Anatolution, an online platform for consensus morphology
Introduction: Reliable cell annotations are essential for supervised image analysis, yet producing the annotations required for quantitative cellular morphometry is laborious, and results can vary with both tissue preparation and expert interpretation of ambiguous morphology. Nissl staining has long underpinned comparative neuroanatomy, generating extensive and taxonomically diverse records of cellular organization across the nervous system. We developed Anatolution to make this resource more amenable to standardized quantitative analysis by integrating independent annotations alongside explicit measures of rater agreement for quality control under computer vision bookkeeping. Method: Anatolution is an open-source web platform for annotating single images or 3D volumes (https://anatolution.herokuapp.com/public-tool/). Multiple raters independently outline cellular structures using a shared set of labels or conventions, and the platform enables consensus analysis alongside conventional image segmentation. Result: We evaluated 603,297 annotations produced by 48 raters across 1,024 image tiles and tested a subset of 68 fully annotated tiles against unsupervised segmentation. Boundary overlap in cell drawings, scored from 0 (“no overlap”) to 1 (“identical”), rose from 0.30 for 1 rater to 0.79 with 7 users, and in a subset of 68 annotated tiles, 88% contained unsupervised segmentation results. Conclusion: Agreement between groups of raters increased as raters were added to an optimum of ∼7 users, and within a subset of images nearly 9 in 10 annotations were spatially anchored to unsupervised methods. Anatolution supports the production of consensus labels whose stability can be quantified and improved by adding independent raters, providing structured reference data for training and evaluating cell segmentation models for quantitative histology.
Authors
- Blaise Gratton (ORCID: https://orcid.org/0000-0002-3424-2433)
- Jon H. Kaas (ORCID: https://orcid.org/0000-0002-9985-1192)
- Zoe Genevieve LeBlanc (ORCID: https://orcid.org/0000-0003-2012-8805)
- Daniel James Miller (ORCID: https://orcid.org/0000-0002-4844-0828)
Institutions
- University of Illinois Urbana-Champaign (US)
- Vanderbilt University (US)
Publication Details
- Journal
- Brain Behavior and Evolution
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1159/bbe/abqag015
- Primary Topic
- Cell Image Analysis Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00