IADR Abstract Archives
Artificial Intelligence for Diagnostic Processing of Dental Panoramic Tomograms.
Objectives: The dental panoramic tomogram (DPT) has an estimated frequency of 2.7 million exposures per year within UK primary care dentistry alone. DPTs are complex and clinicians are often unsure how to interpret incidental findings. The objectives of this work are: (1) To appraise the literature that describes artificial intelligence (AI) applications that diagnose disease on DPTs. (2) To develop and implement an AI-based clinical application that assists the diagnosis of disease presenting on DPTs.
Methods: Scopus, Medline and Web of Knowledge were systematically searched to retrieve records that describe the application of AI to diagnose diseases that present on adult DPTs between 2012-2022. 674 DPTs showing biopsy-proven disease and 505 normal controls were anonymised and used to develop a classification model. Performance was assessed with precision, recall and F1-score. A multidisciplinary team of clinicians and computer scientists defined user requirements, system requirements and potential hazards to develop a clinical application that deploys the classification model.
Results: 61 records were included in the review. These records applied AI to the diagnosis of jaw cysts and tumours (12), periodontal disease (9), caries (6), endodontic diseases (6), osteoporosis (5), soft tissue calcification (5), maxillary sinus disease (5), maxillofacial trauma (4), developmental anomalies (3), temporomandibular joint disease (3), osteonecrosis (1), dental implant failure (1) and multiple radiological findings (1). The included records described AI models that applied classification (72%), object detection (48%) and segmentation (21%) to diagnose disease, with some records involving multiple tasks. Our classification model performed with an F1-score of 0.9. Further details of the clinical application will be presented, pending review by our Trust Intellectual Property Team.
Conclusions: AI has the potential to address clinical demand for assistance in the interpretation of DPTs and could have a role in prioritisation of referrals for patients with significant disease.
Methods: Scopus, Medline and Web of Knowledge were systematically searched to retrieve records that describe the application of AI to diagnose diseases that present on adult DPTs between 2012-2022. 674 DPTs showing biopsy-proven disease and 505 normal controls were anonymised and used to develop a classification model. Performance was assessed with precision, recall and F1-score. A multidisciplinary team of clinicians and computer scientists defined user requirements, system requirements and potential hazards to develop a clinical application that deploys the classification model.
Results: 61 records were included in the review. These records applied AI to the diagnosis of jaw cysts and tumours (12), periodontal disease (9), caries (6), endodontic diseases (6), osteoporosis (5), soft tissue calcification (5), maxillary sinus disease (5), maxillofacial trauma (4), developmental anomalies (3), temporomandibular joint disease (3), osteonecrosis (1), dental implant failure (1) and multiple radiological findings (1). The included records described AI models that applied classification (72%), object detection (48%) and segmentation (21%) to diagnose disease, with some records involving multiple tasks. Our classification model performed with an F1-score of 0.9. Further details of the clinical application will be presented, pending review by our Trust Intellectual Property Team.
Conclusions: AI has the potential to address clinical demand for assistance in the interpretation of DPTs and could have a role in prioritisation of referrals for patients with significant disease.