatic artifact removal with an SSA, an EEMD or MEMD in combination with three similarity measures: Pearson correlation, Jensen-Shannon divergence or Kull back-Leibler divergence. As demonstrated with illustrative examples, EMTLAB offers a complete and almost fully automatic signal processing chain for an analysis of EMT data sets collected during a HDR-BT. In addition, EMTLAB represents a user-friendly graphical user interface (GUI), which also provides convenient means to visualize the results in illustrative graphs. A number of screen shots helps in understanding the functioning of the signal processing chain and the use of the GUI.
Conclusion: EMTLAB is a reliable, efficient and automated solution for processing and analyzing EMT sensor data from a HDR-BT, while employing different physical models of system dynamics. This sensor tracking by particle filtering allows to adapt the analysis to different dynamical models and the SSA and EMD algorithms provide an easy means to remove from the data artifacts stemming from breathing modes or measurement device malfunctioning.