Abstract:Deep learning (DL), an emerging subset of artificial intelligence (AI), use biologically inspired neural networks to extract different features from input data. It is often used for image-related tasks such as automatic detection, segmentation, and classification of anatomical patterns or pathological findings in 2D and 3D images.
Recently, there has been an upsurge in AI research in endodontics, DL models has been used for assessing root morphology, root canal anatomy, case difficulty assessment, detection of pulp and periapical pathosis, detection of root fracture and root resorption.
The aim of this lecture is to describe the practical workflow used to develop a DL model for implementation in Endodontics including;
– Definition of the study design and the DL task.
– Data management (collection/ curation/ de-identification/ definition of the reference standard/ annotation/ splitting).
– Model development (data pre-processing/ model selection/ model training/ hyperparameter tuning/ Hardware and software requirements).
– Model assessment (accuracy metrics/ reporting model outcomes).
– Clinical adoption (deployment/ clinical validation, clinical practice application and monitoring).
– Challenges and limitations.