Answers are grounded in this paper and cite their sources.
Averaged order parameters say how ordered a lattice of orientational states is, but not where its organization is unusual. We extend the local order and entropy quantities of the MyUncle maintenance-lattice model from the plane to unit vectors in d dimensions: the local disorder is one minus the length of the mean of the neighbors' unit vectors, and the information held in the local order follows from the von Mises-Fisher distribution; both equal MyUncle's at d = 2 (tested against the unmodified code). On this basis we build a label-free score that judges how rare the state of a site is at several prevalence scales and smooths it over the lattice, and compare it with isolation forest, local outlier factor, nearest-neighbor and Mahalanobis baselines, with settings frozen before the held-out data were generated. On lattices simulated with MyUncle (damaged sites as ground truth; one development and three held-out parameter sets, labeled polymer photodegradation, liquid crystal and magnetic storage decay), smoothing lowered the average precision of every baseline because damaged sites are not compact regions, adding the local disorder as a feature raised it, and the frozen score was within 0.02 of the best on the held-out sets but clearly beaten on the development set. On synthetic stained-tissue images, smoothing raised the spatial coherence of every scorer; the frozen score was the best method with DINOv2 embeddings and a poor one with handcrafted features. All evidence is simulated; validation on experimental micrographs is future work. Code and results are open.
No takes yet. Share an insight, caveat, or question.
Leon Sandler (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: