Despite advances in periodontal disease (PD) research and periodontal treatments, 42% of the US population suffer from periodontitis. PD can be prevented if high-risk patients are identified early to provide preventive care. Prediction models can help assess risk for PD before initiation and progression; nevertheless, utilization of existing PD prediction models is seldom because of their suboptimal performance. This study aims to develop and test the PD prediction model using machine learning (ML) and electronic dental record (EDR) data that could provide large sample sizes and up-to-date information. A cohort of 27,138 dental patients and grouped PD diagnoses into: healthy control, mild PD, and severe PD was generated. The ML model (XGBoost) was trained (80% training data) and tested (20% testing data) with a total of 74 features extracted from the EDR. We used a five-fold cross-validation strategy to identify the optimal hyperparameters of the model for this one-vs.-all multi-class classification task. Our prediction model differentiated healthy patients vs. mild PD cases and mild PD vs. severe PD cases with an average area under the curve of 0.72. New associations and features compared to existing models were identified that include patient-level factors such as patient anxiety, chewing problems, speaking trouble, teeth grinding, alcohol consumption, injury to teeth, presence of removable partial dentures, self-image, recreational drugs (Heroin and Marijuana), medications affecting periodontium, and medical conditions such as osteoporosis, cancer, neurological conditions, infectious diseases, endocrine conditions, cardiovascular diseases, and gastroenterology conditions. This pilot study demonstrated promising results in predicting the risk of PD using ML and EDR data. The model may provide new information to the clinicians about the PD risks and the factors responsible for the disease progression to take preventive approaches. Further studies are warned to evaluate the prediction model's performance on the external dataset and determine its usability in clinical settings.

Convolutional Neural Networks (CNNs) such as U-Net have been widely used for medical image segmentation. Dental restorations are prominent features of dental radiographs. Applying U-Net on the panoramic image is challenging, as the shape, size and frequency of different restoration types vary. We hypothesized that models trained on smaller, equally spaced rectangular image crops (tiles) of the panoramic would outperform models trained on the full image. A total of 1781 panoramic radiographs were annotated pixelwise for fillings, crowns, and root canal fillings by dental experts. We used different numbers of tiles for our experiments. Five-times-repeated three-fold cross-validation was used for model evaluation. Training with more tiles improved model performance and accelerated convergence. The F1-score for the full panoramic image was 0.7, compared to 0.83, 0.92 and 0.95 for 6, 10 and 20 tiles, respectively. For root canals fillings, which are small, cone-shaped features that appear less frequently on the radiographs, the performance improvement was even higher (+294%). Training on tiles and pooling the results thereafter improved pixelwise classification performance and reduced the time to model convergence for segmenting dental restorations. Segmentation of panoramic radiographs is biased towards more frequent and extended classes. Tiling may help to overcome this bias and increase accuracy.

Objective: To develop two automated computer algorithms to extract information from clinical notes, and to generate three cohorts of patients (disease improvement, disease progression, and no disease change) to track periodontal disease (PD) change over time using longitudinal electronic dental records (EDR). Methods: We conducted a retrospective study of 28,908 patients who received a comprehensive oral evaluation between 1 January 2009, and 31 December 2014, at Indiana University School of Dentistry (IUSD) clinics. We utilized various Python libraries, such as Pandas, TensorFlow, and PyTorch, and a natural language tool kit to develop and test computer algorithms. We tested the performance through a manual review process by generating a confusion matrix. We calculated precision, recall, sensitivity, specificity, and accuracy to evaluate the performances of the algorithms. Finally, we evaluated the density of longitudinal EDR data for the following follow-up times: (1) None; (2) Up to 5 years; (3) > 5 and ≤ 10 years; and (4) >10 and ≤ 15 years. Results: Thirty-four percent (n = 9954) of the study cohort had up to five years of follow-up visits, with an average of 2.78 visits with periodontal charting information. For clinician-documented diagnoses from clinical notes, 42% of patients (n = 5562) had at least two PD diagnoses to determine their disease change. In this cohort, with clinician-documented diagnoses, 72% percent of patients (n = 3919) did not have a disease status change between their first and last visits, 669 (13%) patients’ disease status progressed, and 589 (11%) patients’ disease improved. Conclusions: This study demonstrated the feasibility of utilizing longitudinal EDR data to track disease changes over 15 years during the observation study period. We provided detailed steps and computer algorithms to clean and preprocess the EDR data and generated three cohorts of patients. This information can now be utilized for studying clinical courses using artificial intelligence and machine learning methods.

The major significance of the 2018 gingivitis classification criteria is utilizing a simple, objective, and reliable clinical sign, bleeding on probing score (BOP%), to diagnose gingivitis. However, studies report variations in gingivitis diagnoses with the potential to under- or over-estimating disease occurrence. This study determined the agreement between gingivitis diagnoses generated using the 2018 criteria (BOP%) versus diagnoses using BOP% and other gingival visual assessments. We conducted a retrospective study of 28,908 patients' electronic dental records (EDR) from January-2009 to December-2014, at the Indiana University School of Dentistry. Computational and natural language processing (NLP) approaches were developed to diagnose gingivitis cases from BOP% and retrieve diagnoses from clinical notes. Subsequently, we determined the agreement between BOP%-generated diagnoses and clinician-recorded diagnoses. A thirty-four percent agreement was present between BOP%-generated diagnoses and clinician-recorded diagnoses for disease status (no gingivitis/gingivitis) and a 9% agreement for the disease extent (localized/generalized gingivitis). The computational program and NLP performed excellently with 99.5% and 98% f-1 measures, respectively. Sixty-six percent of patients diagnosed with gingivitis were reclassified as having healthy gingiva based on the 2018 diagnostic classification. The results indicate potential challenges with clinicians adopting the new diagnostic criterion as they transition to using the BOP% alone and not considering the visual signs of inflammation. Periodic training and calibration could facilitate clinicians' and researchers' adoption of the 2018 diagnostic system. The informatics approaches developed could be utilized to automate diagnostic findings from EDR charting and clinical notes.

We aimed to assess the effects of hyperparameter tuning and automatic image augmentation for deep learning-based classification of orthodontic photographs along the Angle classes. Our dataset consisted of 605 images of Angle class I, 1038 images of class II, and 408 images of class III. We trained ResNet architectures for classification of different combinations of learning rate and batch size. For the best combination, we compared the performance of models trained with and without automatic augmentation using 10-fold cross-validation. We used GradCAM to increase explainability, which can provide heat maps containing the salient areas relevant for the classification. The best combination of hyperparameters yielded a model with an accuracy of 0.63–0.64, F1-score 0.61–0.62, sensitivity 0.59–0.65, and specificity 0.80–0.81. For all metrics, it was apparent that there was an ideal corridor of batch size and learning rate combinations; smaller learning rates were associated with higher classification performance. Overall, the performance was highest for learning rates of around 1–3 × 10⁻⁶ and a batch size of eight, respectively. Additional automatic augmentation improved all metrics by 5–10% for all metrics. Misclassifications were most common between Angle classes I and II. GradCAM showed that the models employed features relevant for human classification, too. The choice of hyperparameters drastically affected the performance of deep learning models in orthodontics, and automatic image augmentation resulted in further improvements. Our models managed to classify the dental sagittal occlusion along Angle classes based on digital intraoral photos.

Big data has emerged as a pivotal asset in addressing oral health disparities in recent years. Big data encompasses the vast pool of health care–related biomedical information sourced from diverse channels, such as claims data, patient registries, and electronic health records (EHRs). This study is a critical review that synthesizes the evidence, identifies gaps in knowledge, and discusses future implications regarding big data analytics and oral health disparities. Published reports from 2014 to 2023 that studied associations between big data, social determinants of oral health, and oral health disparities, published in English and available in electronic databases, were included. Search engines were MEDLINE via PubMed, Google Scholar, and Web of Science. A total of 23 studies were included in the review, and all were retrospective data analytics. Studies have used a variety of big data sources, including EHRs, claims, and national or regional registries. This study used a framework of data quality dimensions with intrinsic (data attributes) and contextual values (information provided by the data, in this case, oral health disparities) to critically appraise the included studies. Big data revealed disparities in oral health outcomes and dental care utilization based on race, ethnicity, socioeconomic status, geographical location, insurance category, access to care, and other barriers to care. For the intrinsic data dimension, none of the studies addressed or reported data missingness or consistency of the data. The studies clearly provided contextual data dimensions. From a value-added perspective, several studies provided novel and new information related to racial oral health inequities. Several studies used more than one oral health disparities variable or a composite variable. However, the conclusions from several studies were based on association-based analytics, and few studies used artificial intelligence approaches to understand the population’s oral health inequities—gaps were seen in the study designs and causal analytics.