Exploring Advanced Deep Learning Models for Super Resolution of 3D Dental CBCT Volumes

High Resolution plays an important role in digital imaging, but it is even more important in medical imaging. However, due to certain constraints, medical imaging is often captured in low dose radiation which causes noise and lower resolution. As high resolution plays a vital role in diagnostics and model training, we compare efficiency and accuracy of state-of-the-art deep learning models for super-resolution reconstruction in 3D volumes. We have employed various deep learning models, such as CNN, SR-GAN, UNet and Auto-Encoder for super resolution-reconstruction of 3D Dental CBCT volumes. To optimize their performance, we have combined different architectural enhancements such as multipath structure in CNN and SR-GAN for enhanced features extraction, Mamba with UNet to capture global dependencies, and Diffusion with Auto-En-coder for feature refinement. The performance of CNN is highest with Peak Sig-nal-to-Noise Ratio and Structural Similarity Index of 35.35 and 0.952 respec-tively; however, Auto-Encoder is the fastest with training time of 28.63 hours.