A study published on 12 October in npj Imaging by University of Ottawa researchers introduces SlideFiLM, a method designed to segment gigapixel pathology slides with fewer pixel-level labels. It passes context from an entire slide into the processing of each image tile through a shared modulation layer. The authors evaluated it on glomeruli in PAS-stained kidney slides and atherosclerotic plaque lesions in mouse aortic-root sections.

In the published supplementary tables, using 20% of labeled training data produced whole-slide Dice scores of 87.19% on the KPIs kidney dataset and 86.48% on the MARA vascular dataset; the compared HookNet scores were 75.36% and 79.52%, respectively. The results are limited to these benchmark datasets and the authors’ training setup; the method is not presented as a diagnostic system validated in clinical workflows or independent hospitals. The journal marks the paper as an accepted early version.