Interactive segmentation tools accelerate annotation but often operate on a single image at a time and lack robust initialization. We extend nnInteractive with (i) dataset-aware traversal that supports sequential/on-the-fly loading of images and volumes, similar to MONAI Label’s iteration approach, and (ii) model-assisted pre-segmentation via nnU-Net to warm-start the interaction. Our traversal module adds persistent progress tracking and resume-from-last functionality, enabling efficient, dataset-scale labeling sessions. The nnU-Net integration generates class-wise candidate masks (2D/3D) that are overlaid in the UI for immediate acceptance, editing, or rejection. Users can toggle structures, adjust thresholds, and refine them with standard nnInteractive tools. The system supports batch precomputation or on-the-fly inference, and logs edits to facilitate auditability and reproducibility. We illustrate end-to-end workflows on the pediatric BraTS dataset of 2024 and outline evaluation protocols (clicks, correction time, Dice-Sørensen score after refinement) to quantify annotation efficiency. Together, these additions turn nnInteractive into a scalable, dataset-centric labeling environment that leverages state-of-the-art segmentation priors while preserving expert control. The code is available at: https://github.com/Clinical-Computational-Medical-Imaging/label-refiner