Evaluating the impact of preprocessing on deep learningbased dental decision support systems

Automated dental diagnosis utilizing deep learning has demonstrated considerable potential. However, challenges such as image noise, low contrast, and anatomical variability in panoramic X-rays still hinder accurate model predictions. This study offers a thorough evaluation of how various image pre-processing techniques influence deep learning models within dental decision support systems. Initially, we examined multiple variants of YOLO for teeth instance segmentation. Subsequently, we trained the DenseNet121-based mandibular canal classification model on the dental dataset to assess the effects of various digital image processing methods—including histogram equalization, CLAHE, Gaussian blur, morphological operations, and high-pass filters—on model performance. The findings highlight the importance of pre-processing and the integration of optimized pre-processing strategies into deep learning pipelines to enhance outcomes in the development of dental decision support systems.