Page 116 - Demo

Chapter 7114Materials and MethodsData collectionAn ex-vivo measurement campaign was designed to collect relevant data. Seven fresh frozen cadavers were obtained from the clinical anatomy and embryology section of the department of medical biology of the Amsterdam university medical center (Amsterdam UMC). The donation process was in accordance with Dutch legislation and the regulations of the medical ethical committee of the Amsterdam UMC. Extractions were performed by three senior oral and maxillofacial surgeons. An extensive measurement setup was used, as described in more detail in previous work [7]. An overview of the setup is presented in Figure 1. In short, data on position, orientation and movements was gained through a compliant robot arm (KUKA LBR iiwa 7 R800) passively following the movements of an OMF-surgeon (frequency 100Hz). A sixaxis force/torque (FT) sensor (ATI industrial automation 16-bit Delta transducer) was used to register forces and torques at 20Hz. An open-source framework was used for integration of the components (Robot Operating System, ROS). A custom graphical user interface (GUI) was designed to allow for the addition of metadata on the experiments itself (e.g., reason in case of any failed measurements) and on the clinical status of the teeth (e.g., restorative and periodontal state). In total, the setup gathers thirteendimensional time series for each individual tooth removal procedure. Six-dimensional time series from the force/torque sensor, consisting of three dimensions (‘XYZ’) for both forces and torques. A further seven-dimensional time series are gathered from the robot arm - three dimensions for the position of the end-effector (‘XYZ’) and a fourdimensional representation of the orientation of the end-effector in quaternions [8]. For data analysis, Python was used (Python Software Foundation. Python Language Reference, version 3.9) [9] and the Scikit-learn 1.0.1 module [10]. A calibration step was performed just prior to each experiment, to determine the position and orientation of each tooth [7]. Reporting guidelines were used to structure this report [11, 12].Preprocessing the dataBecause each measurement started and stopped manually, some meaningless data was gathered just prior and after each experiment. Raw data was therefore manually trimmed, using the custom user interface, directly after each experiment. Using data from the calibration step, raw data from the force/torque sensor and robot arm were mathematically transformed from their own reference frames to the clinically relevant tooth frame [7]. It results in one unified reference frame in which, for example, a Tom van Riet.indd 114 26-10-2023 11:59