Enhancing Localization and Orientation of Dental Tools Using Isolation and Deep Learning Algorithms: A Case Study Utilizing YOLOv5, GrabCut, and PCA

Abstract

Dental instrument retrieval during surgeries demands precise identification and localization to select the correct tool per the surgeon's requirements. Challenges with human dental assistants include tool misidentification, accidental contact with sharp tips, and potential infection exposure. This research proposes practical solutions for detecting, categorizing, isolating, and localizing specific dental instruments based on the dentist's needs.

We explore the application of the YOLOv5 deep learning algorithm for tool detection and classification, using predefined classes to determine Bounding Boxes (BBs). The Grabcut algorithm is then used to isolate the tool and create a foreground mask. Next, Principal Component Analysis (PCA) is employed for precise localization of the detected instruments.

Our approach leverages deep learning algorithms to accelerate the detection and classification of dental tools, integrating them with background isolation algorithms to obtain an ideal tool mask. Additionally, we determine the tool's orientation using eigenvectors to obtain the general orientation via PCA. The proposed model aims to ensure the lowest position error and minimum calculation time, even as the environment changes.