Enhancing volume reconstruction of spinal cord from medical images using Res-V-Net-XL and adaptive interpolation

Reconstructing a 3D volume of the spinal cord from medical images presents many challenges, including the preservation of fine structural details and handling variable slice orientations and gaps between slices. Conventional strategies, such as direct, cubic, nearest-neighbour, B-spline, Lanczos, and radial basis function (RBF) interpolation, find it hard to preserve spatial consistency and frequently come about in obscured or mutilated recreations. Deep learning models like U-Net variations fall flat to capture the complex highlights of the spinal cord structure completely, driving problematic execution in division and recreation assignments. To overcome these challenges, we propose a crossover strategy that coordinates an improved V-Net with versatile addition strategies, called Res-V-Net-XL, which stands for residual volume network with extended loss. The Res-V-Net-XL is developed for the exact division of spinal cord highlights, utilising deep residual connections with an added layer and spatial attention mechanisms to extricate anatomical information from medical images. This is then followed by adaptive interpolation, which dynamically adjusts the slice orientation and stitches them based on the size of the input to balance both the processing time as well as accuracy. The suggested hybrid approach works noticeably better than both conventional interpolation methods and the most advanced U-Net variations, according to our testing data. Res-V-Net-XL has a reconstruction accuracy of 98.89%, dice score of 98.98%, and precision of 99.2%. These findings outperform existing models, making our technique an important addition to the precise and trustworthy 3D reconstruction of the spinal cord.