Whole-image + Metadata Model Weights and Metadata-Ablation Outputs for BTXRD Three-Class Bone Lesion Classification

This record contains fold-level trained model weights and metadata-ablation outputs for the BTXRD three-class bone lesion classification study. The uploaded model weights correspond to EfficientNetV2-S Whole-image + 11-dimensional clinical metadata models trained across three random seeds and five cross-validation folds. The metadata-ablation outputs include real, permuted, and zeroed metadata evaluations used to quantify the contribution of clinical metadata while holding the trained model weights and image inputs fixed. The uploaded ZIP archive includes the model weight files, metadata-ablation result files, fold-level result summaries, and checksum/README files.