Back to search
2109.12688

Nesterov Accelerated ADMM for Fast Diffeomorphic Image Registration

Alexander Thorley, Xi Jia, Hyung Jin Chang, Boyang Liu, Karina Bunting, Victoria Stoll, Antonio de Marvao, Declan P. O’Regan, Georgios Gkoutos, Dipak Kotecha, Jinming Duan

correctmedium confidence
Category
Not specified
Journal tier
Specialist/Solid
Processed
Sep 28, 2025, 12:56 AM

Audit review

For parts (a)–(c), the candidate solution’s closed-form ADMM updates for v (s=1 clamped shrinkage; s=2 normal equations) and the FFT-diagonalizable w-update exactly match the paper’s stated formulas and algorithmic setup (Eqs. (3)–(6)), including the z-hat clamping rule and the Fourier symbol of Δ^n . The paper asserts diffeomorphism via compositions of sufficiently small, smooth velocity fields (a standard claim) but does not supply a precise hypothesis or proof beyond citing prior topological results and showing empirical det(Jφ)≥0; the candidate supplies a rigorous sufficient condition ||∇v_i||∞<1 with a constructive proof via the inverse function theorem and contraction mapping, thereby strengthening the paper’s informal justification . Net: both are correct; the candidate gives fuller mathematical details for (d).

Referee report (LaTeX)

\textbf{Recommendation:} minor revisions

\textbf{Journal Tier:} specialist/solid

\textbf{Justification:}

Technically solid and practically impactful. The accelerated ADMM with closed-form v/w updates is well-motivated and clearly presented; experiments are convincing. The main gap is that the diffeomorphism guarantee is stated informally. Adding a standard sufficient condition and a short proof (or precise citation) would make the argument complete. A small normalization ambiguity in the w-update and a mis-pointed citation should be corrected.