AI predicts severity of periodontal disease post-treatment

Search Dental Tribune

Artificial intelligence takes early steps in predicting post-treatment periodontal disease severity

An initial foray into machine learning for predicting the severity of periodontal disease post-treatment has given researchers much to learn from and to hope for. (Image: Jonathan Hevia/Shutterstock)

LAHORE, Pakistan: Because treating periodontal disease requires a nuanced approach, a means of predicting the most likely treatment requirements could help dental professionals tailor treatments to individual patients. In a new study, researchers in Pakistan have evaluated whether using a self-developed machine learning model—a form of artificial intelligence (AI)—could help predict the level of severity of even the most complex courses of periodontal disease after treatment. Though the data employed was limited, they found that AI could influence the periodontal disease treatment carried out, indicating a greater role for machine learning in patient care.

In recent years, the use of AI in healthcare has been studied a great deal, and its potential to improve diagnosis and treatment outcomes has been demonstrated. For example, it has been successfully applied to diagnosis of periodontal disease using panoramic radiographs. However, there has been little research regarding its use to predict the course and outcomes of periodontal disease.

After generating a synthetic data set of 1,000 patients, focusing on variables like age, smoking status and disease severity before and after treatment, the researchers in this study employed a linear regression machine learning model for predictive analysis. The artificial patients were aged 20–80 years and had a median age of 45 years. Half of them were smokers, and roughly half had received periodontal treatment. The severity of periodontal disease ranged from 0 (healthy) to 10 (severe), and post-treatment observations showed a general decrease in disease severity.

Correlation analyses found no significant relationship between smoking habits, age and disease severity either before or after treatment. There was a weak correlation between age and treatment outcome, a surprising lack of a significant relationship between smoking and post-treatment disease severity, and a positive correlation between disease severity pre- and post-treatment. Notably, those with severe disease before treatment often showed severe disease post-treatment too, suggesting that more severe cases might be more difficult to treat effectively. The model illuminated the nuanced interactions of demographic and disease variables, but had limited predictive success, in part because the research assumed that the treatment given was universally effective, and it did not consider clinician’s skill, for example.

Clinically, the findings underscore the need for personalised care, factoring in individual patient nuances. From an AI perspective, the study highlights challenges in healthcare predictions, emphasising the potential and need for continuous AI refinement. The study authors recommended that future research should address the study limitations, such as the artificially generated data, and possibly incorporate advanced AI methodologies, as we inch closer to AI-driven predictive healthcare. They additionally suggested that future research could explore other ways of training models, like gradient boosting or neural networks, for comparison.

The study, titled “Role of artificial intelligence in periodontology”, was published on 27 May 2023 in Pakistan Journal of Medical and Health Sciences.

To post a reply please login or register