A Novel Methodology to Predict Dental Cavities based on Intensive Deep Learning Strategy

Quantitative Light-Induced Fluorescence (Qua-LF) is an objective method for detecting and measuring dental cavity based on plaque using the natural fluorescence properties of bacterial porphyrins. Additionally, complementary artificial intelligence modeling, such as XGBoost, has demonstrated superior predictive accuracy compared to traditional regression models in assessing oral health data. However, the effectiveness of these models heavily relies on the quality and quantity of training data, with larger and more diverse datasets improving accuracy and robustness. Supervised learning, the primary learning technique used in these studies, requires manually annotated datasets, which is time-consuming and labor-intensive. Inconsistencies in image quality, lighting conditions, and the inherent variability of plaque characteristics among individuals and clinical settings further complicate these challenges. Addressing these issues is crucial for improving the reliability and generalizability of machine learning-based plaque quantification using intraoral photography. Practical challenges, such as the costs associated with acquiring and maintaining necessary hardware and software, as well as the required training and expertise for clinicians, pose barriers to widespread implementation. To address these considerations, we propose an innovative protocol for dental cavity prediction using a Modified YOLO-based Deep Learning Model (MYoD-M), which aims to increase the reliability of the prediction mechanism. In our proposed model having two stages of implementation procedure, in primary we perform the super-pixilation based image enhancement with Gaussian-MRFi based segmentation. In stage 2, MYoD-Model is used to predict the dental cavity stages in four classes with boundary lines. Our proposed model outperformed the key performance indices with the overall accuracy nearly 94.5%.