Artificial Intelligence Evaluation of Periodontal Health Using Selfie Intraoral Photography
Artificial Intelligence Evaluation of Periodontal Health Using Selfie Intraoral Photography
Periodontal (gum) disease affecting more than 90% of the population globally. The soft and hard tissues that supporting the teeth are being affected. If untreated, the disease progresses from the mild and reversible form (i.e. gingivitis) that involves superficial gum only to the more severe and irreversible form (i.e. periodontitis) that involves loss of periodontal ligament and bone. Teeth will be lost eventually and significantly impairs the function and the oral health related quality of health. Moreover, periodontal disease has been strongly linked to the systemic diseases via centrally or local routes with significant health implications.
Periodontal disease is initiated by bacteria (plaque) adhere on tooth surfaces and the body defense by eliciting inflammatory response. The disease and associated inflammation is site-specific and the affected gum is manifested with the cardinal signs of inflammation such as redness (color), swelling (increased volume), edema (loss of surface characteristics) and bleeds easily. Dentists are trained to identify the disease sites by visual (redness and swelling) and by mechanical probing (bleeding). Traditionally these give rises to clinical gum indices showing the degree of inflammation and are important to the clinical monitoring and management of gum diseases.
The management of periodontal disease involves the removal of bacteria plaque by both dentists' tooth cleaning and maintained by patients' daily home-care. Home-care plaque removal has been shown by many studies to prevent and/or cease the development of periodontitis. However, most patients do not able to remove plaque effectively and it only takes few days for a health site to development inflammation. Professional monitoring and feedback are highly desirable but many patients only have monthly or even yearly appointment which is too late. Such "non-compliance" increase the treatment cost/time, patient discomfort and reduce treatment efficacy. Moreover, many patients do not receive regular dental checkup and they seek dentists when the gum problem becomes irreversible that complicated and expensive treatment such as tooth extraction and rehabilitation is required.
This study consists of 1) collecting the standardized clinical photography with clinical gum indices/ marked by an experienced clinician, 2) import into computer for training, and 3) longitudinally monitoring and analyze of gum condition in a group of gingivitis patients receiving gum treatment.
The first part of this study will collect buccal view of intraoral photo and clinical gum indices from 300 patients visiting Prince Philip Dental Hospital (PPDH). These intraoral photo will be taken by patients using smartphone (iphone) mounted with apparatus (Scanbox) in a standardized way and the quality supervised by clinical staff. Clinical gum indices will be taken by a trained and calibrated clinical staff. Alternatively clinical photo may be assessed by an experienced clinician to locate the inflamed/health gum. This synthesized data set imported into neural network to learn the image characteristics of inflamed and health gum and train the computer.
The second part of this study will recruit 150 patients with gingivitis and longitudinally follow up their gum condition pre- and post- gum treatment in PPDH using standardized intraoral photo and clinical gum indices. The clinical gum indices are the gold standard reflecting the changes in the gum condition.
All photos collected would be analyzed by computer (software: MatLab) to identify key features that help with screening and monitoring of gum condition. Data would then be collected for building a program that quantifies the identified features which may include the following:
Data and sample size analysis: The computer will be trained with the photo and clinical indices/assessment obtained in the part I of this study. The trained computer will then use to monitor changes of periodontal condition of patients in part II of this study based on photo only. The sensitivity (detect true disease), specificity (detect true health) and the area under these curves (with reference to the gold standard clinical gum indices) will be calculated and are the main outcome of this study. The training of computer is a continuous process and the endpoint comes when the sensitivity, specificity and the area under these curves (AUC) reach plateau (saturation). Therefore, the training of computer will be staged into three phrases with 100, 200 and 300 of cases in the part I of this study and observe the changes in the sensitivity, specificity and AUC.
Patient reported outcome on the use of smartphone selfie would be recorded in visual analogue scale (VAS) as secondary outcome.
Sample size determination: Each tooth represents one unit and assume each subject has 20 teeth, there will be more than 300 subjects X 20 teeth =6000 units for training. For the part 2 study, 150 subjects will be recruited and 130X20=2600 units for testing of the computer.
All new patients attending Prince Philip Dental Hospital will be screen for gum inflammation and periodontal diseases by clinical examination.
Basic periodontal examination (BPE, including visual examination and mechanical probing of gum) will be performed for all of them and assign to the undergraduate students for patient care according to the severity.
Computer record can retrieve patients with gum inflammation (BPE score 2). Data handling All data will be kept reviewed by unblended member (YHL) to check the quality. Only the data that related to the study outcomes will be disclosed. All the data will be kept for another two years since the study finishes. After that, the data will be destroyed completely. Primary investigator will have access to the personal data during and after the study.
Inclusion Criteria:
Exclusion Criteria: