Year: 2026
Type: article
Abstract: This work assesses the effectiveness of a co-registration-based workflow as a multi-contrast neuroimaging merging strategy. For a valid publicly available NiBabel sample dataset containing a structural MRI volume and a four-dimensional echo-planar imaging functional MRI series. After implementing NaN management, robust intensity clipping, min-max normalization, and bicubic resizing, the mean EPI volume was aligned with the reference of structural MRI. Using controlled translations of three axial slices, 24 deterministic perturbation cases were created to evaluate registrations stability. The co-registrationwas done using a mutual-information-based exhaustive translation search followed by five fusion strategies, namely, weighted averaging, principal component analysis fusion, wavelet max-detail fusion, Laplacian pyramid fusion, and edge-adaptive wavelet fusion. The methods used in the study were assessed using various metrics including mutual information, normalized cross-correlation, rmse, ssim, psnr, entropy, edge preservation and normalized composite score. After co-registration, mutual information improved from 0.391 to 0.440, and RMSE reduced from 0.386 to 0.379. Weighted averaging gives the highest mean composite fusion value (0.522) while PCA fusion comes next with the score of 0.518. Results indicate that a transparent registration and fusion pipeline can produce measurable alignment and fusion-quality benefits, although clinical generalization is limited by the small sample dataset.
Author: Sonia Gupta, Sandesh Gupta
Language: English
Subfield: Media Technology
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7216675627
