peds-dmri-lfcl: Label Fusion and Corrective Learning for Pediatric Dynamic MRI

Abstract

This dataset accompanies the study “Multi-Atlas-Based Segmentation of Pediatric Vocal Tract Anatomy in Dynamic MRI”. It provides a demonstration of our segmentation pipeline, which includes atlas-to-subject registration, multiple label fusion strategies (MVLF, MALF, STAPLE), and a corrective learning framework (boundary learning and shape/appearance refinement). The dataset contains a small, de-identified subset of dynamic MRI scans: Cookie atlas: 11 consecutive frames with manual labels Buy subject: 11 frames, with manual labels available for tf01 and tf11 These demo data enable researchers to reproduce the workflow. The full pediatric dataset cannot be released due to IRB restrictions. The accompanying scripts implement the full pipeline, including ANTs registration, PICSL-MALF label fusion, corrective learning, and figure generation. Disclaimer: This software and data are provided for research use only and have not been certified for clinical applications.