DENTAL DISEASE DETECTION USING ADVANCED DEEP LEARNING TECHNIQUES

Abstract

Dental diseases are prevalent clinical conditions that can adversely affect oral function, quality of life, and overall health. Dental radiography is an essential component of dental diagnosis; however, the interpretation of radiographic images can be time-consuming and may be influenced by image quality, disease presentation, anatomical complexity, and clinician experience. Recent advances in deep learning (DL), particularly convolutional neural networks (CNNs) and transfer-learning architectures, have demonstrated considerable potential for automated analysis of medical and dental images. This study proposes a comparative deep learning framework for automated classification of diseased and normal dental X-ray images using DenseNet121, EfficientNet-B5, and U-Net. The proposed framework incorporates image preprocessing, resizing, pixel normalization, data augmentation, model training, and systematic performance evaluation. The dataset comprises 525 dental X-ray images, including 294 diseased and 231 normal images, with 404 images allocated for training and 121 images for testing. Model performance is assessed using accuracy, precision, recall, F1-score, training and testing loss, and confusion-matrix analysis. Experimental results demonstrate that DenseNet121 achieved a test accuracy of 98.35%, whereas EfficientNet-B5 achieved 99.15%. U-Net is incorporated as the third architecture to investigate the applicability of an encoder-decoder network to dental X-ray analysis; its quantitative performance should be established through independent experimental training and evaluation. The comparative framework provides a systematic basis for examining the effectiveness of different deep learning architectures in automated dental image classification. Nevertheless, the high performance obtained on a relatively small dataset should be interpreted cautiously, as model generalizability may be affected by dataset composition, patient-level data dependence, imaging variability, and potential overfitting. Therefore, patient-level data partitioning, independent external validation, calibrated evaluation, clinically established reference standards, and model explainability are recommended before clinical deployment. The proposed system is consequently positioned as a computer-aided decision-support research prototype rather than a replacement for professional dental diagnosis.