AnnotateAnyCell: Open-Source AI Framework for Efficient Annotation in Digital Pathology

Abstract

AO_SCPLOWBSTRACTC_SCPLOWManual annotation of histopathological whole slide images remains a critical bottleneck for computational pathology and clinical AI deployment, requiring prohibitive expert time at scale. Here we present an open-source semi-supervised framework combining active contrastive learning with iterative human-in-the-loop feedback for efficient cellular annotation and classification. The pipeline integrates Cellpose segmentation, UMAP-based latent space visualization, and contrastive learning with pseudolabel propagation, evaluated on five whole slide images of canine invasive urothelial carcinoma across low, intermediate, and high histological grades at 40x magnification. Latent space clustering-guided annotation required 47 minutes compared to 63 minutes for sequential annotation, a 25% reduction (95% CI 18-32%). Classification accuracy reached 96.3% {+/-} 1.2% for mitotic figures and 98.3% {+/-} 1.4% for nucleoli using 1,075 labeled samples, with nucleoli classification achieving 95.5% {+/-} 1.5% accuracy from only 215 samples. Inter-annotator agreement was high for chromatin ({kappa} = 1.00) and nucleoli ({kappa} = 0.95) but moderate for mitotic figures ({kappa} = 0.58) and nuclear shape ({kappa} = 0.36), reflecting intrinsic morphological ambiguity in these categories. This framework substantially reduces annotation burden while achieving expert-level accuracy for well-defined morphological features, providing a scalable path toward AI-assisted diagnostics in resource-constrained pathology settings.