Classifying a Sensorimotor Skill of Periodontal Probing

Abstract

Currently available dental simulators provide a wide range of visual, auditory, and haptic cues to play back the pre-recorded skill, however, they do not extract skill descriptors and do not attempt to model the skill. To ensure efficient communication of a sensorimotor skill, a model that captures the skill's main features and provides real-time feedback and guidance based on the user's expertise is desirable. To develop this model, a complex periodontal probing skill can be considered as a composition of primitives, that can be extracted from the recordings of several professionals performing the probing task. This model will be capable of evaluating the user's proficiency level to ensure adaptation and providing corresponding guidance and feedback. We developed a SVM model that characterizes the sensorimotor skill of periodontal probing by detecting the specific region of the tooth being probed. We explore the features affecting the accuracy of the model and provide a reduced feature set capable of capturing the regions with relatively high accuracy. Finally, we consider the problem of periodontal pocket detection. The SVM model trained to detect pockets was able to achieve a recall around 0.68. We discuss challenges associated with pocket detection and propose directions for future work.