Page 115 - Demo

7A multiclass classification model for tooth removal procedurest113IntroductionAulus Cornelius Celsus (c. 25BC-50AD) described tooth removal procedures for the first time in his ‘De Medicina’ with an instruction: ‘it is to be shook; which must be continued till it move easily’ [1]. In modern textbooks, descriptions of these complex procedures have not changed significantly [2]. Being one of the oldest and most commonly performed surgical procedures worldwide, the lack of scientific progress in this field is surprising. Scientific attempts to increase our understanding of these procedures are relatively rare, heterogeneous and mostly focused on extraction forces [3-6]. Analyzing different aspects of tooth removal, especially in clinical situations, requires measurements of subtle movements and high forces in a confined space (intra-orally), which might explain the knowledge gap in this field [7]. Through a collaboration between computer scientists, mechanical engineers, and oral- and maxillofacial (OMF-) surgeons, a setup was designed to measure different aspects of tooth removal procedures [7]. With the use of compliant robotics, data was gathered on (rotational) forces and movements in all their dimensions, directions, in high detail and at a high frequency. Whilst individual parts of data can be explained and understood with traditional statistical methods, analyzing their combination is complex. Machine learning can be particularly useful to understand and analyze complex or large datasets with many variables, in which it has the potential to detect relationships. It can be considered essential to make use of the data as a whole. A classification model is an example of machine learning technology, which consists of an algorithm capable of predicting which tooth was removed, based on a variety of complex data. It could aid in finding which variables are most relevant in tooth removal procedures and to evaluate how procedures differ between certain teeth. This can be of use for, amongst others, the development of evidence-based education material. The goal of this project was to build and validate a first and exploratory classification model for tooth removal based on force, torque and movement data. By evaluating which variable (or ‘feature’) is selected by the algorithm, a unique insight in this ancient procedure is presented. This manuscript describes our methods data collection using robot technology, the feature design process as well as the models’ performance. Tom van Riet.indd 113 26-10-2023 11:59