A Three-Stage Caries Detection Approach From Radiographic Dental Images

Dental caries remains one of the most common and painful oral health issues globally. If not detected early, it can extend beyond the enamel, damaging the dental pulp and surrounding gum tissues, often requiring invasive and costly treatments. Traditional diagnostic techniques, such as Fiber-Optic Transillumination, visual inspection, and radiographic analysis, are heavily reliant on clinical expertise and are time-intensive. Therefore, an automated, efficient system to assist clinicians in accurately detecting caries boundaries from dental radiographs is both timely and necessary for improved treatment planning and monitoring. In this context, we propose a novel three-stage method for caries lesion detection in dental radiographic images. The core principle is progressively reducing the computational area by removing irrelevant background regions at each stage, thereby lowering computational cost and energy consumption. In the first stage, a hybrid graph cut technique is used to remove non-essential regions and localize a rough area of interest. The second stage employs Grasshopper Optimization to refine the lesion boundary with greater precision. In the final stage, a modified level set method is applied to accurately delineate the caries-affected regions. Experimental validation shows that the proposed method achieves an average accuracy of 90.62% across caries region detection. Although deep learning-based models may offer marginally better accuracy, they come with substantially higher energy and computational demands. The proposed method offers a balanced trade-off between accuracy and energy efficiency, making it well-suited for clinical environments where power consumption and processing resources are limited. These results affirm the potential of the proposed method as an effective, energy-efficient tool for early caries diagnosis.