Dental caries remains a significant oral health issue requiring advancements in diagnostic methods. This study presents a deep learning framework integrating convolutional neural networks (CNNs) with recurrent neural networks (RNNs) to enhance caries detection and classification. Using a dataset of 785 CBCT images, the hybrid model achieved sensitivity and specificity rates of 95.3% and 96.3%, respectively. The model excelled in classifying non-cavitated and cavitated occlusal, proximal, cervical, and multi-caries, particularly in Type IV (multiple caries) with 98.8% accuracy. However, challenges were noted in detecting Type III (cervical) caries due to inconsistent data. The integration of spatially dependent structures in CNNs and temporally dependent structures in RNNs improved diagnostic performance, underscoring the need for hybrid AI systems to enhance dental diagnostics, especially in areas with limited expert availability.