{
  "id": 741238,
  "ownerSlug": "philgear",
  "slug": "rsna-knee-convnext-large",
  "title": "Pocketgull RSNA Knee ConvNeXt-Large",
  "subtitle": "Pre-trained ConvNeXt-Large multi-label abnormality detector for 3D DICOM knee MRI series.",
  "isPrivate": false,
  "description": "# Pocketgull RSNA Knee ConvNeXt-Large Model\n\n## Model Overview\nThis model repository contains pre-trained multi-label abnormality classification weights trained on de-identified 3D knee MRI scans as part of Pocketgull's clinical decision intelligence pipeline.\n\n## Architecture & Training\n- **Backbone**: ConvNeXt-Large (FP16 ONNX export support)\n- **Loss Function**: Asymmetric Loss (ASL) with gamma_minus=4.0, gamma_plus=1.0\n- **Validation Protocol**: 5-Fold GroupKFold anchored by patient_id\n\n## HIPAA & Safe Harbor Compliance\nAll data and model inputs comply strictly with HIPAA \u00a7164.514(b)(2) Safe Harbor de-identification rules.",
  "publishTime": null
}