Periodontal diseases are the largest cause of tooth loss among people of all ages and are also correlated with systemic diseases such as endocarditis. Advanced periodontal disease comprises degradation of surrounding tooth structures, severe inflammation and gingival bleeding. Inflammation is an early indicator of periodontal disease. Early detection and preventive measures can help prevent serious occurrences of periodontal diseases and in most cases restore oral health. We report a machine learning classifier, trained with annotations from dental professionals, that successfully provides pixel-wise inflammation segmentations of color-augmented intraoral images. The classifier successfully distinguishes between inflamed and healthy gingiva and its area under the receiver operating characteristic curve is 0.746, with precision and recall of 0.347 and 0.621 respectively. Dental professionals and patients can benefit from automated point-of-care early diagnosis of periodontal diseases provided by this classifier using oral images acquired by intraoral imaging devices.
Why is this work important?
Visual inspection and probing techniques have been traditionally used for diagnosis of oral diseases in patients. These traditional methods are subjective and not scalable. We describe the use of RGB color images acquired by low-cost camera devices coupled with machine learning to detect areas with poor oral health.
What has been done before?
Currently the gold standard for oral diagnosis is visual inspections by a dentist followed by X-rays. These methods are expensive and invasive.
What are our contributions?
We implement a novel technique to combine medical expert knowledge with biomarker signatures. We use RGB color images taken directly at the point-of-care, using low-cost hand-held devices, to provide a first level machine-learning powered screening for patients.
What are the next steps?
We are expanding the repertoire of biomarkers that can be detected in RGB color images acquired at the point-of-care and pairing them with automated machine learning exams.