Enhancing the Accuracy of 3D Brain MRI Reconstruction via Super Resolution by Utilizing Residual Variational Autoencoder

To improve the quality of a medical image, it is essential to employ equipment that is equipped with a higher-intensity magnetic field. This poses unique challenges for healthcare providers, particularly those who have limited resources and an urgent need for efficiency. Interpolation is the most efficient technique for transforming low-resolution photos into high-resolution images through a straightforward calculation. The interpolation technique yields images with reduced clarity, especially along seamless boundaries, leading to the omission of crucial details. This study suggests employing the residual variational autoencoder model for the purpose of reconstructing high-resolution images. The model comprises three components: an encoder, a decoder, and a latent space. According to the test findings, the suggested model performs better than state-of-the-art techniques already in use, including interpolation, multi-level densely connected super-resolution networks, and variational autoencoders. The evaluation of Structural Similarity and Peak signal-to-noise ratio metrics reveals a significant improvement, with an approximate 5–10% rise in Structural Similarity and a 10–15% increase in Peak signal-to-noise ratio compared to the state-of-the-art.