Accuracy of an AI-clinical Knowledge-based Hybrid System for Detecting Periodontitis in OPG Images

Periodontitis is a major public health problem due to its high prevalence worldwide, substantial socio-economic impacts, and considerable effects on individuals' quality of life. However, periodontitis in the population remains largely undetected. It is crucial to raise awareness about periodontal health and enhance early diagnosis of periodontitis to ensure timely intervention.

The 2018 Classification of Periodontal and Peri-implant Diseases and Conditions defines four stages of periodontitis ranging from the initial stage (stage I) to the advanced stage (stage IV). In stages II-IV, comprehensive treatment procedures are essential otherwise there is a high risk of tooth or even the entire dentition loss. Although clinical examinations are regarded as the gold standard for determining the stage of periodontitis, the process is laborious and time-consuming, demanding highly experienced specialists. Therefore, alternative cost-effective but reliable and valid approaches for differentiating stage II-IV periodontitis diagnosis, particularly in public communities are highly needed.

Orthopantomography (OPG), also known as panoramic radiography, is a non-invasive and low-dose imaging technique that provides a comprehensive view of the maxillofacial region in one procedure. As an extraoral radiograph, it has advantages in capturing the image, especially in cases where patients struggle to open their mouths or exhibit a pronounced gag reflex that hinders the use of intraoral films. Thus, OPG is likely the most frequently taken dental radiograph around the world and may potentially serve as an effective tool for differentiating stage II-IV periodontitis in populations. Recently, several investigations have been carried out to utilize OPG images for periodontitis diagnosis. However, the strategies of these studies rely on the radiographic annotations for specific landmarks by clinicians which may lack compelling accuracy. Furthermore, only the radiographs with high quality could be a valuable adjunct for the periodontitis diagnosis, so many available OPG images with the superimposition of anatomical structures, disproportionate image magnification, distortion, and blur may decrease the generalization of the developed system.

Artificial Intelligence (AI) has emerged as a powerful tool in various fields of medicine, including dentistry. AI-based algorithms, particularly deep learning techniques, have shown remarkable capabilities in image analysis, pattern recognition, and decision-making. In recent years, the integration of AI technology in dentistry has opened new avenues for enhancing the accuracy and efficiency of diagnosis. AI-based algorithms may be able to recognize some features in OPG images that are imperceptible to the human eye, allowing for the detection of subtle bone loss and achieving a more accurate diagnosis of periodontal staging.

Notably, findings from our recent study revealed that a hybrid system combining AI algorithms and clinical knowledge has good performance for differentiating stage II-IV periodontitis. In the development process of this hybrid system, only clinical information provided by experienced specialists was utilized and no radiographic annotations were employed. Despite the promising potential of the hybrid system developed from our initial investigation, it is essential to further train and validate it in different independent populations because a prediction rule derived from one sample could perform better in another sample/population. Besides, it is reasonable to assume that the OPG images taken from different machines may greatly influence the accuracy of the developed hybrid system. Therefore, it is logical to conduct a multi-center study to collect different OPG images from various centers worldwide and the dataset will be utilized to train further and validate the hybrid system ensuring its accuracy and efficacy in periodontitis diagnosis.

In this study, we will compare the diagnostic characteristics of a novel AI-clinical-based hybrid system (Index test) with a panel of experts (reference standard). Experts will independently assess all radiographs and reach agreement if any discrepancy among them is found.